CN118013399B - AI model-based user portrait processing method and device - Google Patents
AI model-based user portrait processing method and device Download PDFInfo
- Publication number
- CN118013399B CN118013399B CN202410410709.2A CN202410410709A CN118013399B CN 118013399 B CN118013399 B CN 118013399B CN 202410410709 A CN202410410709 A CN 202410410709A CN 118013399 B CN118013399 B CN 118013399B
- Authority
- CN
- China
- Prior art keywords
- user
- vector
- behavior
- classification
- type
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 13
- 239000013598 vector Substances 0.000 claims abstract description 446
- 238000012549 training Methods 0.000 claims abstract description 208
- 238000013145 classification model Methods 0.000 claims abstract description 156
- 238000012545 processing Methods 0.000 claims abstract description 108
- 238000000034 method Methods 0.000 claims abstract description 46
- 230000006399 behavior Effects 0.000 claims description 430
- 238000004891 communication Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 26
- 230000008859 change Effects 0.000 claims description 22
- 238000010606 normalization Methods 0.000 claims description 14
- 230000011218 segmentation Effects 0.000 claims description 12
- 238000002360 preparation method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 4
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 12
- 230000008569 process Effects 0.000 description 8
- 238000005457 optimization Methods 0.000 description 6
- 238000012423 maintenance Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention relates to a processing method and a device for user portraits based on an AI model, wherein the method comprises the following steps: configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database; and after training the first, second and third classification models, processing the initial representation of the user using the first classification model, processing the user tracking representation using the second classification model, and processing the user representation adjustment based on the task feedback using the third classification model. The invention can improve the real-time performance and the processing efficiency of the user portrait.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a user portrait processing method and device based on an AI model.
Background
The user portraits are classified according to the behavior habits of the user and give corresponding grade information in each class, so the user portraits are actually the type and grade of the user. The conventional way to process the user image is to rely on manual processing, for example: 1) When a user is newly added, a front-end service person firstly carries out behavior data investigation on the newly added user, and then a background service director carries out initial portrait on the user according to investigation data and manual experience and reports the initial portrait; 2) In the process of user maintenance, front-end service personnel regularly and continuously acquire daily behavior data of a user through user revisions or through data tracking by an information engineer and store the daily behavior data, a background service manager judges whether the type and grade information of the user changes according to the user behavior tracking data and manual experience in the latest period, and if so, the corresponding user portrait information is updated; 3) The service provider (such as financial service field, insurance service field and telecommunication service field) in each service field can customize different user task plans for users with different portrait features (different types and grades), one service manager is communicated with a plurality of service personnel to implement the service, when the service provider is implemented, each service personnel executes different subtask plans, when the task execution is completed, the current task is scored and fed back based on the current execution status (mainly the type reaction of the user side), and after all service personnel complete the feedback, the service manager judges whether the portrait of the user of the current user needs to be updated or not according to all feedback information. The limitation of the artificial image mode with each level of business director as the core is obvious: the real-time performance of task processing is poor and the efficiency is low. The technical problem to be solved by the invention is how to improve the real-time performance and the processing efficiency of the user portrait through the artificial intelligent model (ARTIFICIAL INTELLIGENCE, AI).
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a processing method, a device, electronic equipment and a computer readable storage medium for carrying out user portrait based on an AI model. Firstly, constructing three classification models, namely a first classification model, a second classification model and a third classification model based on a structure of a multi-layer perception network (Multilayer Perceptron, MLP) +softmax, planning data formats of input vectors/tensors of the first classification model, the second classification model and the third classification model, and training the first classification model, the second classification model and the third classification model based on a stock user database which is stored by a business party and is used for recording manual processing information of a stock client; and after training, processing the initial portraits of newly added users by using a first classification model, processing the tracking portraits by using a second classification model in the continuous maintenance stage of the users, and processing user portraits adjustment based on feedback by using a third classification model when all feedback information of the user task plans is received. The invention can improve the real-time performance and the processing efficiency of the user portrait.
To achieve the above object, a first aspect of the present invention provides a method for processing a user portrait based on an AI model, where the method includes:
Configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database; the first classification model is used for carrying out user tag type prediction processing according to the input user behavior vector to obtain a corresponding user tag classification vector; the second classification model is used for carrying out user tag type prediction processing according to the input user behavior time sequence tensor to obtain the corresponding user tag classification vector; the third classification model is used for carrying out user tag type prediction processing according to the input user task feedback vector to obtain a corresponding user tag classification vector;
After training the first, second and third classification models, processing an initial image of a user by using the first classification model; processing the user tracking image by using the second classification model; and processing user portrayal adjustment based on task feedback using the third classification model.
Preferably, the user behavior vector is a one-dimensional vector with a shape of 1×w 1, and the vector length W 1 is an integer greater than or equal to 1; the user behavior vector comprises W 1 behavior type grades; each behavior type grade corresponds to one behavior type; each behavior type corresponds to a behavior grade value range; each behavior grade value range comprises a plurality of behavior grade values; each behavior grade takes an integer greater than zero; each behavior type grade is one of the behavior grade value ranges corresponding to the corresponding behavior type; the total number of the behavior grade values of all the behavior types corresponding to the user behavior vector is matched with the vector length W 1;
The user behavior time sequence tensor is a two-dimensional tensor with the shape of H multiplied by W 2, the tensor height H is an integer greater than 1, and the tensor width W 2 is an integer greater than or equal to 2, and W 2=1+W1; the user behavior time sequence tensor consists of H user behavior time sequence vectors; the user behavior time sequence vector consists of a time stamp and the user behavior vector corresponding to the current time stamp;
The user task feedback vector is a one-dimensional vector with the shape of 1 XW 3, and the vector length W 3 is an integer greater than or equal to 1; the user task feedback vector comprises W 3 task type feedback levels; each task type feedback level corresponds to one task type; each task type corresponds to a task feedback level value range; each task feedback level value range comprises a plurality of task feedback level values; each task feedback level takes an integer greater than zero; the task type feedback level is one of the task feedback level value ranges corresponding to the corresponding task type; the total number of the task feedback grade values of all the task types corresponding to the user task feedback vector is matched with the vector length W 3;
The user tag classification vector is a one-dimensional vector with the shape of 1 XW 4, and the vector length W 4 is an integer greater than 1; the user behavior vector comprises W 4 user tag prediction probabilities; each user tag prediction probability corresponds to one user type and one user grade; each user type corresponds to a user grade value range; each user grade value range comprises a plurality of user grade values; each user grade value is an integer greater than zero; each user grade is one user grade value in the user grade value range corresponding to the corresponding user type; the prediction probability of each user label is the prediction probability of the corresponding user type plus the user grade; and the total number of the user grade values of all the user types corresponding to the user label classification vector is matched with the vector length W 4.
Preferably, the first classification model includes a first MLP network and a first Softmax layer; the first MLP network consists of an input layer, a hidden layer and an output layer; the first Softmax layer is formed by a Softmax function; the first MLP network is used for extracting features of the user behavior vector input by the model to obtain a corresponding first feature vector and sending the corresponding first feature vector to the first Softmax layer; the first Softmax layer is used for predicting classification probability according to the first feature vector to obtain the corresponding user tag classification vector;
The second classification model includes a second MLP network and a second Softmax layer; the second MLP network is composed of an input layer, a hidden layer and an output layer; the second Softmax layer is composed of a Softmax function; the second MLP network is used for carrying out feature extraction on the user behavior time sequence tensor input by the model to obtain a corresponding second feature vector and sending the corresponding second feature vector to the second Softmax layer; the second Softmax layer is used for predicting classification probability according to the second feature vector to obtain the corresponding user tag classification vector;
The third classification model includes a third MLP network and a third Softmax layer; the third MLP network is composed of an input layer, a hidden layer and an output layer; the third Softmax layer is composed of a Softmax function; the third MLP network is used for extracting features of the user task feedback vector input by the model to obtain a corresponding third feature vector and sending the corresponding third feature vector to the third Softmax layer; and the third Softmax layer is used for predicting the classification probability according to the third feature vector to obtain the corresponding user tag classification vector.
Preferably, the stock user database comprises a first user information acquisition library, a first row data acquisition library and a first task feedback data acquisition library;
the first user information collection library comprises a plurality of first user information tables; each first user information table corresponds to a first stock user; the first user information table includes one or more first type level update records;
the first type level update record comprises a first update time field, a first latest type field and a first latest level field; the time information of the first updating time field comprises year, month, day, time and second information; the first latest type field is one user type; the first latest grade field is one user grade value in the user grade value range corresponding to the corresponding first latest type field;
The first row of data acquisition library comprises a plurality of first data acquisition tables; the first data acquisition table corresponds to the first stock users one by one; the first data acquisition table comprises a plurality of first acquisition records;
the first acquisition record comprises a first acquisition time field and a plurality of first behavior acquisition fields;
the time information of the first acquisition time field comprises year, month, day, time and second information;
the first behavior acquisition field comprises a first behavior type and first acquisition data;
The first behavior type is one behavior type and at least comprises an information browsing behavior, an information focusing behavior, an information sharing behavior and a business communication behavior;
When the first behavior type is an information browsing behavior, the corresponding first acquired data is a total browsing duration obtained by counting the browsing total duration of the corresponding first stock user browsing the information of the appointed field before the current time point at the time point of the corresponding first acquired time field; the appointed field at least comprises a financial service field, an insurance service field and a telecommunication service field; the appointed domain information at least comprises an appointed domain article, an appointed domain image, an appointed domain audio and video, an appointed domain public number, an appointed domain micro signal, an appointed domain micro blog number and an appointed domain website;
When the first behavior type is information attention behavior, the corresponding first acquired data is total attention quantity obtained by counting the total number of the appointed field information currently focused by the corresponding first stock user at the time point of the corresponding first acquisition time field;
when the first behavior type is an information sharing behavior, the corresponding first acquired data is a total sharing quantity obtained by counting the total number of the specified domain information shared by the corresponding first stock users before the current time point at the time point of the corresponding first acquisition time field;
When the first behavior type is a business communication behavior, the corresponding first acquired data is a total communication duration obtained by counting the total duration of communication between the corresponding first stock user and business personnel in the appointed field before the current time point at the time point of the corresponding first acquisition time field;
The first task feedback data acquisition library comprises a plurality of second data acquisition tables; the second data acquisition table corresponds to the first stock users one by one; the second data acquisition table comprises a plurality of second acquisition records; each second acquisition record corresponds to a user task plan; each user mission plan comprises a plurality of sub-mission plans and a user type grade change suggestion; each subtask plan comprises a subtask type, a subtask flow description and a subtask feedback score; each user task plan is completed by cooperation of a business director and one or more business personnel; each subtask plan is executed by a business person according to the corresponding subtask flow description, an execution object is the corresponding first stock user, when the task is ended, the corresponding subtask feedback scores are fed back by the current business person according to the task execution result, and after all the subtask plans are fed back, the corresponding user type grade change suggestions are set by a business director according to all the subtask feedback scores; the user type grade change suggestion comprises a hold-up state and a new user type grade; the new user type grade consists of one user type and the corresponding user grade value;
The second acquisition record comprises a second acquisition time field, a plurality of first feedback acquisition fields and a first type level change suggestion field;
the time information of the second acquisition time field comprises year, month, day, time and second information;
The first feedback acquisition field corresponds to the corresponding subtask plan of the user task plan one by one; the first feedback acquisition field comprises a first task type and first feedback data; the first task type corresponds to the corresponding subtask type of the subtask plan; the first feedback data corresponds to the subtask feedback score of the corresponding subtask plan;
the first type level change suggestion field includes a hold-down and the new user type level.
Preferably, the training the first, second and third classification models based on the preset stock user database specifically includes:
Constructing a training data set based on the stock user database to obtain a first model training data set, a second model training data set and a third model training data set which correspond to the first model training data set, the second model training data set and the third model training data set; the first model training data set includes a plurality of first model training data; the first model training data comprises a first model training vector and a first model label vector; the second model training data set includes a plurality of second model training data; the second model training data includes a second model training tensor and a second model label vector; the third model training data set includes a plurality of third model training data; the third model training data comprises a third model training vector and a third model label vector;
And training the first, second and third classification models based on the first, second and third model training data sets, respectively.
Further, the constructing the training data set based on the stock user database to obtain a corresponding first model training data set, second model training data set and third model training data set specifically includes:
Carrying out structural initialization on the user tag classification vector based on all the user types of the stock user database and the user grade value ranges corresponding to the user types; the initialized vector length W 4 of the user tag classification vector is the total number of the user grade values of all the user types; the initialized user tag prediction probabilities of the user tag classification vectors correspond to one user type and one user grade value;
Designating a corresponding collected data level standardization standard for the first collected data corresponding to each type of the first behavior type of the stock user database; the collected data level standardization standard consists of a plurality of classified behavior level corresponding relations and a classified behavior level value range, each classified behavior level corresponding relation comprises a collected data range and a corresponding classified behavior level value, and the classified behavior level value range consists of a plurality of classified behavior level values; when the first behavior type is specifically an information browsing behavior, each classification behavior grade corresponding relation comprises a browsing duration range and a corresponding browsing behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the browsing behavior grade value one by one; when the first behavior type is information attention behavior, each classification behavior grade corresponding relation comprises an attention quantity range and a corresponding attention behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the attention behavior grade value one by one; when the first behavior type is information sharing behavior, each classification behavior grade corresponding relation comprises a sharing quantity range and a corresponding sharing behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the sharing behavior grade value one by one; when the first behavior type is a business communication behavior, each classification behavior grade corresponding relation comprises a communication duration range and a corresponding communication behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the communication behavior grade value one by one;
The user behavior vector is structurally initialized based on all the first behavior types and the classified behavior grade value ranges corresponding to the first behavior types; the initialized vector length W 1 of the user behavior vector is the total number of the classified behavior grade values of all the first behavior types; each behavior type grade of the initialized user behavior vector is composed of one first behavior type and one corresponding classified behavior grade value;
Initializing the tensor height H of the user behavior time sequence tensor based on a preset second model input Zhang Lianggao degrees, and initializing the tensor width W 2 of the user behavior time sequence tensor by the initialized vector length W 1 of the user behavior vector; the tensor height H of the initialized user behavior time sequence tensor is consistent with the input Zhang Lianggao degrees of the second model, and the tensor width W 2 is the sum of the vector length W 1 and 1;
Designating a corresponding feedback data level standardization standard for the subtask feedback scores corresponding to the subtask types of the stock user database; the feedback data level standardization standard comprises a plurality of subtask feedback level corresponding relations and a subtask feedback level value range, each subtask feedback level corresponding relation comprises a feedback scoring range and a corresponding subtask feedback level value, and all subtask feedback level values form the corresponding subtask feedback level value range;
The user task feedback vector is structurally initialized based on all the subtask types and the subtask feedback grade value ranges corresponding to the subtask types; the initialized vector length W 3 of the user task feedback vector is the total number of the subtask feedback grade values of all subtask types; each task type feedback level of the initialized user task feedback vector is composed of one subtask type and a subtask feedback level value corresponding to the subtask type;
traversing all the first acquisition records of the stock user database; and traversing, wherein the first acquisition record of the current traversal is used as a corresponding current record; the first stock user corresponding to the current record is used as a corresponding current user; the first acquisition time field of the current record is used as the corresponding current time; the first type level update record of the first user information collection library of the stock user database, which corresponds to the current user, is marked as a corresponding first early update record, and the first latest type field and the first latest level field of the first early update record, which are closest to the current time, in all the first early update records are extracted to be used as corresponding matched user types and matched user levels to form corresponding matched user type levels; setting a user tag classification vector for model training according to the matching user type grade based on the vector structure of the user tag classification vector as a corresponding first model tag vector; performing corresponding level normalization processing on each corresponding first acquired data in the current record based on the acquired data level normalization standard corresponding to each first behavior type to obtain a corresponding first classification behavior level value; and the corresponding first behavior type grade is formed by each first behavior type of the current record and the corresponding first classification behavior grade value; setting a user behavior vector for model training as a corresponding first model training vector according to all first behavior type grades recorded currently based on a vector structure of the user behavior vector; the first model training vector and the first model label vector are obtained to form corresponding first model training data; when the traversing is finished, the corresponding first model training data set is formed by all the obtained first model training data;
And forming a corresponding first vector set by all the obtained first model training vectors; the first model training vectors belonging to the first stock clients in the first vector set are gathered into a first user vector set corresponding to the first model training vectors; sequencing all the first model training vectors of each first user vector set according to the sequence from far to near of the first acquisition time fields of the corresponding first acquisition records to obtain a corresponding first user vector sequence;
Traversing all the obtained first user vector sequences; traversing, wherein the first user vector sequence traversed currently is used as a corresponding current sequence; the tensor height H of the user behavior time sequence tensor is used as the corresponding segmentation sequence length; starting from a first model training vector of the current sequence, carrying out sub-sequence sliding segmentation processing on the current sequence by using a preset segmentation sliding step length and the segmentation sequence length to obtain a plurality of corresponding first sub-sequences; setting a corresponding training label vector as a corresponding second model label vector based on the vector structure of the matching user type grade corresponding to the last first model training vector of each first subsequence and the user label classification vector; and polling each of the first sub-sequences; when polling, the first subsequence of the current polling is used as a corresponding current subsequence, a corresponding first user behavior time sequence vector is formed by each first model training vector in the current subsequence and the corresponding first acquisition time field of the first acquisition record, and the corresponding second model training tensor is formed by all the first user behavior time sequence vectors corresponding to the current subsequence; forming corresponding second model training data by each second model training tensor and the corresponding second model label vector; when the traversing is finished, the second model training data sets corresponding to all the obtained second model training data are formed;
traversing all the second acquisition records of the stock user database; and traversing, wherein the second acquisition record of the current traversal is used as a corresponding current record; the first stock user corresponding to the current record is used as a corresponding current user; and taking the second acquisition time field of the current record as the corresponding current time; recording the first type grade update record of which the first update time field is earlier than the current time in the first user information table corresponding to the current user in the first user information collection library as a corresponding second early update record, and forming the first latest type field and the first latest grade field of the first early update record of which the first update time field is closest to the current time in all the second early update records into a corresponding original user type grade; identifying whether the first type level change suggestion field of the current record is the hold state, if so, setting a user tag classification vector for model training according to the original user type level to be a corresponding third model tag vector based on the vector structure of the user tag classification vector, otherwise, setting a user tag classification vector for model training according to the first type level change suggestion field of the current record to be a corresponding third model tag vector based on the vector structure of the user tag classification vector; performing corresponding level normalization processing on the corresponding first feedback data in the current record based on the feedback data level normalization standards corresponding to the subtask types to obtain corresponding first subtask feedback level values; the corresponding first subtask type grade is formed by each subtask type of the current record and the corresponding first subtask feedback grade value; setting a user task feedback vector for model training as a corresponding third model training vector according to all the first subtask type grades recorded currently based on the vector structure of the user task feedback vector; the obtained third model training vector and the corresponding third model label vector form corresponding third model training data; and at the end of the traversal, forming a corresponding third model training data set by all the obtained third model training data.
Preferably, the processing the initial image of the user by using the first classification model specifically includes:
Recording the current newly added user as a corresponding first user when one user is newly added each time; performing primary initial behavior data statistics on the first user according to all the behavior types corresponding to the user behavior vector to generate corresponding first initial behavior data; vector setting is carried out according to the first initial behavior data according to the vector structure of the user behavior vector to obtain a corresponding first user behavior vector; the first classification model predicts the user tag type according to the first user behavior vector to obtain a corresponding first user tag classification vector; and the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the first user label classification vector form a corresponding first user portrait and are stored.
Preferably, the processing the user tracking image using the second classification model specifically includes:
Counting the behavior data of the first user once according to all the behavior types corresponding to the user behavior vector at intervals of a designated time to generate corresponding first time behavior data; and forming a corresponding first time behavior data sequence by a plurality of first time behavior data in the last specified time length at intervals of the specified time length; tensor setting is carried out according to the tensor structure of the user behavior time sequence tensor according to the first time behavior data sequence to obtain a corresponding first user behavior time sequence tensor; the second classification model predicts the user tag type according to the first user behavior time sequence tensor to obtain a corresponding second user tag classification vector; and the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the second user label classification vector form a corresponding current user portrait; and identifying whether the current user representation matches the stored first user representation; if the first user portrait is not matched, updating the stored first user portrait to the current user portrait; the specified time length is greater than the specified time interval and is an integer multiple of the specified time length.
Preferably, the processing the user portrait adjustment based on task feedback by using the third classification model specifically includes:
Periodically customizing a first user mission plan with a plurality of first sub-mission plans for the first user according to the latest first user portrait; selecting a corresponding first business person for each first sub-task plan; distributing each first subtask plan to a corresponding first business person for task execution, and receiving a first subtask feedback score fed back by each first business person when the task is finished; vector setting is carried out according to the vector structure of the user task feedback vector and the obtained first subtask feedback scores to obtain corresponding first user task feedback vectors; the third classification model predicts the user tag type according to the first user task feedback vector to obtain a corresponding third user tag classification vector; and the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the third user label classification vector form a corresponding current user portrait; and identifying whether the current user representation matches the stored first user representation; if not, updating the stored first user portrait to the current user portrait.
A second aspect of the embodiment of the present invention provides an apparatus for implementing the method for processing a user portrait based on an AI model according to the first aspect, where the apparatus includes: an AI model preparation module and an AI model representation module;
The AI model preparation module is used for configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database; the first classification model is used for carrying out user tag type prediction processing according to the input user behavior vector to obtain a corresponding user tag classification vector; the second classification model is used for carrying out user tag type prediction processing according to the input user behavior time sequence tensor to obtain the corresponding user tag classification vector; the third classification model is used for carrying out user tag type prediction processing according to the input user task feedback vector to obtain a corresponding user tag classification vector;
The AI model portrayal module is used for processing an initial user image by using the first classification model after training the first, second and third classification models; processing the user tracking image by using the second classification model; and processing user portrayal adjustment based on task feedback using the third classification model.
A third aspect of an embodiment of the present invention provides an electronic device, including: memory, processor, and transceiver;
the processor is configured to couple to the memory, and read and execute the instructions in the memory, so as to implement the method steps described in the first aspect;
The transceiver is coupled to the processor and is controlled by the processor to transmit and receive messages.
A fourth aspect of the embodiments of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions of the method of the first aspect.
The embodiment of the invention provides a processing method and device for user portrait based on an AI model, electronic equipment and a computer readable storage medium. As can be seen from the foregoing, in the embodiment of the present invention, three classification models, i.e., the first, second and third classification models, are first constructed based on the structure of MLP network+softmax, and the data formats of the input vectors/tensors of the first, second and third classification models are planned, and the first, second and third classification models are trained based on the stock user database stored by the business party and used for recording the manual processing information of the stock clients; and after training, processing the initial portraits of newly added users by using a first classification model, processing the tracking portraits by using a second classification model in the continuous maintenance stage of the users, and processing user portraits adjustment based on feedback by using a third classification model when all feedback information of the user task plans is received. The embodiment of the invention improves the real-time performance and the processing efficiency of the user portrait.
Drawings
FIG. 1 is a schematic diagram of a user portrait processing method based on an AI model according to an embodiment of the present invention;
FIG. 2a is a block diagram of a first classification model according to an embodiment of the invention;
FIG. 2b is a block diagram of a second classification model according to an embodiment of the invention;
FIG. 2c is a block diagram of a third classification model according to an embodiment of the invention;
FIG. 3 is a block diagram of a processing device for user portrayal based on an AI model according to a second embodiment of the present invention;
Fig. 4 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The first embodiment of the invention provides a processing method for user portraits based on an AI model, and a service provider can improve the real-time performance and the processing efficiency of the user portraits through the first embodiment of the invention; fig. 1 is a schematic diagram of a user portrait processing method based on an AI model according to a first embodiment of the present invention, where, as shown in fig. 1, the method mainly includes the following steps:
Step 1, configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database;
the method specifically comprises the following steps: step 11, configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector;
Here, the input/output vector/tensor structure corresponding to the three types of models in the embodiment of the present invention is as follows:
1) The user behavior vector is a one-dimensional vector with the shape of 1 XW 1, and the vector length W 1 is an integer greater than or equal to 1; the user behavior vector comprises W 1 behavior type grades; each behavior type level corresponds to a behavior type; each behavior type corresponds to a behavior grade value range; each behavior grade value range comprises a plurality of behavior grade values; each behavior grade takes the value as an integer greater than zero; each behavior type grade is one behavior grade value in a behavior grade value range corresponding to the corresponding behavior type; the total number of the behavior grade values of all behavior types corresponding to the user behavior vector is matched with the vector length W 1; the user behavior vector of the embodiment of the invention adopts a single-hot coding mechanism for coding, namely, the vector data of the effective position in the vector is 1, and the vector data of the ineffective position is 0;
2) The user behavior time sequence tensor is a two-dimensional tensor with the shape of H multiplied by W 2, the tensor height H is an integer greater than 1, and the tensor width W 2 is an integer greater than or equal to 2, and W 2=1+W1; the user behavior time sequence tensor consists of H user behavior time sequence vectors; the user behavior time sequence vector consists of a time stamp and a user behavior vector corresponding to the current time stamp;
3) The user task feedback vector is a one-dimensional vector with the shape of 1 XW 3, and the vector length W 3 is an integer greater than or equal to 1; the user task feedback vector comprises W 3 task type feedback levels; each task type feedback level corresponds to one task type; each task type corresponds to a task feedback level value range; each task feedback level value range comprises a plurality of task feedback level values; each task feedback level takes an integer greater than zero; the task type feedback level is a task feedback level value in a task feedback level value range corresponding to the corresponding task type; the total number of the task feedback grade values of all task types corresponding to the user task feedback vector is matched with the vector length W 3; the user task feedback vector of the embodiment of the invention adopts a single-hot coding mechanism for coding, namely, the vector data of the effective position in the vector is 1, and the vector data of the ineffective position is 0;
4) The user tag classification vector is a one-dimensional vector with the shape of 1 XW 4, and the vector length W 4 is an integer greater than 1; the user behavior vector comprises W 4 user tag prediction probabilities; each user tag prediction probability corresponds to a user type and a user grade; each user type corresponds to a user grade value range; each user level value range includes a plurality of user level values; each user level takes a value of an integer greater than zero; each user grade is one user grade value in a user grade value range corresponding to the corresponding user type; the prediction probability of each user label is the prediction probability of the corresponding user type plus the user grade; the total number of the user grade values of all the user types corresponding to the user label classification vector is matched with the vector length W 4;
Step 12, constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model;
Here, the first classification model of the embodiment of the present invention is used for performing a user tag type prediction process according to an input user behavior vector to obtain a corresponding user tag classification vector; as shown in fig. 2a, which is a block diagram of a first classification model according to a first embodiment of the present invention, the first classification model includes a first MLP network and a first Softmax layer; the first MLP network is composed of an input layer, a hidden layer and an output layer; the first Softmax layer is formed by a Softmax function; the first MLP network is used for carrying out feature extraction on the user behavior vector input by the model to obtain a corresponding first feature vector and sending the corresponding first feature vector to the first Softmax layer; the first Softmax layer is used for predicting classification probability according to the first feature vector to obtain a corresponding user tag classification vector;
The second classification model of the embodiment of the invention is used for carrying out user tag type prediction processing according to the input user behavior time sequence tensor to obtain a corresponding user tag classification vector; as shown in fig. 2b, which is a block diagram of a second classification model according to a first embodiment of the present invention, the second classification model includes a second MLP network and a second Softmax layer; the second MLP network is composed of an input layer, a hidden layer and an output layer; the second Softmax layer is formed by a Softmax function; the second MLP network is used for carrying out feature extraction on the user behavior time sequence tensor input by the model to obtain a corresponding second feature vector and sending the corresponding second feature vector to the second Softmax layer; the second Softmax layer is used for predicting the classification probability according to the second feature vector to obtain a corresponding user tag classification vector;
the third classification model of the embodiment of the invention is used for carrying out user tag type prediction processing according to the input user task feedback vector to obtain a corresponding user tag classification vector; as shown in fig. 2c, which is a block diagram of a third classification model according to a first embodiment of the present invention, the third classification model includes a third MLP network and a third Softmax layer; the third MLP network is composed of an input layer, a hidden layer and an output layer; the third Softmax layer is formed by a Softmax function; the third MLP network is used for extracting the characteristics of the user task feedback vector input by the model to obtain a corresponding third characteristic vector and sending the corresponding third characteristic vector to the third Softmax layer; the third Softmax layer is used for predicting the classification probability according to the third feature vector to obtain a corresponding user tag classification vector;
Step 13, training the first, second and third classification models based on a preset stock user database;
here, before explaining the detailed processing procedure of the current step, the stock user database needs to be explained as follows:
The stock user database in the embodiment of the invention is a database stored by a current service provider and used for recording the manual processing information of the stock clients, and comprises a first user information acquisition library, a first row of data acquisition library and a first task feedback data acquisition library; the first user information collection library is used for recording historical information of each stock client type level in a manual mode; the first row of data acquisition library is used for storing historical information for carrying out behavior data tracking acquisition on each stock client; the first task feedback data acquisition library is used for storing feedback scores of each user task plan of each stock client and manually estimated type grade change suggestions;
1) The first user information acquisition library comprises a plurality of first user information tables;
Each first user information table corresponds to a first stock user;
the first user information table includes one or more first type level update records;
The first type level update record includes a first update time field, a first latest type field, and a first latest level field; wherein the time information of the first update time field comprises year, month, day, time and second information; the first latest type field is a user type; the first latest grade field is one user grade value in a user grade value range corresponding to the corresponding first latest grade field;
2) The first row of data acquisition library comprises a plurality of first data acquisition tables;
the first data acquisition table corresponds to the first stock users one by one;
the first data acquisition table comprises a plurality of first acquisition records;
The first acquisition record comprises a first acquisition time field and a plurality of first behavior acquisition fields; the time information of the first acquisition time field comprises year, month, day, time and second information; the first behavior acquisition field comprises a first behavior type and first acquisition data; the first behavior type is a behavior type and at least comprises an information browsing behavior, an information focusing behavior, an information sharing behavior and a service communication behavior; the first collected data is behavior collected data corresponding to a first behavior type, and specifically: a) If the first behavior type is an information browsing behavior, the corresponding first acquired data is a total browsing duration obtained by counting the browsing total duration of browsing the information of the appointed field of the corresponding first stock user before the current time point on the time point of the corresponding first acquisition time field; wherein the designated domain at least comprises a financial service domain, an insurance service domain and a telecommunication service domain; the appointed domain information at least comprises an appointed domain article, an appointed domain image, an appointed domain audio/video, an appointed domain public number, an appointed domain micro signal, an appointed domain micro blog number and an appointed domain website; b) If the first behavior type is information attention behavior, the corresponding first acquired data is the total attention quantity obtained by counting the total number of the information of the appointed field which is currently concerned by the corresponding first stock user on the time point of the corresponding first acquisition time field; c) If the first behavior type is an information sharing behavior, the corresponding first acquired data is a total sharing number obtained by counting the total number of the appointed field information shared by the corresponding first stock users before the current time point on the time point of the corresponding first acquisition time field; d) If the first behavior type is a business communication behavior, the corresponding first acquired data is a total communication duration obtained by counting the total duration of communication between the corresponding first stock user and business personnel in the appointed field before the current time point at the time point of the corresponding first acquisition time field;
3) The first task feedback data acquisition library comprises a plurality of second data acquisition tables;
the second data acquisition table corresponds to the first stock users one by one;
the second data acquisition table comprises a plurality of second acquisition records;
Each second acquisition record corresponds to a user task plan; each user mission plan includes a plurality of sub-mission plans and a user type level change suggestion; each subtask plan includes a subtask type, a subtask flow specification, and a subtask feedback score; each user mission plan is cooperatively completed by a business director and one or more business personnel; each subtask plan is executed by a business person according to the corresponding subtask flow description, an execution object is a corresponding first stock user, when the task is ended, the current business person feeds back corresponding subtask feedback scores according to the task execution result, and after all the subtask plans are fed back, a business director sets corresponding user type grade change suggestions according to all the subtask feedback scores; the user type level change suggestions include keeping the user type level and new user type level; the new user type grade consists of a user type and a corresponding user grade value;
The second acquisition record comprises a second acquisition time field, a plurality of first feedback acquisition fields and a first type level change suggestion field; the time information of the second acquisition time field comprises year, month, day, time and second information; the first feedback acquisition field corresponds to the corresponding subtask plan of the user task plan one by one; the first feedback acquisition field comprises a first task type and first feedback data; the first task type corresponds to a subtask type of the corresponding subtask plan; the first feedback data corresponds to the subtask feedback scores of the corresponding subtask plans; the first type level change suggestion field includes a hold-down and a new user type level;
the following is a specific processing step of the current step 13:
The method specifically comprises the following steps: step 131, constructing a training data set based on the stock user database to obtain a corresponding first model training data set, second model training data set and third model training data set;
Wherein the first model training data set comprises a plurality of first model training data; the first model training data includes a first model training vector and a first model tag vector; the second model training data set includes a plurality of second model training data; the second model training data includes a second model training tensor and a second model label vector; the third model training data set includes a plurality of third model training data; the third model training data includes a third model training vector and a third model tag vector;
The method specifically comprises the following steps: step 1311, initializing a user tag classification vector based on all user types of the stock user database and user grade value ranges corresponding to the user types;
The vector length W 4 of the initialized user tag classification vector is the total number of user grade values of all user types; the user label prediction probability of the initialized user label classification vector corresponds to a user type and a user grade value;
step 1312, and designating a corresponding collected data level standardization standard for the first collected data corresponding to each type of first behavior of the stock user database;
Wherein, the collected data grade standardization standard consists of a plurality of classification behavior grade corresponding relations and a classification behavior grade value range, each classification behavior grade corresponding relation comprises a collected data range and a corresponding classification behavior grade value, wherein the classification behavior grade value range consists of a plurality of classification behavior grade values;
If the first behavior type is specifically an information browsing behavior, each classification behavior grade corresponding relation comprises a browsing duration range and a corresponding browsing behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the browsing behavior grade value one by one;
If the first behavior type is information attention behavior, each classification behavior grade corresponding relation comprises an attention quantity range and a corresponding attention behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the attention behavior grade value one by one;
if the first behavior type is an information sharing behavior, each classification behavior grade corresponding relation comprises a sharing quantity range and a corresponding sharing behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the sharing behavior grade value one by one;
If the first behavior type is a business communication behavior, each classification behavior grade corresponding relation comprises a communication duration range and a corresponding communication behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the communication behavior grade value one by one;
step 1313, performing structural initialization on the user behavior vector based on all the first behavior types and the classified behavior class value ranges corresponding to the first behavior types;
The vector length W 1 of the initialized user behavior vector is the total number of the classified behavior grades of all the first behavior types; each behavior type grade of the initialized user behavior vector consists of a first behavior type and a corresponding classified behavior grade value;
1314, initializing the tensor height H of the user behavior time sequence tensor based on the preset second model input Zhang Lianggao degrees, and initializing the tensor width W 2 of the user behavior time sequence tensor by the initialized vector length W 1 of the user behavior vector;
Wherein the second model input Zhang Lianggao degrees is a preset integer greater than 1; the tensor height H of the initialized user behavior time sequence tensor is consistent with the input Zhang Lianggao degrees of the second model, and the tensor width W 2 is the sum of the vector length W 1 and 1;
Step 1315, assigning a corresponding feedback data level normalization standard to the subtask feedback scores corresponding to the subtask types of the stock user database;
The feedback data level standardization standard comprises a plurality of subtask feedback level corresponding relations and a subtask feedback level value range, each subtask feedback level corresponding relation comprises a feedback scoring range and a corresponding subtask feedback level value, and all subtask feedback level values form the corresponding subtask feedback level value range;
Step 1316, performing structural initialization on the user task feedback vector based on all subtask types and subtask feedback level value ranges corresponding to the subtask types;
The vector length W 3 of the initialized user task feedback vector is the total number of subtask feedback grade values of all subtask types; each task type feedback level of the initialized user task feedback vector consists of a subtask type and a subtask feedback level value corresponding to the subtask type;
Step 1317, traversing all first acquisition records of the stock user database; the first acquisition record of the current traversal is used as a corresponding current record in the traversal process; the first stock user corresponding to the current record is used as the corresponding current user; taking the first acquisition time field of the current record as the corresponding current time; recording a first type grade update record with a first update time field earlier than the current time in a first user information table corresponding to the current user in a first user information acquisition library of the stock user database as a corresponding first early update record, and extracting a first latest type field and a first latest grade field of a first early update record with the first update time field closest to the current time in all the first early update records as corresponding matched user types and matched user grades to form corresponding matched user type grades; setting a user tag classification vector for model training as a corresponding first model tag vector according to the class of the matched user type based on the vector structure of the user tag classification vector; corresponding level normalization processing is carried out on each corresponding first acquired data in the current record based on acquired data level normalization standards corresponding to each first behavior type, so as to obtain corresponding first classification behavior level values; and the corresponding first behavior type grade is formed by each first behavior type and the corresponding first classification behavior grade value recorded at present; setting a user behavior vector for model training according to all first behavior type grades recorded currently based on a vector structure of the user behavior vector, and recording the user behavior vector as a corresponding first model training vector; the first model training vector and the first model label vector form corresponding first model training data; when the traversing is finished, forming a corresponding first model training data set by all the obtained first model training data;
Step 1318, and forming a corresponding first vector set from all the obtained first model training vectors; the first model training vectors of the first vector sets, which belong to all first stock clients, are gathered into a class to form corresponding first user vector sets; sequencing all the first model training vectors of each first user vector set according to the sequence from far to near of the first acquisition time fields of the corresponding first acquisition records to obtain a corresponding first user vector sequence;
Step 1319, and traversing all the obtained first user vector sequences; traversing, and taking the first user vector sequence traversed currently as a corresponding current sequence; the tensor height H of the user behavior time sequence tensor is used as the corresponding segmentation sequence length; starting from a first model training vector of the current sequence, carrying out sub-sequence sliding segmentation processing on the current sequence by using a preset segmentation sliding step length and a preset segmentation sequence length to obtain a plurality of corresponding first sub-sequences; setting a corresponding training label vector as a corresponding second model label vector based on the vector structure of the matched user type grade and the user label classification vector corresponding to the last first model training vector of each first subsequence; and polling each first sub-sequence; when polling, the first subsequence of the current polling is used as a corresponding current subsequence, each first model training vector in the current subsequence and a first acquisition time field of a corresponding first acquisition record form a corresponding first user behavior time sequence vector, and all first user behavior time sequence vectors corresponding to the current subsequence form a user behavior time sequence tensor for model training and are recorded as a corresponding second model training tensor; forming corresponding second model training data by each second model training tensor and corresponding second model label vector; when the traversing is finished, forming a corresponding second model training data set by all the obtained second model training data;
Here, the slicing sliding step length is an integer which is preset to be greater than or equal to 1;
Step 1320, traversing all second acquisition records of the stock user database; traversing, and taking the second acquisition record of the current traversal as a corresponding current record; the first stock user corresponding to the current record is used as the corresponding current user; taking the second acquisition time field of the current record as the corresponding current time; recording a first type grade update record of which the first update time field is earlier than the current time in a first user information table corresponding to the current user in a first user information acquisition library as a corresponding second early update record, and forming a first latest type field and a first latest grade field of the first early update record of which the first update time field is closest to the current time in all the second early update records into a corresponding original user type grade; identifying whether a first type level change suggestion field of the current record is kept as it is, if so, setting a user tag classification vector for model training according to the original user type level based on the vector structure of the user tag classification vector to be a corresponding third model tag vector, otherwise, setting a user tag classification vector for model training according to the first type level change suggestion field of the current record based on the vector structure of the user tag classification vector to be a corresponding third model tag vector; corresponding level normalization processing is carried out on corresponding first feedback data in the current record based on feedback data level normalization standards corresponding to various subtask types, so that corresponding first subtask feedback level values are obtained; the corresponding first subtask type grade is formed by each subtask type recorded currently and the corresponding first subtask feedback grade value; setting a user task feedback vector for model training according to all the first subtask type grades recorded currently based on the vector structure of the user task feedback vector, and marking the user task feedback vector as a corresponding third model training vector; and forming corresponding third model training data by the obtained third model training vector and the corresponding third model label vector; when the traversing is finished, forming a corresponding third model training data set by all the obtained third model training data;
step 132, training the first, second and third classification models based on the first, second and third model training data sets, respectively;
The method specifically comprises the following steps: step 1321, sequentially traversing first model training data of the first model training data set; traversing, wherein the first model training data in the current traversal is used as corresponding current training data; inputting a first model training vector of the current training data into a first classification model to perform user label type prediction processing to obtain a corresponding first model prediction vector; inputting a first model predictive vector and a first model label vector of current training data into a preset first loss function, and carrying out primary parameter optimization on the first classification model based on a preset first model parameter optimizer towards the direction of enabling the first loss function to reach the minimum value; when the parameter optimization is finished, the next first model training data is transferred to continue training until the current training data is the last first model training data of the first model training data set;
Wherein the first loss function is a cross entropy loss function; the first model parameter optimizer comprises an SGD optimizer and an Adam optimizer;
Step 1322, and traversing the second model training data of the second model training data set in sequence; traversing, and taking the second model training data of the current traversal as corresponding current training data; inputting a second model training tensor of the current training data into a second classification model to perform user label type prediction processing to obtain a corresponding second model prediction vector; inputting a second model predictive vector and a second model label vector of the current training data into a preset second loss function, and carrying out primary parameter optimization on the second classification model based on a preset second model parameter optimizer towards the direction of enabling the second loss function to reach the minimum value; when the parameter optimization is finished, the next second model training data is transferred to continue training until the current training data is the last second model training data of the second model training data set;
Wherein the second loss function is a cross entropy loss function; the second model parameter optimizer comprises an SGD optimizer and an Adam optimizer;
Step 1323, and traversing the third model training data of the third model training data set in sequence; traversing, and taking the currently traversed third model training data as corresponding current training data; inputting a third model training vector of the current training data into a third classification model to perform user label type prediction processing to obtain a corresponding third model prediction vector; inputting a third model predictive vector and a third model label vector of the current training data into a preset third loss function, and carrying out primary parameter optimization on a third classification model based on a preset third model parameter optimizer towards the direction of enabling the third loss function to reach the minimum value; when the parameter optimization is finished, the method is switched to the next third model training data to continue training until the current training data is the last third model training data of the third model training data set;
wherein the third loss function is a cross entropy loss function; the third model parameter optimizer includes an SGD optimizer and Adam optimizer.
Step 2, after training the first, second and third classification models, processing an initial image of the user by using the first classification model; processing the user tracking image by using the second classification model; and processing user portrayal adjustment based on task feedback using a third classification model;
The method specifically comprises the following steps: step 21, after training the first, second and third classification models, processing the initial image of the user by using the first classification model;
the method specifically comprises the following steps: step 211, recording the current newly added user as a corresponding first user when one user is newly added each time;
step 212, performing primary initial behavior data statistics on the first user according to all behavior types corresponding to the user behavior vector to generate corresponding first initial behavior data;
For example, the appointed domain is a financial service domain, and the behavior types corresponding to the user behavior vector include an information browsing behavior of the financial service domain, an information attention behavior of the financial service domain, an information sharing behavior of the financial service domain and a business communication behavior of the financial service domain; the first user performs primary initial behavior data statistics, namely, estimates or counts the browsing total duration of browsing financial service domain information in the last time period of the first user, estimates or counts the total attention quantity of financial service domain information in the last time period of the first user, estimates or counts the total sharing quantity of financial service domain information shared in the last time period of the first user, and estimates or counts the total communication duration of communication with business personnel in the financial service domain in the last time period of the first user; then, based on four types of collected data grade standardization standards respectively corresponding to the four types of behaviors, carrying out grade division according to the counted four types of behavior statistical data (browsing total duration, total attention quantity, total sharing quantity and total communication duration) to obtain corresponding four types of behavior grade values; the corresponding behavior type grade can be formed by various behavior types and corresponding behavior grade values, and the first initial behavior data is formed by four behavior type grades;
step 213, performing vector setting according to the first initial behavior data according to the vector structure of the user behavior vector to obtain a corresponding first user behavior vector;
For example, given that the total number of behavior types of the user behavior vector is 4, the range of values of the classification behavior levels corresponding to each class of behavior includes 4 values of the classification behavior levels, and then the vector length of the user behavior vector is 4×4=16, in the embodiment of the present invention, vector encoding is performed by adopting one-hot encoding by default, that is, in the user behavior vector with the length of 16, only four vector data corresponding to four behavior type levels of the first initial behavior data are 1, and the rest are all 0;
step 214, performing user tag type prediction processing according to the first user behavior vector by the first classification model to obtain a corresponding first user tag classification vector;
Step 215, forming a corresponding first user portrait by the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the first user label classification vector and storing the first user portrait;
step 22, processing the user tracking image by using the second classification model;
The method specifically comprises the following steps: step 221, counting the behavior data of the first user once according to all behavior types corresponding to the user behavior vector at intervals of a specified time to generate corresponding first time behavior data;
here, the specified time interval is a sampling time interval set in advance;
The data format of the first time behavior data is similar to the data format of the first initial behavior data and is also composed of four behavior type grades; however, the data meaning of the four types of behavior statistics (the browsing total time length, the total attention amount, the total sharing amount and the total communication time length) for evaluating the four types of behavior grades is slightly different from the data meaning of the four types of behavior statistics in the first initial behavior data; the browsing total duration of the current step is the total duration of the financial service field information browsed by the first user before the current time point, and the data can be automatically acquired through the service APP of the current service provider installed by the current user; the total attention quantity of the current step is the total quantity of the information of the financial service field which is paid attention to by the first user at the current point, and the data can be automatically acquired through the service APP; the total sharing quantity of the current step is the total quantity of the financial service field information which is already shared by the first user at the current time point, and the data can be automatically acquired through the service APP; the total communication duration of the current step is the total duration of the first user communicating with the service personnel of the current service provider before the current time point, and the data can be automatically acquired through the service APP;
Step 222, forming a corresponding first time behavior data sequence by a plurality of first time behavior data within the last specified time length at intervals of the specified time length;
Here, the specified time length is a preset time length parameter; the specified time length is greater than the specified time interval and is an integer multiple of the specified time length;
Step 223, performing tensor setting according to the first time behavior data sequence according to the tensor structure of the user behavior time sequence tensor to obtain a corresponding first user behavior time sequence tensor;
Step 224, performing a user tag type prediction process by the second classification model according to the first user behavior time sequence tensor to obtain a corresponding second user tag classification vector;
225, forming a corresponding current user portrait by the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the second user label classification vector;
Step 226, and identify whether the current user representation matches the stored first user representation; if the first user portrait is not matched, updating the stored first user portrait to the current user portrait;
step 23, processing user portrait adjustment based on task feedback by using a third classification model;
The method specifically comprises the following steps: step 231, customizing a first user mission plan with a plurality of first sub-mission plans for the first user according to the latest first user portrait; selecting a corresponding first business person for each first sub-task plan; distributing each first subtask plan to a corresponding first business person for task execution, and receiving a first subtask feedback score fed back by each first business person when the task is finished;
Step 232, performing vector setting according to the obtained feedback scores of all the first subtasks according to the vector structure of the user task feedback vector to obtain a corresponding first user task feedback vector;
Here, in the embodiment of the present invention, the first subtask feedback scores are classified based on the feedback data class normalization criteria corresponding to each first subtask feedback score, so that a corresponding task type class can be obtained, and thus, a user task feedback vector can be set;
for example, given that the total number of subtask types of the user task feedback vector is 3, the subtask feedback level value range corresponding to each type of subtask includes 5 subtask feedback level values, and then the vector length of the user task feedback vector is 3×5=15; as is also known, the first user task plan only includes the first sub-task plans corresponding to the two sub-task types, that is, two task type levels are obtained, and the embodiment of the invention adopts the one-hot coding to perform vector coding by default, that is, only 2 vector data in the user task feedback vector with the length of 15 are 1, and the rest are all 0;
Step 233, performing user tag type prediction processing by the third classification model according to the first user task feedback vector to obtain a corresponding third user tag classification vector;
Step 234, the corresponding current user portrait is formed by the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the third user label classification vector;
step 235, identifying whether the current user portrait matches the stored first user portrait; if the user images do not match, the stored first user image is updated to the current user image.
Fig. 3 is a block diagram of a processing apparatus for performing user portrait based on AI model according to a second embodiment of the present invention, where the apparatus is a terminal device or a server for implementing the foregoing method embodiment, or may be an apparatus capable of enabling the foregoing terminal device or the server to implement the foregoing method embodiment, and the apparatus may be an apparatus or a chip system of the foregoing terminal device or the server, for example. As shown in fig. 3, the AI model-based processing device for user representation includes: an AI model preparation module 201 and an AI model representation module 202.
The AI model preparation module 201 is configured to configure a data structure of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database; the first classification model is used for carrying out user tag type prediction processing according to the input user behavior vector to obtain a corresponding user tag classification vector; the second classification model is used for carrying out user tag type prediction processing according to the input user behavior time sequence tensor to obtain a corresponding user tag classification vector; and the third classification model is used for carrying out user tag type prediction processing according to the input user task feedback vector to obtain a corresponding user tag classification vector.
The AI model portrayal module 202 is configured to process the user initial portrayal using the first classification model after training the first, second, and third classification models; processing the user tracking image by using the second classification model; and processing user portrayal adjustment based on task feedback using a third classification model.
The processing device for performing user portrait based on the AI model provided by the embodiment of the invention can execute the method steps in the method embodiment, and the implementation principle and the technical effect are similar and are not repeated here.
It should be noted that, it should be understood that the division of the modules of the above apparatus is merely a division of a logic function, and may be fully or partially integrated into a physical entity or may be physically separated. And these modules may all be implemented in software in the form of calls by the processing element; or can be realized in hardware; the method can also be realized in a form of calling software by a processing element, and the method can be realized in a form of hardware by a part of modules. For example, the AI model portrait module may be a processing element which is set up separately, may be implemented in a chip of the above-described apparatus, or may be stored in a memory of the above-described apparatus in the form of program codes, and may be called by a processing element of the above-described apparatus to execute the functions of the above-described determination module. The implementation of the other modules is similar. In addition, all or part of the modules can be integrated together or can be independently implemented. The processing element described herein may be an integrated circuit having signal processing capabilities. In implementation, each step of the above method or each module above may be implemented by an integrated logic circuit of hardware in a processor element or an instruction in a software form.
For example, the modules above may be one or more integrated circuits configured to implement the methods above, such as: one or more Application SPECIFIC INTEGRATED Circuits (ASIC), or one or more digital signal processors (DIGITAL SIGNAL Processor, DSP), or one or more field programmable gate arrays (Field Programmable GATE ARRAY, FPGA), etc. For another example, when a module above is implemented in the form of processing element scheduler code, the processing element may be a general purpose processor, such as a central processing unit (Central Processing Unit, CPU) or other processor that may invoke the program code. For another example, the modules may be integrated together and implemented in the form of a System-on-a-chip (SOC).
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces, in whole or in part, the processes or functions described in connection with the foregoing method embodiments. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, fiber optic, digital subscriber line ((Digital Subscriber Line, DSL)), or wireless (e.g., infrared, wireless, bluetooth, microwave, etc.), or a wireless (e.g., infrared, wireless, bluetooth, microwave, etc.), the computer-readable storage medium may be any available medium that can be accessed by the computer or a data storage device such as a server, data center, etc., that contains an integration of one or more available media.
Fig. 4 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention. The electronic device may be a terminal device or a server implementing the method of the foregoing embodiment, or may be a terminal device or a server implementing the method of the foregoing embodiment, which is connected to the foregoing terminal device or server. As shown in fig. 4, the electronic device may include: a processor 301 (e.g., a CPU), a memory 302, a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiving actions of the transceiver 303. The memory 302 may store various instructions for performing the various processing functions and implementing the processing steps described in the methods of the previous embodiments. Preferably, the electronic device according to the embodiment of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to implement communication connections between the elements. The communication port 306 is used for connection communication between the electronic device and other peripheral devices.
The system bus 305 referred to in fig. 4 may be a peripheral component interconnect standard (PERIPHERAL COMPONENT INTERCONNECT, PCI) bus, or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, or the like. The system bus may be classified into an address bus, a data bus, a control bus, and the like. For ease of illustration, the figures are shown with only one bold line, but not with only one bus or one type of bus. The communication interface is used to enable communication between the database access apparatus and other devices (e.g., clients, read-write libraries, and read-only libraries). The Memory may include random access Memory (Random Access Memory, RAM) and may also include Non-Volatile Memory (Non-Volatile Memory), such as at least one disk Memory.
The processor may be a general-purpose processor, including a Central Processing Unit (CPU), a network processor (Network Processor, NP), a graphics processor (Graphics Processing Unit, GPU), etc.; but may also be a digital signal processor DSP, an application specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.
It should be noted that, the embodiments of the present invention also provide a computer readable storage medium, where instructions are stored, when the computer readable storage medium runs on a computer, to cause the computer to perform the method and the process provided in the above embodiments.
The embodiment of the invention also provides a chip for running the instructions, and the chip is used for executing the processing steps described in the embodiment of the method.
The embodiment of the invention provides a processing method and device for user portrait based on an AI model, electronic equipment and a computer readable storage medium; as can be seen from the above summary of the invention, in the embodiment of the present invention, three classification models, namely, a first classification model, a second classification model and a third classification model, are first constructed based on the structure of MLP network+softmax, the data formats of input vectors/tensors of the first classification model, the second classification model and the third classification model are planned, and the first classification model, the second classification model and the third classification model are trained based on an inventory user database stored by a business party and used for recording manual processing information of an inventory client; and after training, processing the initial portraits of newly added users by using a first classification model, processing the tracking portraits by using a second classification model in the continuous maintenance stage of the users, and processing user portraits adjustment based on feedback by using a third classification model when all feedback information of the user task plans is received. The embodiment of the invention improves the real-time performance and the processing efficiency of the user portrait.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative elements and steps are described above generally in terms of function in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, in a software module executed by a processor, or in a combination of the two. The software modules may be disposed in Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the invention, and is not meant to limit the scope of the invention, but to limit the invention to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (8)
1. A method for processing a user portrait based on an AI model, the method comprising:
Configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database; the first classification model is used for carrying out user tag type prediction processing according to the input user behavior vector to obtain a corresponding user tag classification vector; the second classification model is used for carrying out user tag type prediction processing according to the input user behavior time sequence tensor to obtain the corresponding user tag classification vector; the third classification model is used for carrying out user tag type prediction processing according to the input user task feedback vector to obtain a corresponding user tag classification vector;
After training the first, second and third classification models, processing an initial image of a user by using the first classification model; processing the user tracking image by using the second classification model; and processing user portrayal adjustment based on task feedback using the third classification model;
Wherein the user behavior vector is a one-dimensional vector with the shape of 1 XW 1, and the vector length W 1 is an integer greater than or equal to 1; the user behavior vector comprises W 1 behavior type grades; each behavior type grade corresponds to one behavior type; each behavior type corresponds to a behavior grade value range; each behavior grade value range comprises a plurality of behavior grade values; each behavior grade takes an integer greater than zero; each behavior type grade is one of the behavior grade value ranges corresponding to the corresponding behavior type; the total number of the behavior grade values of all the behavior types corresponding to the user behavior vector is matched with the vector length W 1;
The user behavior time sequence tensor is a two-dimensional tensor with the shape of H multiplied by W 2, the tensor height H is an integer greater than 1, and the tensor width W 2 is an integer greater than or equal to 2, and W 2=1+W1; the user behavior time sequence tensor consists of H user behavior time sequence vectors; the user behavior time sequence vector consists of a time stamp and the user behavior vector corresponding to the current time stamp;
The user task feedback vector is a one-dimensional vector with the shape of 1 XW 3, and the vector length W 3 is an integer greater than or equal to 1; the user task feedback vector comprises W 3 task type feedback levels; each task type feedback level corresponds to one task type; each task type corresponds to a task feedback level value range; each task feedback level value range comprises a plurality of task feedback level values; each task feedback level takes an integer greater than zero; the task type feedback level is one of the task feedback level value ranges corresponding to the corresponding task type; the total number of the task feedback grade values of all the task types corresponding to the user task feedback vector is matched with the vector length W 3;
The user tag classification vector is a one-dimensional vector with the shape of 1 XW 4, and the vector length W 4 is an integer greater than 1; the user behavior vector comprises W 4 user tag prediction probabilities; each user tag prediction probability corresponds to a user type and a user grade; each user type corresponds to a user grade value range; each user grade value range comprises a plurality of user grade values; each user grade value is an integer greater than zero; each user grade is one user grade value in the user grade value range corresponding to the corresponding user type; the prediction probability of each user label is the prediction probability of the corresponding user type plus the user grade; the total number of the user grade values of all the user types corresponding to the user label classification vector is matched with the vector length W 4;
The processing the initial image of the user by using the first classification model specifically includes:
recording the current newly added user as a corresponding first user when one user is newly added each time; performing primary initial behavior data statistics on the first user according to all the behavior types corresponding to the user behavior vector to generate corresponding first initial behavior data; vector setting is carried out according to the first initial behavior data according to the vector structure of the user behavior vector to obtain a corresponding first user behavior vector; the first classification model predicts the user tag type according to the first user behavior vector to obtain a corresponding first user tag classification vector; and the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the first user label classification vector form a corresponding first user portrait and are stored;
The processing the user tracking image by using the second classification model specifically comprises the following steps:
counting the behavior data of the first user once according to all the behavior types corresponding to the user behavior vector at intervals of a designated time to generate corresponding first time behavior data; and forming a corresponding first time behavior data sequence by a plurality of first time behavior data in the last specified time length at intervals of the specified time length; tensor setting is carried out according to the tensor structure of the user behavior time sequence tensor according to the first time behavior data sequence to obtain a corresponding first user behavior time sequence tensor; the second classification model predicts the user tag type according to the first user behavior time sequence tensor to obtain a corresponding second user tag classification vector; and the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the second user label classification vector form a corresponding current user portrait; and identifying whether the current user representation matches the stored first user representation; if the first user portrait is not matched, updating the stored first user portrait to the current user portrait; the specified time length is greater than the specified time interval and is an integer multiple of the specified time length;
the processing of user portrayal adjustment based on task feedback by using the third classification model specifically comprises:
Periodically customizing a first user mission plan with a plurality of first sub-mission plans for the first user according to the latest first user portrait; selecting a corresponding first business person for each first sub-task plan; distributing each first subtask plan to a corresponding first business person for task execution, and receiving a first subtask feedback score fed back by each first business person when the task is finished; vector setting is carried out according to the vector structure of the user task feedback vector and the obtained first subtask feedback scores to obtain corresponding first user task feedback vectors; the third classification model predicts the user tag type according to the first user task feedback vector to obtain a corresponding third user tag classification vector; and the user type and the user grade corresponding to the user label prediction probability with the maximum probability value in the third user label classification vector form a corresponding current user portrait; and identifying whether the current user representation matches the stored first user representation; if not, updating the stored first user portrait to the current user portrait.
2. The method for processing a user portrait based on an AI model as claimed in claim 1,
The first classification model includes a first MLP network and a first Softmax layer; the first MLP network consists of an input layer, a hidden layer and an output layer; the first Softmax layer is formed by a Softmax function; the first MLP network is used for extracting features of the user behavior vector input by the model to obtain a corresponding first feature vector and sending the corresponding first feature vector to the first Softmax layer; the first Softmax layer is used for predicting classification probability according to the first feature vector to obtain the corresponding user tag classification vector;
The second classification model includes a second MLP network and a second Softmax layer; the second MLP network is composed of an input layer, a hidden layer and an output layer; the second Softmax layer is composed of a Softmax function; the second MLP network is used for carrying out feature extraction on the user behavior time sequence tensor input by the model to obtain a corresponding second feature vector and sending the corresponding second feature vector to the second Softmax layer; the second Softmax layer is used for predicting classification probability according to the second feature vector to obtain the corresponding user tag classification vector;
The third classification model includes a third MLP network and a third Softmax layer; the third MLP network is composed of an input layer, a hidden layer and an output layer; the third Softmax layer is composed of a Softmax function; the third MLP network is used for extracting features of the user task feedback vector input by the model to obtain a corresponding third feature vector and sending the corresponding third feature vector to the third Softmax layer; and the third Softmax layer is used for predicting the classification probability according to the third feature vector to obtain the corresponding user tag classification vector.
3. The method for processing a user portrait based on an AI model as claimed in claim 1,
The stock user database comprises a first user information acquisition library, a first behavior data acquisition library and a first task feedback data acquisition library;
the first user information collection library comprises a plurality of first user information tables; each first user information table corresponds to a first stock user; the first user information table includes one or more first type level update records;
the first type level update record comprises a first update time field, a first latest type field and a first latest level field; the time information of the first updating time field comprises year, month, day, time and second information; the first latest type field is one user type; the first latest grade field is one user grade value in the user grade value range corresponding to the corresponding first latest type field;
The first row of data acquisition library comprises a plurality of first data acquisition tables; the first data acquisition table corresponds to the first stock users one by one; the first data acquisition table comprises a plurality of first acquisition records;
the first acquisition record comprises a first acquisition time field and a plurality of first behavior acquisition fields;
the time information of the first acquisition time field comprises year, month, day, time and second information;
the first behavior acquisition field comprises a first behavior type and first acquisition data;
The first behavior type is one behavior type and at least comprises an information browsing behavior, an information focusing behavior, an information sharing behavior and a business communication behavior;
When the first behavior type is an information browsing behavior, the corresponding first acquired data is a total browsing duration obtained by counting the browsing total duration of the corresponding first stock user browsing the information of the appointed field before the current time point at the time point of the corresponding first acquired time field; the appointed field at least comprises a financial service field, an insurance service field and a telecommunication service field; the appointed domain information at least comprises an appointed domain article, an appointed domain image, an appointed domain audio and video, an appointed domain public number, an appointed domain micro signal, an appointed domain micro blog number and an appointed domain website;
When the first behavior type is information attention behavior, the corresponding first acquired data is total attention quantity obtained by counting the total number of the appointed field information currently focused by the corresponding first stock user at the time point of the corresponding first acquisition time field;
when the first behavior type is an information sharing behavior, the corresponding first acquired data is a total sharing quantity obtained by counting the total number of the specified domain information shared by the corresponding first stock users before the current time point at the time point of the corresponding first acquisition time field;
When the first behavior type is a business communication behavior, the corresponding first acquired data is a total communication duration obtained by counting the total duration of communication between the corresponding first stock user and business personnel in the appointed field before the current time point at the time point of the corresponding first acquisition time field;
The first task feedback data acquisition library comprises a plurality of second data acquisition tables; the second data acquisition table corresponds to the first stock users one by one; the second data acquisition table comprises a plurality of second acquisition records; each second acquisition record corresponds to a user task plan; each user mission plan comprises a plurality of sub-mission plans and a user type grade change suggestion; each subtask plan comprises a subtask type, a subtask flow description and a subtask feedback score; each user task plan is completed by cooperation of a business director and one or more business personnel; each subtask plan is executed by a business person according to the corresponding subtask flow description, an execution object is the corresponding first stock user, when the task is ended, the corresponding subtask feedback scores are fed back by the current business person according to the task execution result, and after all the subtask plans are fed back, the corresponding user type grade change suggestions are set by a business director according to all the subtask feedback scores; the user type grade change suggestion comprises a hold-up state and a new user type grade; the new user type grade consists of one user type and the corresponding user grade value;
The second acquisition record comprises a second acquisition time field, a plurality of first feedback acquisition fields and a first type level change suggestion field;
the time information of the second acquisition time field comprises year, month, day, time and second information;
The first feedback acquisition field corresponds to the corresponding subtask plan of the user task plan one by one; the first feedback acquisition field comprises a first task type and first feedback data; the first task type corresponds to the corresponding subtask type of the subtask plan; the first feedback data corresponds to the subtask feedback score of the corresponding subtask plan;
the first type level change suggestion field includes a hold-down and the new user type level.
4. The AI-model-based user representation processing method of claim 3, wherein the training of the first, second, and third classification models based on a pre-set stock user database specifically comprises:
Constructing a training data set based on the stock user database to obtain a first model training data set, a second model training data set and a third model training data set which correspond to the first model training data set, the second model training data set and the third model training data set; the first model training data set includes a plurality of first model training data; the first model training data comprises a first model training vector and a first model label vector; the second model training data set includes a plurality of second model training data; the second model training data includes a second model training tensor and a second model label vector; the third model training data set includes a plurality of third model training data; the third model training data comprises a third model training vector and a third model label vector;
And training the first, second and third classification models based on the first, second and third model training data sets, respectively.
5. The AI-model-based user representation processing method of claim 4, wherein constructing the training data set based on the stock user database results in corresponding first, second, and third model training data sets, and specifically comprises:
Carrying out structural initialization on the user tag classification vector based on all the user types of the stock user database and the user grade value ranges corresponding to the user types; the initialized vector length W 4 of the user tag classification vector is the total number of the user grade values of all the user types; the initialized user tag prediction probabilities of the user tag classification vectors correspond to one user type and one user grade value;
Designating a corresponding collected data level standardization standard for the first collected data corresponding to each type of the first behavior type of the stock user database; the collected data level standardization standard consists of a plurality of classified behavior level corresponding relations and a classified behavior level value range, each classified behavior level corresponding relation comprises a collected data range and a corresponding classified behavior level value, and the classified behavior level value range consists of a plurality of classified behavior level values; when the first behavior type is specifically an information browsing behavior, each classification behavior grade corresponding relation comprises a browsing duration range and a corresponding browsing behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the browsing behavior grade value one by one; when the first behavior type is information attention behavior, each classification behavior grade corresponding relation comprises an attention quantity range and a corresponding attention behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the attention behavior grade value one by one; when the first behavior type is information sharing behavior, each classification behavior grade corresponding relation comprises a sharing quantity range and a corresponding sharing behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the sharing behavior grade value one by one; when the first behavior type is a business communication behavior, each classification behavior grade corresponding relation comprises a communication duration range and a corresponding communication behavior grade value, and each classification behavior grade value of the classification behavior grade value range corresponds to the communication behavior grade value one by one;
The user behavior vector is structurally initialized based on all the first behavior types and the classified behavior grade value ranges corresponding to the first behavior types; the initialized vector length W 1 of the user behavior vector is the total number of the classified behavior grade values of all the first behavior types; each behavior type grade of the initialized user behavior vector is composed of one first behavior type and one corresponding classified behavior grade value;
Initializing the tensor height H of the user behavior time sequence tensor based on a preset second model input Zhang Lianggao degrees, and initializing the tensor width W 2 of the user behavior time sequence tensor by the initialized vector length W 1 of the user behavior vector; the tensor height H of the initialized user behavior time sequence tensor is consistent with the input Zhang Lianggao degrees of the second model, and the tensor width W 2 is the sum of the vector length W 1 and 1;
Designating a corresponding feedback data level standardization standard for the subtask feedback scores corresponding to the subtask types of the stock user database; the feedback data level standardization standard comprises a plurality of subtask feedback level corresponding relations and a subtask feedback level value range, each subtask feedback level corresponding relation comprises a feedback scoring range and a corresponding subtask feedback level value, and all subtask feedback level values form the corresponding subtask feedback level value range;
The user task feedback vector is structurally initialized based on all the subtask types and the subtask feedback grade value ranges corresponding to the subtask types; the initialized vector length W 3 of the user task feedback vector is the total number of the subtask feedback grade values of all subtask types; each task type feedback level of the initialized user task feedback vector is composed of one subtask type and a subtask feedback level value corresponding to the subtask type;
traversing all the first acquisition records of the stock user database; and traversing, wherein the first acquisition record of the current traversal is used as a corresponding current record; the first stock user corresponding to the current record is used as a corresponding current user; the first acquisition time field of the current record is used as the corresponding current time; the first type level update record of the first user information collection library of the stock user database, which corresponds to the current user, is marked as a corresponding first early update record, and the first latest type field and the first latest level field of the first early update record, which are closest to the current time, in all the first early update records are extracted to be used as corresponding matched user types and matched user levels to form corresponding matched user type levels; setting a user tag classification vector for model training according to the matching user type grade based on the vector structure of the user tag classification vector as a corresponding first model tag vector; performing corresponding level normalization processing on each corresponding first acquired data in the current record based on the acquired data level normalization standard corresponding to each first behavior type to obtain a corresponding first classification behavior level value; and the corresponding first behavior type grade is formed by each first behavior type of the current record and the corresponding first classification behavior grade value; setting a user behavior vector for model training as a corresponding first model training vector according to all first behavior type grades recorded currently based on a vector structure of the user behavior vector; the first model training vector and the first model label vector are obtained to form corresponding first model training data; when the traversing is finished, the corresponding first model training data set is formed by all the obtained first model training data;
and forming a corresponding first vector set by all the obtained first model training vectors; the first model training vectors belonging to the first stock users in the first vector set are gathered into a first user vector set corresponding to the first model training vectors; sequencing all the first model training vectors of each first user vector set according to the sequence from far to near of the first acquisition time fields of the corresponding first acquisition records to obtain a corresponding first user vector sequence;
Traversing all the obtained first user vector sequences; traversing, wherein the first user vector sequence traversed currently is used as a corresponding current sequence; the tensor height H of the user behavior time sequence tensor is used as the corresponding segmentation sequence length; starting from a first model training vector of the current sequence, carrying out sub-sequence sliding segmentation processing on the current sequence by using a preset segmentation sliding step length and the segmentation sequence length to obtain a plurality of corresponding first sub-sequences; setting a corresponding training label vector as a corresponding second model label vector based on the vector structure of the matching user type grade corresponding to the last first model training vector of each first subsequence and the user label classification vector; and polling each of the first sub-sequences; when polling, the first subsequence of the current polling is used as a corresponding current subsequence, a corresponding first user behavior time sequence vector is formed by each first model training vector in the current subsequence and the corresponding first acquisition time field of the first acquisition record, and the corresponding second model training tensor is formed by all the first user behavior time sequence vectors corresponding to the current subsequence; forming corresponding second model training data by each second model training tensor and the corresponding second model label vector; when the traversing is finished, the second model training data sets corresponding to all the obtained second model training data are formed;
traversing all the second acquisition records of the stock user database; and traversing, wherein the second acquisition record of the current traversal is used as a corresponding current record; the first stock user corresponding to the current record is used as a corresponding current user; and taking the second acquisition time field of the current record as the corresponding current time; recording the first type grade update record of which the first update time field is earlier than the current time in the first user information table corresponding to the current user in the first user information collection library as a corresponding second early update record, and forming the first latest type field and the first latest grade field of the first early update record of which the first update time field is closest to the current time in all the second early update records into a corresponding original user type grade; identifying whether the first type level change suggestion field of the current record is the hold state, if so, setting a user tag classification vector for model training according to the original user type level to be a corresponding third model tag vector based on the vector structure of the user tag classification vector, otherwise, setting a user tag classification vector for model training according to the first type level change suggestion field of the current record to be a corresponding third model tag vector based on the vector structure of the user tag classification vector; performing corresponding level normalization processing on the corresponding first feedback data in the current record based on the feedback data level normalization standards corresponding to the subtask types to obtain corresponding first subtask feedback level values; the corresponding first subtask type grade is formed by each subtask type of the current record and the corresponding first subtask feedback grade value; setting a user task feedback vector for model training as a corresponding third model training vector according to all the first subtask type grades recorded currently based on the vector structure of the user task feedback vector; the obtained third model training vector and the corresponding third model label vector form corresponding third model training data; and at the end of the traversal, forming a corresponding third model training data set by all the obtained third model training data.
6. An apparatus for performing the AI-model-based user representation processing method of any of claims 1-5, the apparatus comprising: an AI model preparation module and an AI model representation module;
The AI model preparation module is used for configuring data structures of a user behavior vector, a user behavior time sequence tensor, a user task feedback vector and a user tag classification vector; constructing three types of artificial intelligent models for classifying the user labels, namely a first classification model, a second classification model and a third classification model; training the first, second and third classification models based on a preset stock user database; the first classification model is used for carrying out user tag type prediction processing according to the input user behavior vector to obtain a corresponding user tag classification vector; the second classification model is used for carrying out user tag type prediction processing according to the input user behavior time sequence tensor to obtain the corresponding user tag classification vector; the third classification model is used for carrying out user tag type prediction processing according to the input user task feedback vector to obtain a corresponding user tag classification vector;
The AI model portrayal module is used for processing an initial user image by using the first classification model after training the first, second and third classification models; processing the user tracking image by using the second classification model; and processing user portrayal adjustment based on task feedback using the third classification model.
7. An electronic device, comprising: memory, processor, and transceiver;
The processor being configured to couple to the memory, read and execute instructions in the memory to implement the method of any one of claims 1-5;
The transceiver is coupled to the processor and is controlled by the processor to transmit and receive messages.
8. A computer readable storage medium storing computer instructions which, when executed by a computer, cause the computer to perform the method of any one of claims 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410410709.2A CN118013399B (en) | 2024-04-08 | 2024-04-08 | AI model-based user portrait processing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410410709.2A CN118013399B (en) | 2024-04-08 | 2024-04-08 | AI model-based user portrait processing method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN118013399A CN118013399A (en) | 2024-05-10 |
CN118013399B true CN118013399B (en) | 2024-06-14 |
Family
ID=90954316
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410410709.2A Active CN118013399B (en) | 2024-04-08 | 2024-04-08 | AI model-based user portrait processing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN118013399B (en) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117743848A (en) * | 2023-12-06 | 2024-03-22 | 暗物质(北京)智能科技有限公司 | User portrait generation method and device, electronic equipment and storage medium |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112507116B (en) * | 2020-12-16 | 2023-10-10 | 平安科技(深圳)有限公司 | Customer portrait method based on customer response corpus and related equipment thereof |
CN113220773B (en) * | 2021-05-28 | 2023-09-01 | 华中师范大学 | Map visualization method and system for student evaluation data image |
CN114119058B (en) * | 2021-08-10 | 2023-09-26 | 国家电网有限公司 | User portrait model construction method, device and storage medium |
CN114092264A (en) * | 2021-10-13 | 2022-02-25 | 北京博瑞彤芸科技股份有限公司 | Intelligent passenger obtaining method, system and electronic equipment |
CN115423035A (en) * | 2022-09-26 | 2022-12-02 | 重庆长安新能源汽车科技有限公司 | User portrait generation method based on feature variable scoring, equipment, automobile and storage medium |
CN115936802A (en) * | 2022-11-25 | 2023-04-07 | 上海欣兆阳信息科技有限公司 | Personalized marketing method, device, equipment and storage medium based on user portrait and sequence modeling |
CN116468109B (en) * | 2023-06-19 | 2023-08-29 | 深圳索信达数据技术有限公司 | Training method, using method and related device of prediction model |
-
2024
- 2024-04-08 CN CN202410410709.2A patent/CN118013399B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117743848A (en) * | 2023-12-06 | 2024-03-22 | 暗物质(北京)智能科技有限公司 | User portrait generation method and device, electronic equipment and storage medium |
Non-Patent Citations (1)
Title |
---|
一种用于构建用户画像的二级融合算法框架;李恒超 等;计算机科学;20180115;第45卷(第01期);157-161 * |
Also Published As
Publication number | Publication date |
---|---|
CN118013399A (en) | 2024-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10360517B2 (en) | Distributed hyperparameter tuning system for machine learning | |
US10354201B1 (en) | Scalable clustering for mixed machine learning data | |
US10832174B1 (en) | Distributed hyperparameter tuning system for active machine learning | |
US11151480B1 (en) | Hyperparameter tuning system results viewer | |
US10963802B1 (en) | Distributed decision variable tuning system for machine learning | |
US11816687B2 (en) | Personalized approach to modeling users of a system and/or service | |
CN108108743A (en) | Abnormal user recognition methods and the device for identifying abnormal user | |
CN112446399A (en) | Label determination method, device and system | |
CN112115996B (en) | Image data processing method, device, equipment and storage medium | |
CN113407801A (en) | Cloud computing index recommendation method and system | |
CN110263136B (en) | Method and device for pushing object to user based on reinforcement learning model | |
US20240202057A1 (en) | Methods and systems for determining stopping point | |
CN114219562A (en) | Model training method, enterprise credit evaluation method and device, equipment and medium | |
CN111209105A (en) | Capacity expansion processing method, capacity expansion processing device, capacity expansion processing equipment and readable storage medium | |
CN118013399B (en) | AI model-based user portrait processing method and device | |
CN116957260A (en) | Intelligent allocation method for carrier vehicles based on vehicle demand prediction | |
CN113837807B (en) | Heat prediction method, heat prediction device, electronic equipment and readable storage medium | |
CN110062112A (en) | Data processing method, device, equipment and computer readable storage medium | |
CN113761365B (en) | Data processing system for determining target information | |
CN115905293A (en) | Switching method and device of job execution engine | |
CN111723872B (en) | Pedestrian attribute identification method and device, storage medium and electronic device | |
CN116703046A (en) | Control method and system for real-time dispatching sequence, electronic equipment and storage medium | |
CN113095064A (en) | Code field identification method and device, electronic equipment and storage medium | |
Long et al. | Keras Advanced API | |
CN115062676B (en) | Data processing method, device and computer readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |