CN114896475A - Medium information processing method, medium information processing device, electronic equipment and storage medium - Google Patents
Medium information processing method, medium information processing device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN114896475A CN114896475A CN202210640512.9A CN202210640512A CN114896475A CN 114896475 A CN114896475 A CN 114896475A CN 202210640512 A CN202210640512 A CN 202210640512A CN 114896475 A CN114896475 A CN 114896475A
- Authority
- CN
- China
- Prior art keywords
- media information
- model
- target
- account
- sample
- 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.)
- Pending
Links
- 230000010365 information processing Effects 0.000 title claims abstract description 26
- 238000003672 processing method Methods 0.000 title claims abstract description 15
- 238000012216 screening Methods 0.000 claims abstract description 79
- 238000012163 sequencing technique Methods 0.000 claims abstract description 67
- 238000000034 method Methods 0.000 claims abstract description 48
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 27
- 238000004590 computer program Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 59
- 238000012545 processing Methods 0.000 claims description 36
- 230000006870 function Effects 0.000 claims description 35
- 230000004927 fusion Effects 0.000 claims description 13
- 238000013439 planning Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 description 14
- 238000004364 calculation method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000010276 construction Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000008707 rearrangement Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012384 transportation and delivery Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/9035—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/06—Arrangements for sorting, selecting, merging, or comparing data on individual record carriers
- G06F7/08—Sorting, i.e. grouping record carriers in numerical or other ordered sequence according to the classification of at least some of the information they carry
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present disclosure relates to a media information processing method, apparatus, electronic device, storage medium, and computer program, the method comprising: account characteristics of the target account are obtained. Obtaining a probability prediction result according to the account characteristics and a pre-trained probability prediction model; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sequencing model becomes the target medium information; and determining the screening quantity of the medium information corresponding to each sequencing model according to the probability prediction result and a preset medium quantity calculation algorithm. According to the screening quantity of the media information corresponding to each sequencing model, the candidate media information is obtained from the media information sequence sequenced by each sequencing model; the candidate media information is used for determining target media information corresponding to the target account. By adopting the method, the recommendation accuracy of the target medium information is improved.
Description
Technical Field
The present disclosure relates to the field of data processing technologies, and in particular, to a method and an apparatus for processing media information, an electronic device, a storage medium, and a computer program.
Background
Before the media information is released and displayed, the target media information for releasing and displaying needs to be determined through processing operations such as orientation, recall, sequencing and the like. In this process, the sorting stage is a process that needs to sort and screen the recalled media information data stream according to a preset model. The sorting can be divided into two stages of coarse sorting and fine sorting.
In the current media information sorting method, a coarse sorting stage includes two preset sorting models, the two sorting models can sort recalled media information data streams according to different dimensions respectively to obtain media information sequences corresponding to the two different dimensions, and further, a fixed amount of candidate media information is determined in each media information sequence, so that the fine sorting can determine unique target media information from the candidate media information obtained by the coarse sorting, and the target media information is recommended to an account.
However, in the current media information sorting method, two sorting models in the coarse sorting are media information sequences obtained by sorting according to different dimensions. Different dimensions have different importance degrees for different accounts, so that if each account acquires candidate media information in two media information sequences according to a uniform quantity threshold value, the media information delivery is not targeted, and the accuracy of the determined target media information is low.
Disclosure of Invention
The present disclosure provides a media information processing method, apparatus, electronic device, storage medium, and computer program, to at least solve the problem of low accuracy of target media information determined in related art. The technical scheme of the disclosure is as follows:
according to a first aspect of embodiments of the present disclosure, there is provided a medium information processing method including:
acquiring account characteristics of a target account;
obtaining a probability prediction result according to the account characteristics and a pre-trained probability prediction model; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sequencing model becomes the target medium information;
determining the screening quantity of the medium information corresponding to each sequencing model according to the probability prediction result and a preset medium quantity calculation algorithm; each sequencing model comprises the target sequencing model;
according to the screening quantity of the media information corresponding to each sequencing model, the candidate media information is obtained from the media information sequence sequenced by each sequencing model; the candidate media information is used for determining target media information corresponding to the target account.
In one embodiment, the obtaining account characteristics of the target account includes:
acquiring account portrait characteristics and context characteristics of a target account;
and performing feature splicing on the account portrait features and the context features to obtain fusion features, and performing encoding processing on the fusion features to obtain encoded account features.
In an embodiment, the determining, according to the probability prediction result and a preset intermediary amount calculation algorithm, the intermediary information screening amount corresponding to each ranking model includes:
acquiring the total quantity of preset candidate medium information;
taking the total amount of the candidate medium information as a constraint condition, and constructing a target function according to the probability prediction result and a preset target planning algorithm; the target function comprises medium information screening parameters corresponding to the sequencing models;
determining parameter values of medium information screening parameters corresponding to the sequencing models contained in the objective function according to the objective function and the constraint conditions; and the parameter values are the screening quantity of the medium information corresponding to each sequencing model.
In one embodiment, the method further comprises:
acquiring sample account characteristics and training samples of a target account, wherein the training samples comprise sample target media information and model marks of a sequencing model corresponding to the sample target media information;
and performing model training on the initial probability prediction model according to the sample account characteristics and the training samples to obtain a trained probability prediction model.
In one embodiment, the obtaining training samples includes:
acquiring a media information stream; the media information stream comprises a plurality of media information sets;
inputting each media information set into a sorting model to obtain candidate media information corresponding to each sorting model, and adding model marks to the candidate media information, wherein the model marks represent the sorting models corresponding to the candidate media information;
determining sample target media information in each candidate media information according to a preset value estimation model and the candidate media information, and determining a training sample according to the sample target media information and the model mark of the sample target media information.
In one embodiment, the training samples include positive samples and negative samples; determining a training sample according to the sample target media information and the model label of the sample target media information, comprising:
if the sample target media information carries a first model mark, determining the sample target media information as positive sample data;
and if the sample target media information carries a second model mark, determining the sample target media information as negative sample data.
According to a second aspect of the embodiments of the present disclosure, there is provided a medium information processing apparatus including:
an acquisition unit configured to perform acquisition of an account characteristic of a target account;
the processing unit is configured to execute a probability prediction model trained in advance according to the account characteristics to obtain a probability prediction result; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sequencing model becomes the target medium information;
the determining unit is configured to execute a calculation algorithm according to the probability prediction result and a preset medium quantity, and determine a medium information screening quantity corresponding to each sequencing model; each sequencing model comprises the target sequencing model;
the screening unit is configured to execute screening quantity of the media information corresponding to each sorting model, and candidate media information of the screening quantity of the media information is obtained in the media information sequence sorted by each sorting model; the candidate media information is used for determining target media information corresponding to the target account.
In one embodiment, the obtaining unit includes:
an acquisition subunit configured to perform acquiring account portrait and contextual features of a target account;
and the coding processing unit is configured to perform feature splicing on the account image features and the context features to obtain fusion features, and perform coding processing on the fusion features to obtain the account features after the coding processing.
In one embodiment, the determining unit includes:
an acquisition subunit configured to perform acquisition of a preset total number of candidate medium information;
the construction subunit is configured to execute the operation of taking the total quantity of the candidate medium information as a constraint condition and constructing an objective function according to the probability prediction result and a preset objective planning algorithm; the target function comprises medium information screening parameters corresponding to the sequencing models;
the calculation unit is configured to determine parameter values of medium information screening parameters corresponding to the sorting models contained in the objective function through the objective function and the constraint conditions; and the parameter values are the screening quantity of the medium information corresponding to each sequencing model.
In one embodiment, the medium information processing apparatus further includes:
the system comprises a sample acquisition unit, a data processing unit and a data processing unit, wherein the sample acquisition unit is configured to execute acquisition of sample account characteristics and training samples of a target account, and the training samples comprise sample target medium information and model marks of a sequencing model corresponding to the sample target medium information;
and the training unit is configured to perform model training on the initial probability prediction model according to the sample account characteristics and the training samples to obtain a trained probability prediction model.
In one embodiment, the sample acquiring unit includes:
a data acquisition subunit configured to perform acquiring the media information stream; the media information stream comprises a plurality of media information sets;
the sample processing subunit is configured to input each media information set into a ranking model, obtain candidate media information corresponding to each ranking model, and add a model tag to the candidate media information, wherein the model tag represents the ranking model corresponding to the candidate media information;
the training sample determining subunit is configured to determine sample target media information in each candidate media information according to a preset value estimation model and the candidate media information, and determine a training sample according to the sample target media information and the model mark of the sample target media information.
In one embodiment, the training sample determination subunit includes:
a first determining subunit configured to determine the sample target media information as positive sample data if the sample target media information carries a first model flag;
a second determining subunit configured to determine the sample target media information as negative sample data if the sample target media information carries a second model flag.
In a third aspect, a server is provided, which includes:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the target object presentation method of the first aspect or the second aspect.
According to a fourth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium, wherein instructions of the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the target object presentation method of any one of the first aspect or the second aspect.
According to a fifth aspect of the embodiments of the present disclosure, there is provided a computer program product, where the instructions, when executed by a processor of an electronic device, enable the electronic device to perform the target object displaying method of any one of the first aspect or the second aspect.
The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:
the computer equipment acquires account characteristics of the target account; obtaining a probability prediction result according to the account characteristics and a pre-trained probability prediction model; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sequencing model becomes the target medium information; determining the screening quantity of the medium information corresponding to each sequencing model according to the probability prediction result and a preset medium quantity calculation algorithm; according to the screening quantity of the media information corresponding to each sequencing model, the candidate media information is obtained from the media information sequence sequenced by each sequencing model; the candidate media information is used for determining target media information corresponding to the target account. By adopting the method, model operation is carried out through the account characteristics of the target account and the probability prediction model to obtain the probability prediction result for representing the characteristic preference of the target object, and then the media information screening quantity of the sequencing models is determined based on the probability prediction result, so that the media information screening quantity corresponding to each sequencing model in the process of determining the target media information at each time is dynamically changed according to the account characteristics of the target account, and the accuracy of determining the target media information is improved and lower. .
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.
Fig. 1 is a flowchart illustrating a media information processing method according to an exemplary embodiment.
FIG. 2 is a flowchart illustrating one step of obtaining account characteristics for a target account in accordance with an exemplary embodiment.
FIG. 3 is a flow diagram illustrating a media information screening quantity determination for various ranking models in accordance with an exemplary embodiment.
FIG. 4 is a flow diagram illustrating a probabilistic predictive model training method in accordance with an exemplary embodiment.
FIG. 5 is a flow chart illustrating a method of determining training samples in accordance with an exemplary embodiment.
Fig. 6 is a flow chart illustrating a method of dividing positive and negative samples according to an example embodiment.
Fig. 7 is a block diagram illustrating a media information processing apparatus according to an example embodiment.
FIG. 8 is a block diagram illustrating a server in accordance with an exemplary embodiment.
Detailed Description
In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.
It should be further noted that the account information (including but not limited to account device information, account personal information, etc.) and data (including but not limited to data for presentation, analyzed data, etc.) referred to in the present disclosure are information and data that are authorized by the account or sufficiently authorized by the parties.
Fig. 1 is a flowchart illustrating a media information processing method according to an exemplary embodiment, where as shown in fig. 1, the media information processing method may be applied to a server, and specifically includes the following steps.
In step S110, account characteristics of the target account are acquired.
In implementation, an account can browse page information through a client, in the browsing process, the account triggers corresponding page browsing operation, then the client generates a corresponding browsing request and transmits the browsing request to a server (referred to as a server for short) for media information recommendation, and then the server responds to the browsing request and acquires traffic data carried in the browsing request. The server may extract account characteristics of the target account (i.e., the target account) from the traffic information. The traffic data includes all communication data between the client and the server, for example, browsing request data; the account features extracted from the traffic data may reflect feature preferences of a target account (i.e., a target object), and the account features may include, but are not limited to, an account portrait feature (user feature), an account context feature (context feature), and the like, which is not limited in the embodiment of the present disclosure.
In step S120, a probability prediction result is obtained according to the account feature and the pre-trained probability prediction model.
The probability prediction result reflects the probability that the candidate medium information corresponding to the target sorting model becomes the target medium information. The determination process of the target medium information needs four processes of orientation, recall, rough arrangement and fine arrangement. In the rough ranking stage, the server side presets two ranking models, which can be called a first ranking model and a second ranking model, and the two ranking models rank each media information in the media information flow according to different dimensions (also called characteristic dimensions) to obtain two media information sequences. For example, the dimension of the ordering of the first ordering model is the value of the media information, so the first ordering model may also be referred to as a value ordering model, and the dimension of the ordering of the second ordering model is the conversion rate of the media information, so the second ordering model may also be referred to as a conversion rate ordering model, so the server can obtain candidate media information from two media information sequences respectively, and then, in a fine ordering stage, determine target media information in the candidate media information, where the target media information is used for recommending to a target account. That is, the determined target intermediary information may be from the candidate intermediary information corresponding to the first ranking model or from the candidate intermediary information corresponding to the second ranking model.
In implementation, the server stores a pre-trained probability prediction model. The output of the probabilistic predictive model (i.e., the probabilistic predictive result) is used to characterize the probability that the target intermediary information is from the candidate intermediary information corresponding to the target ranking model. Optionally, the target ranking model may be a first ranking model or a second ranking model, and the embodiment of the present disclosure is not limited.
Specifically, the probabilistic predictive model is trained from sample data of the current account, and the feature preference of the current account is learned. For example, the probabilistic predictive model takes the first ranking model of the rearrangement stage as the target ranking model, and the probabilistic predictive result corresponding to the first ranking model can be output. Furthermore, in the model application process, the server inputs the account characteristics of the target account into the pre-trained probabilistic prediction model, and through the operation processing of the probabilistic prediction model, the probabilistic prediction result corresponding to the first ranking model can be determined, and the probabilistic prediction result can represent the probability that the target medium information comes from the candidate medium information corresponding to the first ranking model. Optionally, the media information in the embodiment of the present disclosure may be information used for propagation, such as advertisements, and the embodiment of the present disclosure is not limited.
In step S130, the media information screening amount corresponding to each ranking model is determined according to the probability prediction result and a preset media amount calculation algorithm.
In implementation, after a probability prediction result corresponding to the target ranking model is obtained, the server may construct a medium quantity calculation model (or referred to as a medium quantity objective function) according to the probability prediction result, the total quantity of candidate medium information screened by the two ranking models, and a preset medium quantity calculation algorithm. The medium quantity calculation model comprises medium information screening parameters corresponding to each sequencing model, and the parameter values of the medium information screening parameters are the medium information screening quantity. Furthermore, the server can solve the medium quantity calculation model to obtain the medium information screening quantity corresponding to each sequencing model.
In step S140, candidate pieces of media information of the media information screening amount are obtained from the media information sequence sorted by each sorting model according to the media information screening amount corresponding to each sorting model.
The candidate media information is used for determining target media information corresponding to the target account.
In implementation, for each ranking model, the server obtains candidate media information of the media information screening quantity from the ranked media information sequence output by the ranking model according to the media information screening quantity corresponding to the ranking model, so that the server can further determine final target media information from the candidate media information and push the target media information to a target account.
In the media information processing method, the server side obtains the account characteristics of the target account. And then, inputting the account characteristics into a pre-trained probability prediction model to obtain a probability prediction result. The probability prediction result reflects the probability that the medium information screened by the target sorting model becomes the target medium information. And then, the server side determines the media information screening quantity corresponding to each sequencing model according to the probability prediction result and a preset media quantity calculation algorithm. And according to the screening quantity of the medium information, obtaining candidate medium information in the medium information sequence sequenced by each sequencing model. The candidate media information is used for determining target media information of the target account. By adopting the method, model operation is carried out through the account characteristics of the target account and the probability prediction model to obtain the probability prediction result for representing the characteristic preference of the target object, and then the media information screening quantity of the sequencing models is determined based on the probability prediction result, so that the media information screening quantity corresponding to each sequencing model in the process of determining the target media information at each time is dynamically changed according to the account characteristics of the target account, and the accuracy of determining the target media information is improved and lower.
In an exemplary embodiment, as shown in fig. 2, in step S110, the following steps may be specifically implemented:
in step S111, account representation features and context features of the target account are obtained.
In implementation, the server may obtain all traffic data transmitted between the server and the client of the target account, and then, the server may extract, based on a preset feature extraction algorithm, an image feature (u8er feature) and a context feature (context feature) of the target account from the traffic data. The account profile characteristics may include, but are not limited to, age, gender, geographic location of the target account, clicked historical media information, and the like. The contextual characteristics are characteristics carried by the client of the target account, and the contextual characteristics may include, but are not limited to, the time the target account sends the request, the mobile terminal brand of the target account, the operating system, the mobile terminal model, the current network state (e.g., whether 4G, 5G, wifi, etc.).
In step S112, the account image feature and the context feature are feature-spliced to obtain a fusion feature, and the fusion feature is encoded to obtain an encoded account feature.
In implementation, the server side inputs the account portrait characteristics and the context characteristics into a probability prediction model, and the probability prediction model processes the account portrait characteristics and the context characteristics, that is, feature splicing is performed on the account portrait characteristics and the context characteristics to obtain spliced fusion characteristics. And then, transmitting the spliced fusion features to an embedding layer (also called an embedding layer) in a probability prediction model, and coding the fusion features by the embedding layer to obtain the coded account features.
In this embodiment, the account characteristics used for model processing are obtained by performing characteristic extraction and characteristic splicing on the flow data of the target account, and then the probability prediction model may perform operation processing on the account characteristics and output a probability prediction result.
In an exemplary embodiment, as shown in fig. 3, in step S130, the following steps may be specifically implemented:
in step S131, a preset total number of candidate medium information is obtained.
In implementation, the data processing amount of the medium information recommendation model corresponding to the fine ranking stage is limited. Therefore, the total quantity of the candidate media information corresponding to the candidate media information in the rough ranking stage is pre-stored in the server, and the total quantity of the candidate media information can be determined according to the data processing capacity of the media information recommendation model corresponding to the fine ranking stage. Furthermore, in the process of processing the media information, the server may obtain a total number (also denoted as K) of the preset candidate media information as a number constraint condition.
In step S132, the total amount of candidate medium information is used as a constraint condition, and an objective function is constructed according to the probability prediction result and a preset objective planning algorithm.
The target function comprises medium information screening parameters corresponding to the sequencing models.
In implementation, since the number of candidate pieces of media information required by the fine ranking stage is fixed, in one implementation, the number may be used as the total number of candidate pieces of media information, and then, the total number of candidate pieces of media information may be used as a constraint condition for the number of candidate pieces of media information corresponding to each ranking model in the coarse ranking stage, that is, qotaa + qotab ═ K. The method comprises the following steps that (1) quotaA represents a medium information screening parameter corresponding to a first sequencing model in a coarse sequencing stage, and a parameter value of the medium information screening parameter represents the medium information screening quantity corresponding to the first sequencing model; and the quotaB represents medium information screening parameters corresponding to the second sequencing model, and parameter values of the medium information screening parameters represent the medium information screening quantity corresponding to the second sequencing model.
After the constraint condition is determined, the server side can construct an objective function according to the probability prediction result and a preset objective planning algorithm. The target function comprises medium information screening parameters corresponding to the first sequencing model and medium information screening parameters corresponding to the second sequencing model.
Specifically, the probability prediction result reflects the probability that the candidate media information corresponding to the target ranking model becomes the target media information, which can be denoted as P, and if the target ranking model is the first ranking model, P is the probability that the candidate media information corresponding to the first ranking model becomes the target media information. Therefore, the media information filtering amount corresponding to the first ranking model may be denoted as quotaA ═ K × P, where K denotes the total amount of candidate media information, and the media information filtering amount corresponding to the second ranking model may be denoted as K-K × P. And then, the server side constructs an objective function according to the medium information screening parameters corresponding to the first sequencing model, the medium information screening parameters corresponding to the second sequencing model and a preset objective planning algorithm. The target planning algorithm may be various, the target planning algorithm is described as the lagrangian algorithm in the embodiment of the present disclosure, and other algorithms are similar to the lagrangian algorithm and are not described again.
The objective function comprises medium information screening parameters corresponding to the first sequencing model, and an index coefficient alpha corresponding to the medium information screening parameters, medium information screening parameters corresponding to the second sequencing model, and an index coefficient beta corresponding to the medium information screening parameters. Wherein, the exponent coefficient α and the exponent coefficient β are known quantities, for example, the server may perform: max \ sum _ i ^ N { quotaA } { \ alpha } + { quotaB } { \\ beta } \\ newline st \ newline program to realize the construction of the target function.
In step S133, parameter values of media information filtering parameters corresponding to each ranking model included in the objective function are determined by the objective function and the constraint conditions.
The parameter values of the medium information screening parameters represent the screening quantity of the medium information corresponding to each sequencing model.
In implementation, the server performs solution calculation of the objective function under the constraint of the constraint condition. The solution of the objective function is parameter values (i.e., screening quantity of the pieces of media information) of the media information screening parameters corresponding to the first ranking model and parameter values (i.e., screening quantity of the pieces of media information) of the media information screening parameters corresponding to the second ranking model.
In this embodiment, a probability prediction result of a target ranking model representing account feature preference is obtained through account features of a target object, and then a target function is constructed based on the probability prediction result, a preset target planning algorithm and media information screening quantity corresponding to each ranking model. The screening quantity of the candidate media information of the target account is determined by solving the target function, so that the accuracy of determining the candidate media information is improved, and further, the accuracy of determining the target media information is improved.
In an exemplary embodiment, as shown in fig. 4, an embodiment of the present disclosure further discloses a method for training a probabilistic predictive model, where the method further includes:
in step S410, account features and training samples are obtained.
The training sample comprises sample target medium information and a model mark corresponding to the sample target medium information.
In implementation, the account characteristics and the training samples corresponding to the target account are stored in the server in advance. The training sample may be composed of sample target medium information, and the sample target medium information may be determined in the process of processing the historical medium information of the target account. In order to indicate the source of the finally determined sample target media information, the server may add corresponding model marks to candidate media information determined in media information sequences of different ranking models in the rough ranking stage, so that the sample target media information determined in each candidate media information also carries corresponding model marks in the fine ranking stage.
Specifically, for the account characteristics of the target account, the server may obtain all traffic data between the client and the server of the target account. Then, in the traffic data, the server may randomly extract a part of the traffic data, and extract account features from the part of the traffic data according to a preset feature extraction algorithm. In addition, for the processing procedure of the server side obtaining the training sample of the target account, the following detailed description of the embodiment of the disclosure will be provided, and will not be repeated herein.
In step S420, model training is performed on the initial probability prediction model according to the account features and the training samples, so as to obtain a trained probability prediction model.
In implementation, a probability prediction model (which may also be referred to as an initial probability prediction model) is preset in a server, the server inputs account features and training samples into the probability prediction model to obtain a probability prediction result, then the server calculates a loss value between the probability prediction result and a verification set result (a target ranking model corresponding to each target media information included in a training sample) of supervised learning, and when a loss value corresponding to an output result (i.e., the probability prediction result) of the probability prediction model meets a preset loss condition, the server stops training the probability prediction model to obtain a trained probability prediction model.
The probabilistic prediction model may be, but not limited to, a DNN (Deep Neural Networks) model, and the DNN model is taken as an example in the embodiment of the present disclosure for explanation. Wherein the DNN model comprises an input layer, a hidden layer and an output layer. The hidden layer may also comprise multiple layers. The disclosed embodiments are not limiting. Then, the server performs model training on the initial probability prediction model according to the obtained account features and training samples, and obtains probability prediction results corresponding to the target ranking model through operation processing on the account features and the training samples by a plurality of hidden layers (for example, hidden layers with three data processing dimensions [512,128, 2 ]) in the initial probability prediction model and activation of relu activation functions in an output layer of the initial probability prediction model. After the probability prediction result is obtained, the server verifies an output result (namely, the probability prediction result) according to the ranking model represented by the model mark carried by the sample target medium information, calculates a loss function between the output result of the initial probability prediction model and a true value (namely, the category of the ranking model represented by the model mark), and then, based on the size relationship between the loss value of the loss function (for example, the loss function may be a cross entropy loss function) and a preset loss threshold, the server may determine whether to perform parameter adjustment on the initial probability prediction model. If the loss value is greater than the preset loss threshold value (namely, the characteristic loss value does not meet the preset loss condition), adjusting parameters of the initial probability prediction model, and then, continuing to execute the process of calculating the initial probability prediction model to obtain a probability prediction result until the loss of the initial probability prediction model is less than or equal to the preset loss threshold value (namely, the characteristic loss value meets the preset loss condition), determining that the training of the initial probability prediction model is finished.
In the embodiment, the initial probability prediction model is trained on the supervised deep neural network model through the pre-constructed training sample and the extracted sample account characteristics of the target account, so that the accuracy and stability of the output result of the probability prediction model are improved, and a stable and accurate probability prediction result can be obtained based on the trained probability prediction model in the model application process.
In an exemplary embodiment, as shown in fig. 5, in step S410, the specific implementation procedure of the step of obtaining training samples is as follows:
in step S411, a media information stream is acquired.
The media information stream comprises a plurality of media information sets.
In implementation, the server obtains a historical media information stream from the media information database, where the historical media information stream may include multiple media information sets, and the server may determine a target media information in each media information set.
In step S412, each media information set is input into the ranking model, candidate media information corresponding to each ranking model is obtained, and a model tag is added to the candidate media information.
The model mark represents a ranking model corresponding to the candidate medium information.
In implementation, after directional and recall processing is performed on each media information set, the media information set enters a media information rough arrangement stage, the server side inputs the directional and recall processed media information sets into two sequencing models in the rough arrangement stage, and the two sequencing models sequence the media information in the media information sets according to different dimensions respectively to obtain different media information sequences. Then, screening quantity thresholds according to the medium information corresponding to different sorting models, and obtaining candidate medium information of a quantity corresponding to each sorting model in the two medium information sequences. In addition, the server can also add model marks to each candidate media information according to the ranking model source of the candidate media information. For example, the first ranking model corresponds to the model flag a, the second ranking model corresponds to the model flag B, and further, in the medium information sequence ranked by the first ranking model, 100 medium information items obtained by screening are all added with the model flag a, and in the medium information sequence ranked by the second ranking model, 200 medium information items obtained by screening are all added with the model flag B.
In step S413, sample target media information in each candidate media information is determined according to a preset value estimation model and the candidate media information, and a training sample is determined according to the sample target media information and a model label of the sample target media information.
In implementation, the server obtains candidate media information determined in a rough arrangement mode, and can input the candidate media information into the value estimation model to obtain sample target media information. The server may further obtain the model label carried by the sample target media information, and then determine the sample as the training sample according to the sample target media information and the model label carried by the sample target media information.
In this embodiment, the server obtains candidate media information according to the ranking model in the rearrangement stage, then adds a model tag to the candidate media information, and further determines that sample target media information in the candidate media information carries the model tag, where the model tag is used to represent the candidate media information corresponding to which ranking model the sample target media information comes from, so that a training sample constructed by the sample target media information realizes tagging of the sample target media information, and can be used to complete supervised model training.
In an exemplary embodiment, as shown in FIG. 6, the training samples include positive samples and negative samples. The specific processing procedure of step S413 includes:
in step S610, if the sample target media information carries the first model flag, the sample target media information is determined as positive sample data.
In implementation, if the sample target media information carries the first model mark, the server classifies the sample target media information into positive sample data, and then a positive sample is formed based on the sample target media information classified into the positive sample data determined in the plurality of media information sets.
In step S620, if the sample target media information carries the second model flag, the sample target media information is determined as negative sample data.
In implementation, if the sample target media information carries the second model mark, the server classifies the sample target media information as negative sample data, and then a negative sample is formed based on the sample target media information classified as negative sample data determined in a plurality of media information sets.
In this embodiment, the training sample is divided by the model labels carried by the sample target medium information in the training sample, so as to obtain a positive sample of the sample target medium information containing the first model label and a negative sample of the sample target medium information containing the second model label, thereby implementing the construction of the training sample of the supervised neural network model.
It should be understood that although the various steps in the flowcharts of fig. 1-6 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 1-6 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least some of the other steps.
It is understood that the same/similar parts between the embodiments of the method described above in this specification can be referred to each other, and each embodiment focuses on the differences from the other embodiments, and it is sufficient that the relevant points are referred to the descriptions of the other method embodiments.
Fig. 7 is a block diagram illustrating a media information processing apparatus according to an example embodiment. Referring to the figure, the apparatus 700 comprises an acquisition unit 702, a processing unit 704, a determination unit 706 and a screening unit 708.
An acquiring unit 702 configured to perform acquiring account characteristics of a target account;
a processing unit 704 configured to execute a probability prediction model trained in advance according to the account features to obtain a probability prediction result; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sequencing model becomes the target medium information;
a determining unit 706 configured to execute a calculation algorithm according to the probability prediction result and a preset amount of media, and determine a screening amount of media information corresponding to each sequencing model;
a screening unit 708 configured to execute screening quantity of the media information according to the media information corresponding to each sorting model, and obtain candidate media information of the screening quantity of the media information in the media information sequence sorted by each sorting model; the candidate media information is used for determining target media information corresponding to the target account.
In this embodiment, with the intermediary information processing apparatus, model operation is performed through the account features of the target account and the probability prediction model to obtain a probability prediction result for representing the preference of the target object features, and then the intermediary information screening number of the ranking model is determined based on the probability prediction result, so that the intermediary information screening number corresponding to each ranking model in each determination process of the target intermediary information is dynamically changed according to the account features of the target account, thereby improving the accuracy of determining the target intermediary information.
In an exemplary embodiment, the obtaining unit 702 includes:
an acquisition subunit configured to perform acquiring account portrait and contextual features of a target account;
and the coding processing unit is configured to perform feature splicing on the account image features and the context features to obtain fusion features, and perform coding processing on the fusion features to obtain the account features after the coding processing.
In an exemplary embodiment, the determining unit 706 includes:
an acquisition subunit configured to perform acquisition of a preset total number of candidate medium information;
the construction subunit is configured to execute the operation of taking the total quantity of the candidate medium information as a constraint condition and constructing an objective function according to the probability prediction result and a preset objective planning algorithm; the target function comprises medium information screening parameters corresponding to the sequencing models;
the calculation unit is configured to determine parameter values of medium information screening parameters corresponding to the sorting models contained in the objective function through the objective function and the constraint conditions; and the parameter values are the screening quantity of the medium information corresponding to each sequencing model.
In an exemplary embodiment, the intermediary information processing apparatus 700 further comprises:
the system comprises a sample acquisition unit, a data processing unit and a data processing unit, wherein the sample acquisition unit is configured to execute acquisition of sample account characteristics and training samples of a target account, and the training samples comprise sample target medium information and model marks of a sequencing model corresponding to the sample target medium information;
and the training unit is configured to perform model training on the initial probability prediction model according to the sample account characteristics and the training samples to obtain a trained probability prediction model.
In an exemplary embodiment, the sample acquiring unit includes:
a data acquisition subunit configured to perform acquiring the media information stream; the media information stream comprises a plurality of media information sets;
the sample processing subunit is configured to input each media information set into a ranking model, obtain candidate media information corresponding to each ranking model, and add a model tag to the candidate media information, wherein the model tag represents the ranking model corresponding to the candidate media information;
the training sample determining subunit is configured to determine sample target media information in each candidate media information according to a preset value estimation model and the candidate media information, and determine a training sample according to the sample target media information and the model mark of the sample target media information.
In an exemplary embodiment, the training sample determination subunit includes:
a first determining subunit configured to determine the sample target media information as positive sample data if the sample target media information carries a first model flag;
a second determining subunit configured to determine the sample target media information as negative sample data if the sample target media information carries a second model flag.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
In an exemplary embodiment, a computer program product is also provided that includes instructions executable by the processor 820 of the electronic device 800 to perform the above-described method.
Fig. 8 is a block diagram illustrating an electronic device 800 for mediating information processing in accordance with an exemplary embodiment. For example, the electronic device 800 may be a server. Referring to fig. 8, electronic device 800 includes a processing component 820 that further includes one or more processors and memory resources, represented by memory 822, for storing instructions, such as applications, that are executable by processing component 820. The application programs stored in memory 822 may include one or more modules that each correspond to a set of instructions. Further, the processing component 820 is configured to execute instructions to perform the above-described methods.
The electronic device 800 may further include: a power component 824 is configured to perform power management for the electronic device 800, a wired or wireless network interface 826 configured to connect the electronic device 800 to a network, and an input/output (I/O) interface 828. The electronic device 800 may operate based on an operating system stored in memory 822, such as Window 88 over, Mac O8X, Unix, Linux, FreeB8D, or the like.
In an exemplary embodiment, a computer-readable storage medium comprising instructions, such as the memory 822 comprising instructions, executable by the processor of the electronic device 800 to perform the above-described method is also provided. The storage medium may be a computer-readable storage medium, which may be, for example, a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
In an exemplary embodiment, a computer program product is also provided, which includes instructions executable by a processor of the electronic device 800 to perform the above-described method.
It should be noted that the descriptions of the above-mentioned apparatus, the electronic device, the computer-readable storage medium, the computer program product, and the like according to the method embodiments may also include other embodiments, and specific implementations may refer to the descriptions of the related method embodiments, which are not described in detail herein.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (10)
1. A method for media information processing, the method comprising:
acquiring account characteristics of a target account;
obtaining a probability prediction result according to the account characteristics and a pre-trained probability prediction model; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sequencing model becomes the target medium information;
determining the screening quantity of the medium information corresponding to each sequencing model according to the probability prediction result and a preset medium quantity calculation algorithm; each sequencing model comprises the target sequencing model;
according to the screening quantity of the media information corresponding to each sequencing model, the candidate media information is obtained from the media information sequence sequenced by each sequencing model; the candidate media information is used for determining target media information corresponding to the target account.
2. The media information processing method of claim 1, wherein the obtaining account characteristics of the target account comprises:
acquiring account portrait characteristics and context characteristics of a target account;
and performing feature splicing on the account portrait features and the context features to obtain fusion features, and performing encoding processing on the fusion features to obtain encoded account features.
3. The method as claimed in claim 1, wherein the determining the media information screening amount corresponding to each ranking model according to the probability prediction result and a predetermined media amount calculation algorithm comprises:
acquiring the total quantity of preset candidate medium information;
taking the total amount of the candidate medium information as a constraint condition, and constructing a target function according to the probability prediction result and a preset target planning algorithm; the target function comprises medium information screening parameters corresponding to the sequencing models;
determining parameter values of medium information screening parameters corresponding to the sequencing models contained in the objective function according to the objective function and the constraint conditions; and the parameter values are the screening quantity of the medium information corresponding to each sequencing model.
4. The media information processing method of claim 1, further comprising:
acquiring sample account characteristics and training samples of a target account, wherein the training samples comprise sample target media information and model marks of a sequencing model corresponding to the sample target media information;
and performing model training on the initial probability prediction model according to the sample account characteristics and the training samples to obtain a trained probability prediction model.
5. The method according to claim 4, wherein the obtaining training samples comprises:
acquiring a media information stream; the media information stream comprises a plurality of media information sets;
inputting each media information set into a sorting model to obtain candidate media information corresponding to each sorting model, and adding model marks to the candidate media information, wherein the model marks represent the sorting models corresponding to the candidate media information;
determining sample target media information in each candidate media information according to a preset value estimation model and the candidate media information, and determining a training sample according to the sample target media information and the model mark of the sample target media information.
6. The media information processing method according to claim 5, wherein the training samples include positive samples and negative samples; determining a training sample according to the sample target media information and the model label of the sample target media information, comprising:
if the sample target media information carries a first model mark, determining the sample target media information as positive sample data;
and if the sample target media information carries a second model mark, determining the sample target media information as negative sample data.
7. A media information processing apparatus, comprising:
an acquisition unit configured to perform acquisition of an account characteristic of a target account;
the processing unit is configured to execute a probability prediction model trained in advance according to the account characteristics to obtain a probability prediction result; the probability prediction result reflects the probability that the candidate medium information corresponding to the target sorting model becomes the target medium information;
the determining unit is configured to execute a calculation algorithm according to the probability prediction result and a preset medium quantity, and determine a medium information screening quantity corresponding to each sequencing model; each sequencing model comprises the target sequencing model;
the screening unit is configured to execute screening quantity of the media information corresponding to each sorting model, and candidate media information of the screening quantity of the media information is obtained in the media information sequence sorted by each sorting model; the candidate media information is used for determining target media information corresponding to the target account.
8. A server, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the media information processing method of any one of claims 1 to 6.
9. A computer-readable storage medium, wherein instructions in the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the media information processing method of any one of claims 1 to 6.
10. A computer program product comprising instructions therein, which when executed by a processor of an electronic device, enable the electronic device to perform the media information processing method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210640512.9A CN114896475A (en) | 2022-06-08 | 2022-06-08 | Medium information processing method, medium information processing device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210640512.9A CN114896475A (en) | 2022-06-08 | 2022-06-08 | Medium information processing method, medium information processing device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114896475A true CN114896475A (en) | 2022-08-12 |
Family
ID=82727277
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210640512.9A Pending CN114896475A (en) | 2022-06-08 | 2022-06-08 | Medium information processing method, medium information processing device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114896475A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112559777A (en) * | 2019-09-25 | 2021-03-26 | 北京达佳互联信息技术有限公司 | Content item delivery method and device, computer equipment and storage medium |
| CN112767053A (en) * | 2021-01-29 | 2021-05-07 | 北京达佳互联信息技术有限公司 | Information processing method, information processing device, electronic equipment and storage medium |
| CN112905896A (en) * | 2021-03-30 | 2021-06-04 | 网易传媒科技（北京）有限公司 | Training method of recommended number model, and mixed content recommendation method and device |
| CN114385916A (en) * | 2022-01-11 | 2022-04-22 | 北京达佳互联信息技术有限公司 | Method and device for determining recommended resources, electronic equipment and storage medium |
-
2022
- 2022-06-08 CN CN202210640512.9A patent/CN114896475A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112559777A (en) * | 2019-09-25 | 2021-03-26 | 北京达佳互联信息技术有限公司 | Content item delivery method and device, computer equipment and storage medium |
| CN112767053A (en) * | 2021-01-29 | 2021-05-07 | 北京达佳互联信息技术有限公司 | Information processing method, information processing device, electronic equipment and storage medium |
| CN112905896A (en) * | 2021-03-30 | 2021-06-04 | 网易传媒科技（北京）有限公司 | Training method of recommended number model, and mixed content recommendation method and device |
| CN114385916A (en) * | 2022-01-11 | 2022-04-22 | 北京达佳互联信息技术有限公司 | Method and device for determining recommended resources, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109547814B (en) | Video recommendation method and device, server and storage medium | |
| CN106202186B (en) | Service recommendation method and device based on artificial intelligence | |
| CN108108743B (en) | Abnormal user identification method and device for identifying abnormal user | |
| CN111310055A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN112468853B (en) | Television resource recommendation method and device, computer equipment and storage medium | |
| CN109214374A (en) | Video classification methods, device, server and computer readable storage medium | |
| CN112528164B (en) | User collaborative filtering recall method and device | |
| CN112288554B (en) | Commodity recommendation method and device, storage medium and electronic device | |
| CN111552835B (en) | File recommendation method, device and server | |
| CN114764471A (en) | Recommendation method, recommendation device and storage medium | |
| CN112364184A (en) | Method, device, server and storage medium for ordering multimedia data | |
| CN112884235A (en) | Travel recommendation method, and training method and device of travel recommendation model | |
| CN113362109A (en) | Recommendation method and device for media resources, electronic equipment and storage medium | |
| CN112214677A (en) | Interest point recommendation method and device, electronic equipment and storage medium | |
| CN113806634B (en) | Service package recommending method, device and server | |
| CN112464106B (en) | Object recommendation method and device | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| CN111782774B (en) | Method and device for recommending problems | |
| CN114896475A (en) | Medium information processing method, medium information processing device, electronic equipment and storage medium | |
| CN113946753B (en) | Service recommendation method, device, equipment and storage medium based on location fence | |
| CN114139046B (en) | Object recommendation method and device, electronic equipment and storage medium | |
| CN114463590A (en) | Information processing method, apparatus, device, storage medium, and program product | |
| CN114723469A (en) | Method, system and electronic device for user satisfaction degree prediction and attribution | |
| CN113672804A (en) | Recommendation information generation method, system, computer device and storage medium | |
| CN113506638A (en) | Online consultation theme and machine learning based network training method and system |
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 |