CN111695037A - Information recommendation method and device based on artificial intelligence and electronic equipment - Google Patents
Information recommendation method and device based on artificial intelligence and electronic equipment Download PDFInfo
- Publication number
- CN111695037A CN111695037A CN202010529198.8A CN202010529198A CN111695037A CN 111695037 A CN111695037 A CN 111695037A CN 202010529198 A CN202010529198 A CN 202010529198A CN 111695037 A CN111695037 A CN 111695037A
- Authority
- CN
- China
- Prior art keywords
- information
- click rate
- piece
- sequence
- determining
- 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
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
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides an information recommendation method, an information recommendation device, electronic equipment and a computer-readable storage medium based on artificial intelligence; the method comprises the following steps: determining the relevance characteristics between the information to be positioned and all information in the information set; wherein the information set comprises at least one of the following information: the located position information; information of a position to be determined; the information of the position is the information of the display position in the sequence of the position which is distributed, and the information of the position to be distributed is the information of the display position in the sequence of the position to be distributed; determining a corresponding first click rate according to the relevance characteristics of each piece of information to be positioned; allocating the display positions which are not allocated in the position sequence and have the highest priority to the information to be positioned with the highest first click rate and marking the information to be positioned as new positioned information; and when the display positions in the position sequence are completely distributed, performing recommendation operation based on each piece of distributed position information and the priority of the corresponding distributed display position. Through the method and the device, the accuracy of the recommendation information can be improved.
Description
Technical Field
The present application relates to information recommendation technologies, and in particular, to an information recommendation method and apparatus based on artificial intelligence, an electronic device, and a computer-readable storage medium.
Background
Cloud computing (cloud computing) is a computing model that distributes computing tasks over a pool of resources formed by a large number of computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The network that provides the resources is referred to as the "cloud". Resources in the "cloud" appear to the user as being infinitely expandable and available at any time, available on demand, expandable at any time, and paid for on-demand.
Artificial Intelligence (AI) is a theory, method and technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results.
Information recommendation is an important application of artificial intelligence, a rearrangement module in the related technology is the final stage of personalized recommendation of a recommendation system, and the rearrangement module breaks up information generated by a sorting module according to a preset rule and then presents the information to a user, so that the problem that recommendation is lack of personalization due to the fact that information with high repetition degree is continuously presented to the user is solved.
Disclosure of Invention
The embodiment of the invention provides an information recommendation method and device based on artificial intelligence, an electronic device and a computer-readable storage medium, which can improve the information recommendation accuracy rate and further improve the user experience.
The technical scheme of the embodiment of the invention is realized as follows:
the embodiment of the invention provides an information recommendation method based on artificial intelligence, which comprises the following steps:
determining the relevance characteristics between the information to be positioned and all information in the information set;
wherein the set of information comprises at least one of: the located position information; the information of the position to be located; the information of the determined position is the information of the display position in the position sequence which is already distributed in the information set, and the information of the waiting position is the information of the display position in the position sequence which is to be distributed in the information set;
determining a corresponding first click rate according to the relevance characteristics of each piece of information to be positioned;
allocating the display positions which are not allocated in the position sequence and have the highest priority to the information to be positioned with the highest first click rate, and marking the information to be positioned as new positioned information;
and when the display positions in the position sequence are distributed, performing recommendation operation based on each piece of the distributed position information and the priority of the corresponding distributed display position.
The embodiment of the invention provides an information recommendation device based on artificial intelligence, which comprises:
the characteristic acquisition module is used for determining the relevance characteristics between the information to be positioned and all the information in the information set;
wherein the set of information comprises at least one of: the located position information; the information of the position to be located; the information of the located position is the information of the display position in the position sequence which is already distributed in the information set, and the information of the waiting position is the information of the display position in the position sequence which is to be distributed in the information set;
the click rate determining module is used for determining a corresponding first click rate according to the relevance characteristic of each piece of information to be positioned;
the position distribution module is used for distributing the display positions which are not distributed in the position sequence and have the highest priority to the information of the positions to be positioned with the highest first click rate and marking the information of the positions to be positioned as new positioned information;
and the recommending module is used for executing recommending operation based on each piece of the positioned information and the priority of the correspondingly distributed display position when the display positions in the position sequence are distributed.
In the above solution, the characteristic obtaining module is further configured to, before determining the relevance characteristics between the information to be located and all information in the information set,
acquiring basic characteristics of each piece of information in an information base;
performing full connection processing on the basic features based on the general full connection parameters of the information base to obtain a corresponding second click rate;
and performing descending sorting processing on the information base based on the second click rate of each piece of information, and selecting a plurality of pieces of information sorted at the front in a descending sorting result to form the information set.
In the foregoing solution, the feature obtaining module is further configured to:
acquiring the characteristics of each piece of information to be positioned and the characteristics of each piece of positioned information in the information set;
executing the following processing for each piece of the information of the position to be positioned:
performing attention coding processing on the characteristics of each piece of information to be positioned to obtain the association degree between the information to be positioned and each piece of information in the information set;
and determining the association characteristics of the information to be positioned based on the association degree between the information to be positioned and each piece of information in the information set.
In the foregoing solution, the feature obtaining module is further configured to:
performing linear transformation processing on the characteristics of each information in the information set to obtain a query vector, a key vector and a value vector corresponding to each information;
and performing dot multiplication on the query vector of the information to be positioned and the key vector of each piece of information in the information set, and performing normalization processing on the result of the dot multiplication processing based on a maximum likelihood function to obtain the association degree between the information to be positioned and each piece of information in the information set.
In the foregoing solution, the feature obtaining module is further configured to:
determining the degree of association as an attention weight of a vector of values corresponding to the each information;
and carrying out weighting processing on the value vector based on the attention weight to obtain the correlation characteristic of the to-be-positioned confidence based on attention coding processing.
In the foregoing solution, the feature obtaining module is further configured to:
acquiring basic characteristics of each piece of information in the information set;
acquiring a position feature of each piece of the positioned information in the information set, wherein the position feature is used for representing a display position of the positioned information;
taking the basic characteristics of the information to be positioned as the characteristics of each piece of information to be positioned;
and fusing the basic characteristics and the position characteristics of the positioned information to obtain the characteristics of the positioned information.
In the above aspect, the basic feature includes at least one of:
basic attribute characteristics used for representing basic information of a user to be recommended; the interest label characteristics are used for representing interest preferences of the user to be recommended; the environment characteristics are used for representing a recommendation environment for recommending the information to the user to be recommended; a category feature for characterizing a category of the information; source characteristics for characterizing a source of the information; content characteristics for characterizing the content of the information.
In the foregoing solution, the feature obtaining module is further configured to:
performing the following for each information in the set of information:
inquiring a plurality of eigenvectors corresponding to the information from a pre-established eigenvector matrix;
and fusing the plurality of feature vectors of the information to obtain the basic features corresponding to the information.
In the foregoing solution, the click rate determining module is further configured to:
acquiring a to-be-positioned position corresponding to each to-be-positioned information respectively, and acquiring full-connection parameters corresponding to the to-be-positioned positions;
and performing full connection processing on the relevance characteristics of each piece of to-be-positioned information based on the full connection parameter of each piece of to-be-positioned information to obtain a first click rate when the to-be-positioned information is displayed at the corresponding to-be-positioned information.
In the foregoing solution, the feature obtaining module is further configured to: determining new correlation characteristics between the new located information and the new information to be located;
the click rate determination module is further configured to: determining a corresponding new first click rate according to the relevance characteristic of each new piece of information to be positioned;
the position assignment module is further configured to: and allocating the display positions which are not allocated in the position sequence and have the highest priority to the new information to be positioned with the highest first click rate until the display positions in the position sequence are allocated.
In the scheme, the first click rate of each piece of information to be positioned in the information set is obtained by calling a click rate prediction model; the device further comprises: a training module to:
prior to determining the correlation characteristic between the determined position information and the information to be positioned,
acquiring an information sample sequence and a real first click rate of each information sample in the information sample sequence from a recommendation log;
training the click rate prediction model aiming at the kth display position of the position sequence based on the information sample sequence and the corresponding real first click rate to update the parameters corresponding to the kth display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model unchanged in the updating process;
wherein k is an integer greater than or equal to 1, and the other display positions are display positions in the position sequence except the first k-1 display positions;
when the parameters corresponding to the kth display position in the click rate prediction model are unchanged, continuing to train the click rate prediction model for the (k + 1) th display position of the position sequence so as to update the parameters corresponding to the (k + 1) th display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model in the updating process;
and when the parameters corresponding to each display position in the click rate prediction model are determined, determining that the click rate prediction model is trained completely.
In the foregoing solution, the training module is further configured to:
performing the following in the training of the click-through rate prediction model for the kth presentation position of the sequence of positions:
determining a first click rate of each information sample except the first k-1 information samples in the information sample sequence through the click rate prediction model;
determining an error between the predicted first click rate and the true first click rate for each information sample, and back-propagating the error in the click rate prediction model according to the loss function corresponding to the kth display position to
Determining a parameter change value corresponding to the kth display position in the click rate prediction model when the loss function corresponding to the kth display position obtains a minimum value;
and updating the parameters corresponding to the kth display position in the click rate prediction model according to the determined parameter change values.
An embodiment of the present invention provides an electronic device, including:
a memory for storing executable instructions;
and the processor is used for realizing the artificial intelligence based information recommendation method provided by the embodiment of the invention when the executable instructions stored in the memory are executed.
The embodiment of the invention provides a computer-readable storage medium, which stores executable instructions and is used for realizing the artificial intelligence-based information recommendation method provided by the embodiment of the invention when being executed by a processor.
The embodiment of the invention has the following beneficial effects:
based on the correlation characteristics among the information to be positioned, all the positioned information in the information set and the information to be positioned, not only can the relation among a plurality of information in the information sequence be modeled, but also the change of user interest in the process that a user browses the information sequence can be modeled, so that a click rate prediction result which can better accord with an actual recommendation scene is obtained, the position of the plurality of information in the information set is adjusted based on the click rate, and the accuracy rate of the recommendation information and the user experience are improved to the maximum extent.
Drawings
FIG. 1 is a schematic diagram of an architecture of an artificial intelligence-based information recommendation system provided by an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a server 200 applying an artificial intelligence-based information recommendation method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of model training of an artificial intelligence-based information recommendation method according to an embodiment of the present invention;
FIG. 4 is a training flowchart of an artificial intelligence-based information recommendation method according to an embodiment of the present invention;
5A-5D are schematic flow diagrams of artificial intelligence based information recommendation methods provided by embodiments of the invention;
FIG. 6 is a diagram of a recommendation system architecture for an artificial intelligence based information recommendation method according to an embodiment of the present invention;
fig. 7 is a schematic diagram of a model application of an artificial intelligence based information recommendation method according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail with reference to the accompanying drawings, the described embodiments should not be construed as limiting the present invention, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict.
In the description that follows, references to the terms "first", "second", and the like, are intended only to distinguish similar objects and not to indicate a particular ordering for the objects, it being understood that "first", "second", and the like may be interchanged under certain circumstances or sequences of events to enable embodiments of the invention described herein to be practiced in other than the order illustrated or described herein.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to be limiting of the invention.
Before further detailed description of the embodiments of the present invention, terms and expressions mentioned in the embodiments of the present invention are explained, and the terms and expressions mentioned in the embodiments of the present invention are applied to the following explanations.
1) The recommendation system comprises: recommendation systems are a tool for automatically contacting users and information, which can help users find information interesting to them in an information overload environment, and can push information to users interested in them.
2) Information personalized recommendation: recommending information interested by the user to the user according to the interest characteristics and the browsing behavior of the user, wherein the information comprises: merchandise, news, etc.;
3) mask (Mask) location awareness: the position perception mainly adds the position information corresponding to the sample into the basic characteristics of the sample explicitly, gradually releases the position information through a Mask mechanism, aims at the positioned position information, the mask of the position characteristic is 1, the position characteristic is added to the basic characteristic of the information, the first click rate prediction is carried out on the basis of the basic characteristic, aiming at the information to be treated, the mask of the position characteristic is 0, the position characteristic is shielded, the first click rate prediction is carried out only based on the basic characteristic, the mask position perception can be used for distinguishing which information is the positioned information, so that the position characteristic is displayed and perceived, which information is the information to be positioned, masking the location features, dynamically adding the location features to the basic features of the information during the process of determining and adjusting the display location of the information, the influence of the preceding information on the following information is effectively taken into account in the rearrangement model (click-through rate prediction model).
The following technical solutions exist for the rearrangement problem in the related art: the information generated by the sorting module is limited by a preset rule through the scattering module and then presented to the user, the task of the information processing system is to prevent information with high repetition degree from being continuously presented to the user so as to enhance the recommendation diversity and effectively control the category of the information obtained by one-time refreshing, the scattering module takes specific rules into consideration, rearranges the information generated by the sorting module based on a recurrent neural network or a pointer, when the rearranging module is used on line, all the information obtained from the sorting module is scattered based on the rules, and finally the scattered information is recommended to the user.
The rearrangement module in the related art performs information scattering mainly based on a single rule, and therefore, the following disadvantages are included:
1. the rearrangement model based on the single rule is mainly used for displaying information to a user through preset strategy control, namely, the information generated by the rearrangement model is simply scattered, the rearrangement model is mainly used for learning the click rate and performing the rearrangement based on a single sample, the influence of the relation between the samples on the rearrangement is not considered, the final effect generated by the rearrangement cannot effectively learn the global sequence characteristics, the position relation between the sample sequences is further ignored, the sample characteristics and the position characteristics cannot be fully utilized to learn the rearrangement model, the rearrangement function of the rearrangement module is weakened, the recurrent neural network type model inputs the initial list into the recurrent neural network type model in sequence, however, the method based on the recurrent neural network has limited capability of modeling the interaction between the information in the list, and the characteristic information of the previous coding item is reduced along with the increase of the coding distance, there are problems of poor interpretability, being affected by the initial input sequence and the coding distance, and finally affecting the sequencing effect presented to the user.
2. The rule learning of the rearrangement module is generated statically, namely the rule learning is predefined before the user experience, cannot change along with the change of the user interest in the browsing process, the position information of each piece of information is not considered explicitly, the change of the interest of the user after the user sees the previous piece of information cannot be simulated sufficiently, and the good experience of the user on information recommendation products is limited to a certain extent.
Aiming at the problems of inaccurate recommendation, incapability of realizing personalized rearrangement and the like of the methods provided by the related technologies, the embodiment of the invention provides an information recommendation method, device, electronic equipment and computer-readable storage medium based on artificial intelligence, which can solve the problems of low recommendation accuracy and lack of personalization in recommendation, and is a position-sensing rearrangement method based on a Mask mechanism, and based on the technical scheme in the related technologies, the position-sensing rearrangement method is improved from the following aspects:
1. the method comprises the steps that position information is dynamically added while information characteristics are utilized, not only is the specific attribute of each information considered, but also the position semantics of the information seen by a user are effectively utilized, so that the recommended information has more integrity, the influence of the information in front on the information behind is learnt explicitly by gradually adding the position information, specifically, the initial characteristics of the bottom layer not only comprise basic characteristics such as information side characteristics, but also directly increase the position characteristics corresponding to the information, the position information is gradually released through a Mask mechanism, the influence of the information in front on the information behind is effectively considered in a model, and the change of the interest of the user is simulated vividly;
2. attention processing is carried out on information learned by a bottom layer, feature expressions of feature vectors of current information after being influenced by feature vectors of other information are contained in the feature vectors of each information, so that the relation among the vectors corresponding to a plurality of information is fully learned, the problem that the relation among the information cannot be fully extracted due to the limitation of coding distance is effectively avoided, and particularly, the relation among the features of the plurality of information is fully expressed by learning the interactivity among the bottom layer features through a converter (Trm, Transformer) module;
3. a full-connection module is added at the high level of a rearrangement model (click rate prediction model), a special full-connection layer is added for each position and used for extracting the position semantics learned by the bottom layer, the parameters of the full-connection layer are matched and adapted with the position information of the position, the full-connection layer does not share the parameters, the position semantics are learned while the click rate semantics are learned by the feature vector of the information, and the position semantics are explicitly sensed at the high level through the full-connection module, so that the loss of the position information is effectively avoided;
4. when the click rate prediction model is trained, in the process of gradually increasing the position information, only the cross entropy loss generated by the information corresponding to the position information vectorization representation which is not determined is calculated, so that the test process is circularly carried out until all the vectorization representations of the position information are learned, the invalid loss is not calculated, the calculation is concentrated on the part which is visible in the position information vectorization representation, the click condition of the part which is invisible in the position information vectorization representation is trained and learned, and therefore the learned model can minimize the click rate loss containing the position semantics and maximize the rearrangement effect.
An exemplary application of the electronic device provided by the embodiment of the present invention is described below, and the electronic device provided by the embodiment of the present invention may be implemented as a server. In the following, an exemplary application will be explained when the electronic device is implemented as a server.
The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as cloud service, a cloud database, cloud computing, a cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and a big data and artificial intelligence platform.
The so-called artificial intelligence cloud Service is also generally called AIaaS (AI as a Service, chinese). The method is a service mode of an artificial intelligence platform, and particularly, the AIaaS platform splits several types of common AI services and provides independent or packaged services at a cloud. This service model is similar to the one opened in an AI theme mall: all developers can access one or more artificial intelligence services provided by the platform through an API (application programming interface), and part of the qualified developers can also use an AI framework and an AI infrastructure provided by the platform to deploy and operate and maintain the self-dedicated cloud artificial intelligence services.
Referring to fig. 1, fig. 1 is a schematic structural diagram of an artificial intelligence based information recommendation system according to an embodiment of the present invention, where the information recommendation system may be used to support recommendation scenes of various information, such as an application scene for recommending news, an application scene for recommending goods, an application scene for recommending videos, and the like, and depending on the application scenes, the information may be news, actual goods, video information, graphics, and the like, in the information recommendation system, a terminal 400 is connected to a server 200 through a network 300, the network 300 may be a wide area network or a local area network, or a combination of the two, in response to receiving a request for recommending information from the terminal 400, the function of the information recommendation system is implemented based on various modules in the server 200, and a second click rate determining unit 25521 in a click rate determining module 2552 in the server determines a second click rate of multiple pieces of information in an information database 500, and performing descending sorting on the plurality of information in the information base based on the second click rate to obtain K pieces of information in the descending sorting, continuing to perform first click rate determination on the first K pieces of information in the descending sorting by using a first click rate determination unit 25522 in the click rate determination module 2552, further performing position distribution on the current information with the highest first click rate by using a position distribution module 2553, repeating the above process until all the information is distributed, and sending the distribution result to the terminal by using a recommendation module 2554 so that the terminal presents according to the distribution result.
Referring to fig. 2, fig. 2 is a schematic structural diagram of a server 200 applying an artificial intelligence-based information recommendation method according to an embodiment of the present invention, where the server 200 includes: at least one processor 210, memory 250, and at least one network interface 220. The various components in server 200 are coupled together by a bus system 240. It is understood that the bus system 240 is used to enable communications among the components. The bus system 240 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 240 in fig. 2.
The Processor 210 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.
The memory 250 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid state memory, hard disk drives, optical disk drives, and the like. Memory 250 optionally includes one or more storage devices physically located remotely from processor 210.
In some embodiments, memory 250 is capable of storing data, examples of which include programs, modules, and data structures, or a subset or superset thereof, to support various operations, as exemplified below.
An operating system 251 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;
a network communication module 252 for communicating to other computing devices via one or more (wired or wireless) network interfaces 220, exemplary network interfaces 220 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), among others.
In some embodiments, the artificial intelligence based information recommendation apparatus provided by the embodiments of the present invention may be implemented in software, and fig. 2 illustrates an artificial intelligence based information recommendation apparatus 255 stored in a memory 250, which includes a plurality of modules of an information recommendation system, where the modules may be software in the form of programs and plug-ins, and include the following software modules: a feature acquisition module 2551, a click-through rate determination module 2552, a location assignment module 2553, a recommendation module 2554, and a training module 2555, which are logical and thus can be arbitrarily combined or further split depending on the implemented functions, which will be described below.
The information recommendation method based on artificial intelligence provided by the embodiment of the present invention will be described below with reference to an exemplary application and implementation of the information recommendation system provided by the embodiment of the present invention, where the information recommendation system includes a training phase and an application phase.
Next, a model used in the artificial intelligence based information recommendation method and training performed by the model according to the embodiment of the present invention will be described.
Referring to fig. 3, fig. 3 is a schematic diagram of model training of an artificial intelligence-based information recommendation method provided in an embodiment of the present invention, where a click rate prediction model for rearrangement includes a feature fusion module, a feature association module, a full connection module, and an output module, and for an information i, the feature fusion module is configured to obtain feature vectors corresponding to respective dimensions of the information i, and perform fusion processing, i.e., addition processing, on the feature vectors corresponding to multiple dimensions of the information i to obtain a basic feature F corresponding to the information i, i.e., the basic feature is obtained based on at least one of the feature vectors of the multiple dimensions, so that the obtained basic feature includes at least one of the feature vectors of the multiple dimensions, in addition, in the feature fusion module, position information of the information i needs to be considered, and when the information i is determined to be information displayed at a position j, the position feature P of the position j is visible, the position feature of the position j is added with the basic feature F of the information i to obtain the feature of the information i, when the display position of the information i is not determined at the moment, the basic feature of the information i is directly used as the feature of the information i, the feature association module is actually a self-attention network, the association feature between the information and each piece of information can be obtained through the self-attention network, and then the full connection module is used for performing full connection processing on the association feature of the information, so that the first click rate of the corresponding information is obtained.
In some embodiments, the first click rate of each piece of information to be positioned in the information set is obtained by calling a click rate prediction model; before determining the relevance features between the determined position information and the information to be positioned, the click-through rate prediction model can be trained by the following scheme: acquiring an information sample sequence and a real first click rate of each information sample in the information sample sequence from the recommendation log; training the click rate prediction model aiming at the kth display position of the position sequence based on the information sample sequence and the corresponding real first click rate to update the parameters corresponding to the kth display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model unchanged in the updating process; wherein k is an integer greater than or equal to 1, and the other display positions are display positions in the position sequence except the first k-1 display positions; when the parameters corresponding to the kth display position in the click rate prediction model are unchanged, continuing to train the click rate prediction model for the (k + 1) th display position of the position sequence so as to update the parameters corresponding to the (k + 1) th display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model in the updating process; and when the parameters corresponding to each display position in the click rate prediction model are determined, determining that the click rate prediction model is trained completely.
By way of example, referring to fig. 4, fig. 4 is a training flowchart of the artificial intelligence based information recommendation method provided by the embodiment of the present invention, in the training process, a vectorized representation for each location is trained, namely, the position characteristics of each position are obtained by actual training, the position characteristics of each position obtained by training are applied to the later online prediction stage, the first K pieces of information generated by the sequencing module are initially input, the position characteristics of the K pieces of information are set as unknown, then, the first click rate prediction is carried out through a position perception rearrangement algorithm, the position characteristics of each position are gradually determined along with the training of the network, and in the process of gradual determination, the sum of the corresponding position characteristics and the basic characteristics of the information is used as input and is sent to the network for repeated training until all the position characteristics are obtained by training, and the training is finished.
In some embodiments, the above training of the click-through rate prediction model for the kth display position of the position sequence to update the parameter corresponding to the kth display position in the click-through rate prediction model may be implemented by the following scheme: the following processing is performed in the training of the click rate prediction model for the kth presentation position of the position sequence: determining a first click rate of each information sample except the first k-1 information samples in the information sample sequence through a click rate prediction model; determining an error between the predicted first click rate and the real first click rate of each information sample, and reversely propagating the error in the click rate prediction model according to the loss function corresponding to the kth display position so as to determine a parameter change value corresponding to the kth display position in the click rate prediction model when the loss function corresponding to the kth display position obtains a minimum value; and updating the parameter corresponding to the kth display position in the click rate prediction model according to the determined parameter change value.
As an example, referring to fig. 3, first learning the position characteristics of the first position, inputting the sequence including K pieces of information into the model, obtaining a first click rate of each piece of information at the corresponding position, specifically, as follows, the information sequence is an information sequence that is actually presented to the user and/or obtains the actual click behavior, assuming that K is 3, the actual information sequence is presented in a manner that the first information in the sequence is presented at the first position, the second information in the sequence is presented at the second position, the third information in the sequence is presented at the third position, the first click rate obtained by the model is actually a 3-by-3 first click rate matrix, representing the first click rate of each piece of information at each position, but in training, only the diagonal first click rate is selected for calculation of the loss function, because the relationship between the information in the information sequence having the actual click behavior and the position is the diagonal relationship (the first information in the sequence is presented at the second position) One position, the second information in the sequence is presented at the second position, the third information in the sequence is presented at the third position), and the first click rate of other cases is only a byproduct obtained by training, for example, the first click rate of the first information in the sequence at the third position, which is useless because there is no corresponding real click data, after obtaining three first click rates S of diagonal lines, the error between each S and the corresponding real click data is calculated, the real click data is set in such a way that the real click rate with click behavior is 1 and the real click rate without click behavior is 0, the first click rates of each information are added and then loss calculation is performed to determine the position characteristic (the parameter corresponding to the first position) of the first position, thereby opening the parameter of the first position, and under the condition that the parameters of the first position are known, predicting a first click rate at which second information in the sequence is presented at the second position and a first click rate at which third information in the sequence is presented at the third position, and calculating a loss function based on the errors between the two S and the corresponding real click data to determine the position characteristics (the parameters corresponding to the second position) of the second position, and then opening the parameters of the second position until the parameters of all the positions are trained and determined, thereby completing the whole training process.
Next, an application of the model in the artificial intelligence based information recommendation method provided by the embodiment of the present invention is described. Referring to fig. 5A, fig. 5A is a flowchart illustrating an artificial intelligence based information recommendation method according to an embodiment of the present invention, which will be described with reference to steps 101-104 shown in fig. 5A.
In step 101, the server determines the correlation characteristics between the information to be located and all the information in the information set.
In some embodiments, the set of information includes at least one of the following information: the located position information; the information of the position to be located; the determined position information is information of display positions in the position sequence allocated in the information set, the to-be-determined position information is information of display positions in the position sequence to be allocated in the information set, and the number of the display positions is the same as that of the information.
As an example, all information in the information set constitutes an information sequence, each information in the information sequence is information to be positioned in an initial state, each information in the information sequence has one information to be positioned, a first click rate of each information at the information to be positioned is determined subsequently, for the information sequence in the initial state, the information in the information sequence is sorted according to a second click rate, the second click rate is a click rate output by a fine-ranking model, or the information sequence is an information sequence obtained after adjustment according to the second click rate and a scattering strategy, the positions of the information in the information sequence sorted according to the second click rate are adjusted through the scattering strategy, so that the position distribution of the adjusted information conforms to the scattering strategy, and the scattering strategy can specify the distribution distance between the information of the same label.
In some embodiments, before determining the relevance characteristics between the information to be located and all information in the information set, the following technical solutions may also be performed: acquiring basic characteristics of each piece of information in an information base; performing full connection processing on the basic characteristics based on the general full connection parameters of the information base to obtain a corresponding second click rate; and performing descending sorting processing on the information base based on the second click rate of each piece of information, and selecting a plurality of pieces of information sorted at the front in a descending sorting result to form an information set.
As an example, the refined ranking model is used for ranking the recalled information based on a second click rate, selecting a plurality of pieces of information ranked at the top in a descending ranking result to form a ranked information set, sorting the information in the information set in a descending ranking manner according to the second click rate, and determining the pending position corresponding to each piece of information in an initial state according to the descending ranking result, so as to subsequently output a first click rate of each piece of information at the pending position, for example, information a, information B, and information C exist in the information set, the second click rate of information a is smaller than that of information B, the second click rate of information B is smaller than that of information C, then the pending position of information C is a display position with the highest priority in the position sequence, the pending position of information a is a display position with the lowest priority in the position sequence, the higher priority positions characterize the positions that the user will preferentially see when browsing, e.g., the positions that are ranked relatively top in the same page (relative to the lower priority positions).
As an example, the association features between the information can obtain global information of the entire information sequence, so that the first click rate is predicted not according to independent information prediction, but also in consideration of mutual influence among the information, and is more suitable for a behavior scene of a user performing actual browsing click, and the information browsed by the user is not a single piece of information, but a plurality of pieces of information are presented simultaneously.
As an example, the correlation characteristic between the located information and the information to be located may dynamically characterize the preference change of the user when the located information is perceived by the user before the information to be located is perceived by the user, for example, for the above-mentioned information a, information B, and information C, when the first click rate prediction is performed, the location of the information a is already determined to be the first location with the highest priority, that is, the information a is located, and then when the information a presented at the first location with the highest priority is browsed by the user, the preference interest of the user is inevitably changed, even if slightly changed, by the correlation characteristic between the located information (information a) and the information to be located, the first click rate of the information to be located at the corresponding location can be more accurately predicted, and the change of the interest behavior of the user when actually browsing is approximately simulated, so that the information after the arrangement mode is adjusted can accord with the dynamic change of the user interest.
Based on fig. 5A and fig. 5B, fig. 5B is a schematic flow chart of the artificial intelligence based information recommendation method according to the embodiment of the present invention, the determining of the relevance characteristics between the information to be located and all information in the information set in step 101 may be implemented in step 1011 and step 1012, and will be described with reference to step 1011 and step 1012 shown in fig. 5B.
In step 1011, the characteristics of each piece of information to be located in the set of information, and the characteristics of each piece of located information are obtained.
In some embodiments, the obtaining of the feature of each piece of to-be-positioned information in the information set and the feature of each piece of positioned information may be implemented by the following technical solutions: acquiring basic characteristics of each piece of information in an information set; acquiring the position characteristics of each piece of positioned information in the information set, wherein the position characteristics are used for representing the display position of the positioned information; taking the basic characteristics of the information to be positioned as the characteristics of each piece of information to be positioned; and fusing the basic characteristics and the position characteristics of the positioned information to obtain the characteristics of the positioned information.
As an example, for an information sequence in an initial state, information a, information B, and information C all belong to information to be located, the features of information a, information B, and information C are all basic features of information a, information B, and information C, for an information sequence in adjustment, for example, the position of information a has been determined to be the first position with the highest priority, that is, information a is located information to be located, the features of information B and information C are still the basic features of information B and information C, the feature of information a is a fusion processing result of the basic feature of information a and the position feature of the first position with the highest priority, that is, an addition processing result, so that when subsequently determining an associated feature representing the association of each information to be located with all information, the change of the behavior of interest of a user at the time of actually browsing can be approximately simulated, so that the information after the arrangement mode is adjusted can accord with the dynamic change of the user interest.
In some embodiments, the above-mentioned basic feature of each piece of information in the information set may be implemented by the following technical solutions: the following processing is performed for each information in the set of information: inquiring a plurality of characteristic vectors corresponding to information from a pre-established characteristic vector matrix; and carrying out fusion processing on a plurality of feature vectors of the information to obtain the basic features of the corresponding information.
As an example, the fusion process of the feature vectors may be a summation process of the feature vectors, thereby obtaining a feature corresponding to each piece of information.
In some embodiments, the base features include at least one of: basic attribute characteristics used for representing basic information of a user to be recommended; the interest label characteristics are used for representing interest preferences of the user to be recommended; the environment characteristics are used for representing a recommendation environment for recommending information to a user to be recommended; a category feature for characterizing a category of information; source characteristics for characterizing a source of information; content characteristics for characterizing the content of the information.
As an example, the basic feature is obtained based on at least one of the features of multiple dimensions, that is, the obtained basic feature includes at least one of the features of multiple dimensions, and the information recommendation method provided in the embodiment of the present invention is a recommendation scheme for a same user, so that features of a user side, such as a gender, an age, a region, a home structure, and the like of the user, features of an information side, such as an information category, an information source, and the like, can be obtained, features of an environment side, such as a network environment where the user is located, a network device used by the user, and the like, can be obtained, and a feature vector of at least one of the features of the dimensions is subjected to fusion processing, so as to obtain the basic feature, that is, the feature of each dimension is obtained to model a first click rate of information, so as to implement more personalized recommendation.
In some embodiments, the feature data of the information may be encoded in the feature engineering to obtain the feature vector of the information, which may specifically be implemented by the following technical solutions: converting the characteristic value of the characteristic data into a characteristic index, and performing hash processing on the characteristic index to obtain a characteristic index code; and carrying out Hash processing on the feature names of the feature data to obtain feature name codes, and combining the feature name codes and the feature index codes to obtain the feature vectors of the information.
As an example, in the process of performing hash coding, it is necessary to calculate an index corresponding to feature data and encode the index, in the machine learning process, in order to facilitate implementation of a correlation algorithm, it is often necessary to convert tag data (generally, a character string) into an integer index, or to restore the integer index into a corresponding tag after the calculation is completed, a converter may encode a list of class attribute features (or tags) so as to digitize the class attribute features, the range of the index starts from 0, the process may index corresponding features so that certain algorithms that cannot accept class type features can be used, and the efficiency of a machine learning algorithm such as a decision tree is improved, if input is numerical data, the input may be converted into character type data and then encoded, and a feature name actually represents a feature class, such as an attribute of a feature, the feature data may be mapped to a 64-bit hash space, with the first 16 bits of the 64-bit hash space representing the feature name and the last 48 bits representing the feature index.
In step 1012, the following is performed for each piece of information to be positioned: performing attention coding processing on the characteristics of each piece of information to be positioned to obtain the association degree between the information to be positioned and each piece of information in the information set; and determining the association characteristics of the information to be positioned based on the association degree between the information to be positioned and each piece of information in the information set.
In some embodiments, in step 1012, the attention coding process is performed on the feature of each piece of information to be located to obtain the association degree between the piece of information to be located and each piece of information in the information set, which may be implemented by the following technical solution: performing linear transformation processing on the characteristics of each information in the information set to obtain a query vector, a key vector and a value vector corresponding to each information; and performing dot multiplication on the query vector of the information to be positioned and the key vector of each piece of information in the information set, and performing normalization processing on the dot multiplication result based on a maximum likelihood function to obtain the association degree between the information to be positioned and each piece of information in the information set.
In some embodiments, in step 1012, determining the association characteristic of the information to be located based on the association degree between the information to be located and each piece of information in the information set may be implemented by the following technical solutions: determining the degree of association as an attention weight of a value vector corresponding to each piece of information; and carrying out weighting processing on the value vector based on the attention weight to obtain the correlation characteristic of the confidence to be positioned based on the attention coding processing.
As an example, the attention coding process is performed on the feature of each information in the information set, the attention score of each information on a certain information (information to be positioned) is obtained, the influence of the positioned information (the position feature and the basic feature) on the positioned information and the influence of all the information to be positioned (the basic feature) on the positioned information are obtained, the attention score of each information on the positioned information is not required to be obtained as the position of the positioned information is determined, the attention score can be used as the association degree between the information (information to be positioned) and each information including the information itself, and the mutual influence between the information can be considered by the above method, the real-time characteristics of the information used for determining the first click rate when the user browses the information in real time can be approximately represented, so that the information after adjustment can be in line with the actual dynamic change of the user interest.
As an example, the calculation of the attention score of each information to a certain information is mainly realized by dot multiplication processing of a query vector and a key vector and maximum likelihood processing, after the dot multiplication processing and before the maximum likelihood processing are carried out, the dot multiplication processing result can be divided by the square root of the dimension of the key vector, which can make the gradient more stable, the maximum likelihood processing is realized by softmax function, the function of the maximum likelihood processing is to normalize the attention scores of all the information, the obtained scores are all positive values and are 1, the attention score determines the contribution of each information in the information set to the information, the query vector, the key vector and the value vector of the characteristics of each information are obtained by carrying out linear transformation on the characteristic vector of the information, namely multiplying the characteristic vector of the information with the query parameter, the key parameter and the value parameter, and the query parameter, The key parameter and the value parameter are obtained by model training, and in order to simplify the training process, the query parameter, the key parameter and the value parameter may be uniformly set to 1, that is, linear transformation processing is to multiply the feature vector of the information by 1, so that the obtained query vector, key vector and value vector are the same and are all feature vectors, and the attention score of each information with respect to a certain information and the value vector corresponding to each information are subjected to weighted summation processing, that is, the associated feature based on attention coding processing of the information is obtained.
Through the attention coding processing, the information dependency relationship inside the information set can be learned, so that the internal information relationship of the information set can be captured, and in the long-distance dependency problem, because the attention coding processing is to calculate the attention scores of all information and each information, the maximum path length is only 1 no matter how long the information is, so that the long-distance dependency relationship can be captured.
In step 102, the server determines a corresponding first click rate according to the relevance characteristic of each piece of information to be positioned.
Based on fig. 5A and fig. 5C, fig. 5C is a schematic flow chart of the artificial intelligence based information recommendation method according to the embodiment of the present invention, and the step 102 of determining the corresponding first click rate according to the relevance feature of each piece of information to be located may be implemented by the step 1021-.
In step 1021, a to-be-positioned position corresponding to each to-be-positioned information is obtained, and full-connection parameters corresponding to the to-be-positioned positions are obtained.
In step 1022, based on the full-link parameter of each piece of information to be located, performing full-link processing on the relevance characteristic of each piece of information to be located, to obtain a first click rate when the information to be located is displayed at the corresponding piece of information to be located.
As an example, as the network deepens, what the click rate prediction model learns becomes more abstract and less interpretable, after passing through the feature association module, the position information (position feature) of the bottom layer is lost, so that a plurality of independent fully-connected layers are used in the last layer to determine the first click rate of different positions, as shown in the fully-connected module in fig. 3, that is, each position has a corresponding fully-connected layer, and parameters in the fully-connected layers are used to learn the feature information specific to the position.
In step 103, the server assigns the unassigned and highest-priority display position in the position sequence to the information of the position to be located with the highest first click rate, and marks the display position as new located information.
As an example, the display positions in the position sequence are sequentially allocated to the information to be positioned according to the descending order of the priority of the display positions, and for the information a, the information B and the information C, in the process of predicting the first click rate (position allocation is not performed yet), the information a, the information B and the information C are predicted at the first click rate corresponding to the information to be positioned, so as to obtain a first click rate PA、PBAnd PCThe corresponding determination mode of the to-be-positioned is as described above, if the second click rate of the information a is smaller than the second click rate of the information B, and the second click rate of the information B is smaller than the second click rate of the information C, the to-be-positioned position of the information C is the display position with the highest priority in the position sequence, the to-be-positioned position of the information a is the display position with the lowest priority in the position sequence, and if P is the display position with the lowest priority in the position sequenceAGreater than PB,PBGreater than PCThen the unassigned and highest priority show position in the sequence of positions is assigned to information a and a is marked as new already-located position information.
In some embodiments, after step 103 is executed, the following technical solutions may also be executed: determining new correlation characteristics between the new located information and the new information to be located; determining a corresponding new first click rate according to the relevance characteristics of each new piece of information to be positioned; and allocating the display positions which are not allocated in the position sequence and have the highest priority to the new information of the positions to be positioned with the highest first click rate until the display positions in the position sequence are allocated.
The allocation of positions according to the present solution is, as an example, actually a process of stepwise allocation of positions, after the first prediction process described above, taking the unallocated display position with the highest priority in the position sequence as the to-be-positioned position of the information B, taking the display position with the lowest priority in the position sequence as the to-be-positioned position of the information C, so as to predict the click rate for the second time, the first click rate of the information B and the information C is output in the second click rate prediction process to carry out second position distribution, after each position is allocated, the information to be positioned is sorted again in descending order according to the latest first click rate prediction result so as to determine the position to be positioned of each information to be positioned at the next first click rate prediction, for the to-be-positioned position, the to-be-positioned position with higher priority is still corresponding to the to-be-positioned position information with higher click rate.
In step 104, when the display positions in the position sequence are allocated, the server executes a recommendation operation based on each allocated position information and the priority of the corresponding allocated display position.
In some embodiments, specifically, a recommendation scenario that requests a plurality of information to be displayed in one page: the number of the information in the information set is consistent with the number of the display positions in the position sequence, namely, the number of the display positions in the page carried by the recommendation request, the background sends the corresponding number of the information and the corresponding priority of the display positions, and the SDK displays the information in the display positions corresponding to the priority of the information.
In some embodiments, it may also be that, in response to the information recommendation request, multiple pieces of information are requested at a time and are displayed in multiple pages, that is, the number of pieces of information requested is greater than the number of display positions in a page, and then the software development component sequentially displays in the display positions in the multiple pages according to the received multiple pieces of information and the priorities of the corresponding display positions, where the priority of the display position in the page that is displayed first is higher than the priority of the display position in the page that is displayed later.
Referring to fig. 5D, fig. 5D is a schematic flowchart of an artificial intelligence-based information recommendation method according to an embodiment of the present invention, in step 200, a server receives an information sequence composed of information in an information set; in step 201, the server determines the relevance characteristics between the information to be positioned and all the information in the information set; in step 202, the server determines a corresponding first click rate according to the relevance characteristic of each piece of information to be positioned; in step 203, the server allocates the display position which is not allocated in the position sequence and has the highest priority to the information of the position to be positioned with the highest first click rate, and marks the information of the position to be positioned as new information of the positioned position; when the unallocated exhibition location still exists, the server performs step 201 again 203, and when the unallocated exhibition location does not exist, the server performs step 204, and the server performs the recommendation operation based on each piece of the already-located information and the priority of the corresponding allocated exhibition location.
An exemplary application of the artificial intelligence based information recommendation method provided by the embodiment of the invention in a practical application scenario will be described below.
The information recommendation method based on artificial intelligence provided by the embodiment of the invention can be applied to a news recommendation system, a model used by the information recommendation method based on artificial intelligence is a rearrangement model (click rate prediction model) constructed based on Mask position perception, the model adjusts the position of information on the basis of an initial ordered list, the initial ordered list is improved by explicitly utilizing the position information of the information, a position feature vector is learned while a basic feature vector (such as an information side feature vector) is learned each time, the learned position feature is a process of gradually releasing through a Mask method, a real physical process of refreshing news from front to back is simulated, the position feature vector represents real position information of a bottom layer, and after the model is learned through a Trm module (feature association module), the feature of each information comprises the cross feature of other information, in order to avoid the loss of bottom-layer position information, a full-connection module for extracting position information of information is introduced at a high layer and is used for sensing the position information of the high layer, so that relevant semantics and position semantics are further extracted, the position semantics are explicit expression of high-layer information vector semantics, the loss function of the model only considers the loss of a sample with unknown position vector, and the specific position characteristic of each position is more fully learned. And finally recommended to the user.
Referring to fig. 6, fig. 6 is a recommendation system architecture diagram of an artificial intelligence based information recommendation method, a news recommendation system includes 4 modules, a user profile service module, a recall service module, a ranking service module and a reordering service module, wherein the profile module mainly includes information clicked by a user in the past and basic registration information, accumulates and stores long-term interest, short-term interest and basic information of the user, provides a basis for recall and ranking, the recall module is responsible for primarily searching information potentially interesting to the user from massive information, the ranking module and the scattering module rank the recalled information and present the recalled information to the user according to a certain rule, the recall algorithm is a filter from an information pool to the user interest information, provides basic data for a subsequent ranking and reordering module, and the ranking module scores the recalled generated information through a corresponding click rate prediction model, the rearrangement module is used as a connection module between the sequencing module and the final user recommendation, the sequenced information is further optimized integrally, the rearrangement module with poor performance influences the information sequence seen by the user, the effect of the rearrangement service in the recommendation system is greatly weakened, and the user experience is further influenced.
The rearrangement service of the related technology breaks up the sorted information based on a specific rule, the information with high dispersion similarity is presented to the position of a user in one brush, the service only considers the influence of the repetition and diversity of the information on final sorting, the average effect of the whole user is good, but the recommendation effect of each user is not optimal, the rearrangement module directly utilizes the sorted information to break up, the sorting result only considers the correlation between single information and the user, the correlation between a plurality of pieces of information aiming at the same user is not considered, the interest of the user can be dynamically changed under the influence of previous information, the information finally recommended to the user by the rearrangement module in the related technology cannot dynamically simulate the change of the interest of the user, and the recommendation accuracy and the user experience are influenced.
In order to solve the problems, the information recommendation method based on artificial intelligence provided by the embodiment of the invention is a Mask position sensing rearrangement method, the rearrangement method based on Mask position sensing is located in a rearrangement module, position information of information is gradually released through a Mask mechanism and is used for rearranging the ordered information, when an information request of a terminal is received, a recall module fully digs the information which is potentially interesting for a user, the recalled information is rearranged through an ordering model of the ordering module and the reordering module, the information is finally recommended to the user, after the user clicks and reads the information according to own preference, real click data is reported to a recommendation system, different modules perform optimization iteration according to the reported user data, and the user experience is continuously improved.
The training process of the artificial intelligence based information recommendation method provided by the embodiment of the invention is shown in fig. 4, an information sequence initially input to an initialization model is a Top-K information descending information sequence which is really presented to a user and generated by a sorting module, the positions of the K information are set as unknown, then the K information sequences are rearranged through a position sensing rearrangement algorithm for multiple times, the vectorization representation of the positions is gradually set as fixed along with the training of a network, and at the moment, the corresponding position vector and the basic feature vector at the information side are added and input to the rearrangement model for repeated training until the vectorization representation of all the positions is determined, so that the training is completed.
Assuming that the number of pieces of recommended information acquired by a news client in response to a refresh request is K, a ranking model of a fine ranking module predicts that K pieces of information are recommended to a user according to a second click rate, where the position of each piece of information is not globally optimal, it is set that the position of the K pieces of information at this time is adjustable, the position of the K pieces of information is set to be unknown, the corresponding position vector is set to be null (that is, the corresponding position vector is masked), it is realized in a network that information presented at the position is already fixed and information presented at the position can be adjusted by introducing a Mask mechanism, as shown in fig. 3, when initial K pieces of information are acquired for the first time (illustrated here as 3 information, where K is 3), position information at all positions are unknown, then the position information at this time will be completely masked, at this time, only a basic feature vector F of a combination of information-side features and user-side features is input at the bottom layer, when the network runs to oneAnd then, selecting the information with the highest first click rate to be displayed at the first position (the position with the highest priority in all the positions), enabling the information at the first position to become known, cancelling the Mask by using the position vector P at the first position, enabling the position vectors at other K-1 positions to be still masked, and then iterating layer by layer until the information displayed at all the positions is known, wherein the sequence of the K pieces of information is also adjusted, and recording the information as0 represents the first information ordering, namely the initial input ordering, when the position information of each information is uncertain, and a vector can be used to record whether the position of the information at the moment is fixed { z }1,z2,…,zKI.e. whether the information to be presented at this position is determined, when z isiWhen the value of (1) is 1, the information to be displayed at the position is fixed, and a position feature vector { P } can be introduced for each position1,P2,…,PKWhen the information to be shown at this position is determined, the position feature vector is visible, whereas the position feature vector is masked, and when the first input is made, the information to be shown at all positions is uncertain, as shown by P in fig. 3, at this time { z }1,z2,…,zKAll are set to 0, i.e. the position vectors are all masked.
The basic feature vector and the position feature vector are subjected to a Trm module (feature association module) to extract the relationship among the feature vectors of a plurality of information, a self-attention network can be used, and the input vectors are subjected to a full connection layer to obtain three vectors: the query vector (query) Q, the key vector (key) K, and the value vector (value) V, the output is the weighting of all the value vectors in V, as shown in equation 1:
wherein, the weight is calculated by the query vector of the characteristic vector and each key vector, and the calculation method comprises three steps: 1) calculating and comparing the similarity of Q and K; 2) carrying out normalization processing based on a maximum likelihood function on the obtained similarity; 3) and (4) carrying out weighted summation on all the value vectors according to the calculated weight to obtain an attention vector (associated characteristic), wherein the attention vector (associated characteristic T) output by the attention network at the moment contains characteristic expression of the characteristic vector of the current information after being influenced by the characteristic vectors of other information, and d is the initial dimension of the input characteristic vector.
Along with the deepening of a network, things learned by a model are more and more abstract, interpretability is more and more poor, after passing through a Trm module, position information designed at the bottom layer is greatly lost, so that a plurality of independent full-connection layers are used at the last layer to extract click rate features and position features of different positions, as shown in a full-connection module in figure 3, namely each position is provided with a corresponding full-connection layer, parameters in the full-connection layers are used for learning characteristic information specific to the position, each full-connection layer not only learns click rate semantics but also learns position semantics, the position semantics is explicit expression of high-level information vector semantics, and for information i, vectors formed by position feature vectors and basic feature vectors are converted through a self-attention network and output associated feature vectors as uiThe full link layer parameter set here is wiAnd then calculating a first click rate based on equation 2:
Ci=Sigmod(wi*ui+bi) (2);
wherein, biFor the bias item, the full link layer at each position outputs K click rates, and the diagonal part is taken as the click rate corresponding to each piece of finally output information:
the Mask of the position is set on the bottom layer based on whether the position of the information is fixed, the information with the fixed position does not participate in the position adjustment of the information, so the loss of the first click rate calculation of the position loses meaning at the moment, and therefore the loss function of each training is calculated based on the formula (3):
wherein k isSubscript starting point of information not fixed at this time, yiRepresenting a first click rate true value for the user.
In the application stage of the model, referring to fig. 7, fig. 7 is a schematic diagram of model application of the artificial intelligence based information recommendation method provided by the embodiment of the present invention, the information is sorted according to the first click rate, the information with the highest first click rate is selected to be located at the first position, and { z } z is set1,z2,…,zKZ in1Set to 1, when the position vector P of the first position is visible, when the information is ordered as1 represents the result after the first round of sequencing, so that the process of the first round of sequencing is completed, the result after the first round of sequencing is input into a network, but the position vector P of the first position at the moment is already visible, before attention processing is not carried out, the characteristic of the information of the first position is the sum of the basic characteristic vector S and the position vector P, after the characteristic association module, each information obtains a corresponding relevance characteristic T, and after the rearrangement network of Mask position perception, the position of the first information is fixed, each information except the information displayed at the first position can obtain a corresponding first click rate S again, so that the rest K-1 information is sequenced, and the information sequence after the sequence is adjusted is obtainedWhereinIs a fixed position in the first wheel,for a fixed position in round 2, { z ] in the third round ordering1,z2,…,zKZ in1、z2Setting the sequence as 1, repeating the steps until the positions of K pieces of information are all fixed, finishing the reordering fine-tuning algorithm, and obtaining the information in the sequence
The training process and the testing process of the model have the following differences: for each piece of training data (information sequence), training can be performed for multiple rounds based on the information sequence, and the position vector of the Mask is gradually released, so that interest change generated after the user sees the previous information is dynamically simulated, and at this time, a loss function for each piece of training data is shown in formula (4):
each user's swipe data is trained K times, with the corresponding number of undetermined positions being 0, 1, 2, 3, …, K-1, respectively.
For each piece of test data, as shown in fig. 7, there is no real data in the prediction process, and the prediction process will gradually determine the information displayed at the positions 1, 2 and 3 according to the position sequence shown in fig. 7 until the display information of each position is determined. After training is completed, parameters of the model are stored, a predicted second click rate of the information is obtained through a sorting module when the model is used on line, K pieces of information which is in front of the second click rate and responds to a primary request is further obtained, characteristics of the K pieces of information are input into the model, a prediction process shown in figure 7 is simulated, finally adjusted K pieces of information are obtained, and then the K pieces of information are presented to a user.
The information recommendation method based on artificial intelligence provided by the embodiment of the invention is applied to a news recommendation system, the rearrangement method of the reordering module is optimized, the position information dynamically enters a network structure through a Mask mechanism, the change of the user interest after seeing the information near the front can be effectively simulated while the relation characteristics between the information are effectively learned, the addition of the characteristic vector and the position characteristic vector of the information side is sent to a Trm module (characteristic association module), so that the characteristic information between different information vectors can be extracted, the relation between the information can be better utilized, and the position information can be perceived at a high level through a unique full connection layer set at each position, the accuracy of the recommendation information of the reordering module is improved, the finally output result can represent the user interest, and the user experience is improved.
Continuing with the exemplary structure of the artificial intelligence based information recommender 255 as implemented as a software module provided in the present invention, in some embodiments, as shown in FIG. 2, the software modules stored in the artificial intelligence based information recommender 255 of the memory 250 may include: a characteristic obtaining module 2551, configured to determine relevance characteristics between the information to be located and all information in the information set; wherein the information set comprises at least one of the following information: the located position information; information of a position to be determined; the information of the position to be allocated is the information of the display position in the sequence of the position to be allocated in the information set; the click rate determining module 2552 is configured to determine a corresponding first click rate according to the relevance characteristic of each piece of information to be located; a position allocating module 2553, configured to allocate, to the to-be-positioned position information with the highest first click rate, an unallocated display position with the highest priority in the position sequence, and mark the to-be-positioned position information as new positioned position information; and a recommending module 2554, configured to, when the display positions in the position sequence are allocated, perform a recommending operation based on each allocated position information and the priority of the corresponding allocated display position.
In some embodiments, the click-through rate determining module 2552 is further configured to, before determining the correlation feature between the determined location information and the information to be located, obtain a basic feature of each information in the information base; performing full connection processing on the basic characteristics based on the general full connection parameters of the information base to obtain a corresponding second click rate; and performing descending sorting processing on the information base based on the second click rate of each piece of information, and selecting a plurality of pieces of information sorted at the front in a descending sorting result to form an information set.
In some embodiments, the feature acquisition module 2551 is further configured to: acquiring the characteristics of each piece of information to be positioned and the characteristics of each piece of positioned information in the information set; for each piece of information to be positioned, the following processing is performed: performing attention coding processing on the characteristics of each piece of information to be positioned to obtain the association degree between the information to be positioned and each piece of information in the information set; and determining the association characteristics of the information to be positioned based on the association degree between the information to be positioned and each piece of information in the information set.
In some embodiments, the feature acquisition module 2551 is further configured to: performing linear transformation processing on the characteristics of each information in the information set to obtain a query vector, a key vector and a value vector corresponding to each information; and performing dot multiplication on the query vector of the information to be positioned and the key vector of each piece of information in the information set, and performing normalization processing on the dot multiplication result based on a maximum likelihood function to obtain the association degree between the information to be positioned and each piece of information in the information set.
In some embodiments, the feature acquisition module 2551 is further configured to: determining the degree of association as an attention weight of a value vector corresponding to each piece of information; and carrying out weighting processing on the value vector based on the attention weight to obtain the correlation characteristic of the confidence to be positioned based on the attention coding processing.
In some embodiments, the feature acquisition module 2551 is further configured to: acquiring basic characteristics of each piece of information in an information set; acquiring the position characteristics of each piece of positioned information in the information set, wherein the position characteristics are used for representing the display position of the positioned information; taking the basic characteristics of the information to be positioned as the characteristics of each piece of information to be positioned; and fusing the basic characteristics and the position characteristics of the positioned information to obtain the characteristics of the positioned information.
In some embodiments, the base features include at least one of: basic attribute characteristics used for representing basic information of a user to be recommended; the interest label characteristics are used for representing interest preferences of the user to be recommended; the environment characteristics are used for representing a recommendation environment for recommending information to a user to be recommended; a category feature for characterizing a category of information; source characteristics for characterizing a source of information; content characteristics for characterizing the content of the information.
In some embodiments, the feature acquisition module 2551 is further configured to: the following processing is performed for each information in the set of information: inquiring a plurality of characteristic vectors corresponding to information from a pre-established characteristic vector matrix; and carrying out fusion processing on a plurality of feature vectors of the information to obtain the basic features of the corresponding information.
In some embodiments, the click rate determination module 2552 is further configured to: acquiring a to-be-positioned position corresponding to each to-be-positioned information respectively, and acquiring full-connection parameters corresponding to the to-be-positioned positions; and performing full connection processing on the relevance characteristics of each piece of to-be-positioned information based on the full connection parameter of each piece of to-be-positioned information to obtain a first click rate when the to-be-positioned information is displayed at the corresponding to-be-positioned position.
In some embodiments, the feature acquisition module 2551 is further configured to: determining new correlation characteristics between the new located information and the new information to be located; click rate determination module 2552, further configured to: determining a corresponding new first click rate according to the relevance characteristics of each new piece of information to be positioned; a position assignment module 2553, further configured to: and allocating the display positions which are not allocated in the position sequence and have the highest priority to the new information of the positions to be positioned with the highest first click rate until the display positions in the position sequence are allocated.
In some embodiments, the first click rate of each piece of information to be positioned in the information set is obtained by calling a click rate prediction model; the apparatus 255 further comprises: a training module 2555 to: before determining the relevance characteristics between the positioned information and the information to be positioned, acquiring an information sample sequence and a real first click rate of each information sample in the information sample sequence from a recommendation log; training the click rate prediction model aiming at the kth display position of the position sequence based on the information sample sequence and the corresponding real first click rate to update the parameters corresponding to the kth display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model unchanged in the updating process; wherein k is an integer greater than or equal to 1, and the other display positions are display positions in the position sequence except the first k-1 display positions; when the parameters corresponding to the kth display position in the click rate prediction model are unchanged, continuing to train the click rate prediction model for the (k + 1) th display position of the position sequence so as to update the parameters corresponding to the (k + 1) th display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model in the updating process; and when the parameters corresponding to each display position in the click rate prediction model are determined, determining that the click rate prediction model is trained completely.
In some embodiments, training module 2555 is further configured to: the following processing is performed in the training of the click rate prediction model for the kth presentation position of the position sequence: determining a first click rate of each information sample except the first k-1 information samples in the information sample sequence through a click rate prediction model; determining an error between the predicted first click rate and the real first click rate of each information sample, and reversely propagating the error in the click rate prediction model according to the loss function corresponding to the kth display position so as to determine a parameter change value corresponding to the kth display position in the click rate prediction model when the loss function corresponding to the kth display position obtains a minimum value; and updating the parameter corresponding to the kth display position in the click rate prediction model according to the determined parameter change value.
Embodiments of the present invention provide a computer-readable storage medium storing executable instructions, which when executed by a processor, cause the processor to perform an artificial intelligence based information recommendation method provided by embodiments of the present invention, for example, the artificial intelligence based information recommendation method shown in fig. 5A-5D.
In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.
In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, may be stored in a portion of a file that holds other programs or data, e.g., in one or more scripts in a HyperText markup Language (H TML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).
By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.
In summary, according to the embodiments of the present invention, based on the correlation characteristics between the information to be located and all the located information in the information set and the information to be located, not only can the relationship among multiple pieces of information in the information sequence be modeled, but also the change of the user interest in the process of browsing the information sequence by the user can be modeled, so as to obtain the click rate prediction result which can better conform to the actual recommended scene, and adjust the location of multiple pieces of information in the information set based on the click rate, thereby maximally improving the accuracy of the recommended information and the user experience.
The above description is only an example of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.
Claims (15)
1. An information recommendation method based on artificial intelligence is characterized by comprising the following steps:
determining the relevance characteristics between the information to be positioned and all information in the information set;
wherein the set of information comprises at least one of: the located position information; the information of the position to be located; the information of the determined position is the information of the display position in the position sequence which is already distributed in the information set, and the information of the waiting position is the information of the display position in the position sequence which is to be distributed in the information set;
determining a corresponding first click rate according to the relevance characteristics of each piece of information to be positioned;
allocating the display positions which are not allocated in the position sequence and have the highest priority to the information to be positioned with the highest first click rate, and marking the information to be positioned as new positioned information;
and when the display positions in the position sequence are distributed, performing recommendation operation based on each piece of the distributed position information and the priority of the corresponding distributed display position.
2. The method of claim 1, wherein prior to determining the association between the information to be located and all information in the set of information, the method further comprises:
acquiring basic characteristics of each piece of information in an information base;
performing full connection processing on the basic features based on the general full connection parameters of the information base to obtain a corresponding second click rate;
and performing descending sorting processing on the information base based on the second click rate of each piece of information, and selecting a plurality of pieces of information sorted at the front in a descending sorting result to form the information set.
3. The method of claim 1, wherein determining the correlation characteristics between the information to be located and all information in the information set comprises:
acquiring the characteristics of each piece of information to be positioned and the characteristics of each piece of positioned information in the information set;
executing the following processing for each piece of the information of the position to be positioned:
performing attention coding processing on the characteristics of each piece of information to be positioned to obtain the association degree between the information to be positioned and each piece of information in the information set;
and determining the association characteristics of the information to be positioned based on the association degree between the information to be positioned and each piece of information in the information set.
4. The method according to claim 3, wherein the performing attention coding processing on the feature of each piece of to-be-positioned information to obtain a degree of association between the piece of to-be-positioned information and each piece of information in the set of information comprises:
performing linear transformation processing on the characteristics of each information in the information set to obtain a query vector, a key vector and a value vector corresponding to each information;
and performing dot multiplication on the query vector of the information to be positioned and the key vector of each piece of information in the information set, and performing normalization processing on the result of the dot multiplication processing based on a maximum likelihood function to obtain the association degree between the information to be positioned and each piece of information in the information set.
5. The method according to claim 3, wherein the determining the association characteristic of the to-be-positioned information based on the association degree between the to-be-positioned information and each information in the information set comprises:
determining the degree of association as an attention weight of a vector of values corresponding to the each information;
and carrying out weighting processing on the value vector based on the attention weight to obtain the correlation characteristic of the to-be-positioned confidence based on attention coding processing.
6. The method according to claim 3, wherein the obtaining the characteristics of each to-be-positioned information and each positioned information in the information set comprises:
acquiring basic characteristics of each piece of information in the information set;
acquiring a position feature of each piece of the positioned information in the information set, wherein the position feature is used for representing a display position of the positioned information;
taking the basic characteristics of the information to be positioned as the characteristics of each piece of information to be positioned;
and fusing the basic characteristics and the position characteristics of the positioned information to obtain the characteristics of the positioned information.
7. The method of claim 6, wherein the base features comprise at least one of:
basic attribute characteristics used for representing basic information of a user to be recommended; the interest label characteristics are used for representing interest preferences of the user to be recommended; the environment characteristics are used for representing a recommendation environment for recommending the information to the user to be recommended; a category feature for characterizing a category of the information; source characteristics for characterizing a source of the information; content characteristics for characterizing the content of the information.
8. The method of claim 6, wherein the obtaining the basic feature of each information in the information set comprises:
performing the following for each information in the set of information:
inquiring a plurality of eigenvectors corresponding to the information from a pre-established eigenvector matrix;
and fusing the plurality of feature vectors of the information to obtain the basic features corresponding to the information.
9. The method according to claim 1, wherein the determining a corresponding first click rate according to the relevance feature of each piece of information to be located comprises:
acquiring a to-be-positioned position corresponding to each to-be-positioned information respectively, and acquiring full-connection parameters corresponding to the to-be-positioned positions;
and performing full connection processing on the relevance characteristics of each piece of to-be-positioned information based on the full connection parameter of each piece of to-be-positioned information to obtain a first click rate when the to-be-positioned information is displayed at the corresponding to-be-positioned information.
10. The method of claim 9, further comprising:
determining new correlation characteristics between the new located information and the new information to be located;
determining a corresponding new first click rate according to the relevance characteristic of each new piece of information to be positioned;
and allocating the display positions which are not allocated in the position sequence and have the highest priority to the new information to be positioned with the highest first click rate until the display positions in the position sequence are allocated.
11. The method of claim 1,
the first click rate of each piece of information to be positioned in the information set is obtained by calling a click rate prediction model;
prior to determining the correlation characteristic between the determined position information and the pending position information, the method further comprises:
acquiring an information sample sequence and a real first click rate of each information sample in the information sample sequence from a recommendation log;
training the click rate prediction model aiming at the kth display position of the position sequence based on the information sample sequence and the corresponding real first click rate to update the parameters corresponding to the kth display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model unchanged in the updating process;
wherein k is an integer greater than or equal to 1, and the other display positions are display positions in the position sequence except the first k-1 display positions;
when the parameters corresponding to the kth display position in the click rate prediction model are unchanged, continuing to train the click rate prediction model for the (k + 1) th display position of the position sequence so as to update the parameters corresponding to the (k + 1) th display position in the click rate prediction model, and fixing the parameters corresponding to other display positions in the click rate prediction model in the updating process;
and when the parameters corresponding to each display position in the click rate prediction model are determined, determining that the click rate prediction model is trained completely.
12. The method of claim 11, wherein the training the click-through rate prediction model for a kth presentation position of a position sequence to update parameters of the click-through rate prediction model for the kth presentation position comprises:
performing the following in the training of the click-through rate prediction model for the kth presentation position of the sequence of positions:
determining a first click rate of each information sample except the first k-1 information samples in the information sample sequence through the click rate prediction model;
determining an error between the predicted first click rate and the true first click rate for each information sample, and back-propagating the error in the click rate prediction model according to the loss function corresponding to the kth display position to
Determining a parameter change value corresponding to the kth display position in the click rate prediction model when the loss function corresponding to the kth display position obtains a minimum value;
and updating the parameters corresponding to the kth display position in the click rate prediction model according to the determined parameter change values.
13. An artificial intelligence-based information recommendation device, comprising:
the characteristic acquisition module is used for determining the relevance characteristics between the information to be positioned and all the information in the information set;
wherein the set of information comprises at least one of: the located position information; the information of the position to be located; the information of the determined position is the information of the display position in the position sequence which is already distributed in the information set, and the information of the waiting position is the information of the display position in the position sequence which is to be distributed in the information set;
the click rate determining module is used for determining a corresponding first click rate according to the relevance characteristic of each piece of information to be positioned;
the position distribution module is used for distributing the display positions which are not distributed in the position sequence and have the highest priority to the information of the positions to be positioned with the highest first click rate and marking the information of the positions to be positioned as new positioned information;
and the recommending module is used for executing recommending operation based on each piece of the positioned information and the priority of the correspondingly distributed display position when the display positions in the position sequence are distributed.
14. An electronic device, comprising:
a memory for storing executable instructions;
a processor for implementing the artificial intelligence based information recommendation method of any one of claims 1 to 12 when executing the executable instructions stored in the memory.
15. A computer-readable storage medium storing executable instructions for implementing the artificial intelligence based information recommendation method of any one of claims 1 to 12 when executed by a processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010529198.8A CN111695037A (en) | 2020-06-11 | 2020-06-11 | Information recommendation method and device based on artificial intelligence and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010529198.8A CN111695037A (en) | 2020-06-11 | 2020-06-11 | Information recommendation method and device based on artificial intelligence and electronic equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111695037A true CN111695037A (en) | 2020-09-22 |
Family
ID=72480328
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010529198.8A Pending CN111695037A (en) | 2020-06-11 | 2020-06-11 | Information recommendation method and device based on artificial intelligence and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111695037A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112528019A (en) * | 2020-12-01 | 2021-03-19 | 清华大学 | Method and device for processing entity relationship in text, electronic equipment and storage medium |
CN112612951A (en) * | 2020-12-17 | 2021-04-06 | 上海交通大学 | Unbiased learning sorting method for income improvement |
CN112765172A (en) * | 2021-01-15 | 2021-05-07 | 齐鲁工业大学 | Log auditing method, device, equipment and readable storage medium |
CN112966577A (en) * | 2021-02-23 | 2021-06-15 | 北京三快在线科技有限公司 | Method and device for model training and information providing |
CN114491319A (en) * | 2022-01-21 | 2022-05-13 | 北京百度网讯科技有限公司 | Information sorting method and device, electronic equipment and storage medium |
CN116910372A (en) * | 2023-09-11 | 2023-10-20 | 腾讯科技(深圳)有限公司 | Information push model processing method and device, information push method and device |
-
2020
- 2020-06-11 CN CN202010529198.8A patent/CN111695037A/en active Pending
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112528019A (en) * | 2020-12-01 | 2021-03-19 | 清华大学 | Method and device for processing entity relationship in text, electronic equipment and storage medium |
CN112612951A (en) * | 2020-12-17 | 2021-04-06 | 上海交通大学 | Unbiased learning sorting method for income improvement |
CN112612951B (en) * | 2020-12-17 | 2022-07-01 | 上海交通大学 | Unbiased learning sorting method for income improvement |
CN112765172A (en) * | 2021-01-15 | 2021-05-07 | 齐鲁工业大学 | Log auditing method, device, equipment and readable storage medium |
CN112966577A (en) * | 2021-02-23 | 2021-06-15 | 北京三快在线科技有限公司 | Method and device for model training and information providing |
CN114491319A (en) * | 2022-01-21 | 2022-05-13 | 北京百度网讯科技有限公司 | Information sorting method and device, electronic equipment and storage medium |
CN116910372A (en) * | 2023-09-11 | 2023-10-20 | 腾讯科技(深圳)有限公司 | Information push model processing method and device, information push method and device |
CN116910372B (en) * | 2023-09-11 | 2024-01-26 | 腾讯科技(深圳)有限公司 | Information push model processing method and device, information push method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111695037A (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
CN108446374B (en) | User's Intention Anticipation method, apparatus, electronic equipment, storage medium | |
CN109902849B (en) | User behavior prediction method and device, and behavior prediction model training method and device | |
WO2021159776A1 (en) | Artificial intelligence-based recommendation method and apparatus, electronic device, and storage medium | |
WO2023061087A1 (en) | Information recommendation method and apparatus, and electronic device, computer-readable storage medium and computer program product | |
Braunhofer et al. | Selective contextual information acquisition in travel recommender systems | |
CN111651692A (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
CN110489582B (en) | Method and device for generating personalized display image and electronic equipment | |
CN112035747A (en) | Information recommendation method and device | |
CN105339933B (en) | Pass through the news results of query expansion | |
CN108345702A (en) | Entity recommends method and apparatus | |
JP2024503774A (en) | Fusion parameter identification method and device, information recommendation method and device, parameter measurement model training method and device, electronic device, storage medium, and computer program | |
KR20190070625A (en) | Method and apparatus for recommending item using metadata | |
US20210109989A1 (en) | Systems and methods for automatically generating and optimizing web pages | |
CN110955831A (en) | Article recommendation method and device, computer equipment and storage medium | |
Ben-Shimon et al. | An ensemble method for top-N recommendations from the SVD | |
CN112883257A (en) | Behavior sequence data processing method and device, electronic equipment and storage medium | |
CN113641835A (en) | Multimedia resource recommendation method and device, electronic equipment and medium | |
CN113793182A (en) | Commodity object recommendation method and device, equipment, medium and product thereof | |
CN110766489A (en) | Method for requesting content and providing content and corresponding device | |
CN112269943B (en) | Information recommendation system and method | |
CN113704620A (en) | User label updating method, device, equipment and medium based on artificial intelligence | |
KR20220071805A (en) | Method, apparatus, and system for production optimization of ad contents | |
CN109299351B (en) | Content recommendation method and device, electronic equipment and computer readable medium | |
US20240005170A1 (en) | Recommendation method, apparatus, electronic device, and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40028508 Country of ref document: HK |
|
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |