CN112035747A - Information recommendation method and device - Google Patents

Information recommendation method and device Download PDF

Info

Publication number
CN112035747A
CN112035747A CN202010913452.4A CN202010913452A CN112035747A CN 112035747 A CN112035747 A CN 112035747A CN 202010913452 A CN202010913452 A CN 202010913452A CN 112035747 A CN112035747 A CN 112035747A
Authority
CN
China
Prior art keywords
information
recommendation
recommendation information
neural network
piece
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.)
Granted
Application number
CN202010913452.4A
Other languages
Chinese (zh)
Other versions
CN112035747B (en
Inventor
卢建东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN202010913452.4A priority Critical patent/CN112035747B/en
Publication of CN112035747A publication Critical patent/CN112035747A/en
Application granted granted Critical
Publication of CN112035747B publication Critical patent/CN112035747B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (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; the method comprises the following steps: acquiring a historical information sequence and a recommendation information set of a user; determining a correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence; determining behavior characteristics of each piece of recommended information corresponding to the user according to the correlation factor of each piece of recommended information in the recommended information set corresponding to the historical information sequence; performing multiple iterative feature extraction processing on the behavior feature of each piece of recommendation information, and determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information; and executing recommendation operation based on the click rate of each piece of recommendation information. Through the method and the device, the recommendation accuracy can be improved.

Description

Information recommendation method and device
Technical Field
The present application relates to artificial intelligence technologies, and in particular, to an information recommendation method and apparatus, an electronic device, and a computer-readable storage medium.
Background
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, and a ranking stage in a recommendation system usually predicts click rate and ranks based on a machine learning model, and takes a high score as a priority recommendation object. Various efforts are made in the related art to improve the click rate prediction accuracy of the machine learning model, for example, a large amount of feature data is constructed in a feature engineering stage to enable the machine learning model to fully learn, but the applicant finds that the utilization mode of the feature data is lack of pertinence and distinctiveness in the process of implementing the embodiment of the application, so that the diversity interest of users is difficult to effectively depict, and the click rate prediction accuracy and further the accuracy of information recommendation are influenced.
Disclosure of Invention
The embodiment of the application provides an information recommendation method and device, an electronic device and a computer-readable storage medium, which can improve recommendation accuracy.
The technical scheme of the embodiment of the application is realized as follows:
the embodiment of the application provides an information recommendation method, which comprises the following steps:
acquiring a historical information sequence and a recommendation information set of a user;
determining a correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence;
determining behavior characteristics of each piece of recommended information corresponding to the user according to the correlation factor of each piece of recommended information in the recommended information set corresponding to the historical information sequence;
performing multiple iterative feature extraction processing on the behavior feature of each piece of recommendation information, and determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information;
and executing recommendation operation based on the click rate of each piece of recommendation information.
An embodiment of the present application provides an information recommendation device, including:
the acquisition module is used for acquiring a historical information sequence and a recommendation information set of a user;
a correlation factor determining module, configured to determine a correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence;
the behavior characteristic determining module is used for determining the behavior characteristic of each piece of recommended information corresponding to the user according to the correlation factor of each piece of recommended information in the recommended information set corresponding to the historical information sequence;
the click rate determining module is used for performing multiple iterative feature extraction processing on the behavior feature of each piece of recommendation information and determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information;
and the recommending module is used for executing recommending operation based on the click rate of each piece of recommending information.
In the above scheme, the correlation factor of each piece of recommendation information corresponding to the history information sequence includes: each piece of historical information in the historical information sequence corresponds to the correlation factor of each piece of recommended information;
the correlation factor determination module is further configured to:
determining the characteristics of each historical information in the historical information sequence;
for any piece of recommendation information in the recommendation information set and any piece of history information in the history information sequence, executing the following processing:
acquiring the characteristics of the recommendation information;
carrying out opposite subtraction processing on the characteristics of the recommendation information and the characteristics of the historical information to obtain corresponding difference value characteristics;
splicing the features of the recommendation information, the features of the historical information and the corresponding difference features to obtain spliced features corresponding to the historical information;
and carrying out full connection processing on the splicing characteristics of the historical information to obtain the correlation factor of the historical information corresponding to the recommendation information.
In the foregoing solution, the behavior feature determination module is further configured to:
determining the characteristics of each historical information in the historical information sequence;
performing the following for each recommendation information in the set of recommendation information:
and weighting the characteristics of the plurality of historical information by taking the correlation factors of the historical information corresponding to the recommendation information as weights to obtain behavior characteristics representing the user aiming at the recommendation information.
In the foregoing solution, the click rate determining module is further configured to:
determining data characteristics of the user, recommendation environment characteristics of the user and characteristics of each recommendation information in the recommendation information set;
splicing the behavior characteristic, the data characteristic, the recommendation environment characteristic and the characteristic of the recommendation information;
and performing iterative feature extraction processing on the splicing processing result.
In the foregoing solution, the click rate determining module is further configured to:
performing feature extraction processing on the input of an nth neural network model in N cascaded neural network models, and
transmitting the nth feature extraction result output by the nth neural network model to an (n + 1) th neural network model to continue feature extraction processing;
wherein N is an integer with the value increasing from 1, the value range of N is more than or equal to 1 and less than or equal to N-1, and N is an integer more than or equal to 2; and when the value of N is more than or equal to 2 and less than or equal to N-1, the input of the nth neural network model is the feature extraction result of the nth neural network model.
In the above scheme, when N is equal to or greater than 1 and equal to or less than N-1, the nth neural network model includes a one-dimensional convolutional layer and a maximum pooling layer;
the click rate determination module is further configured to:
performing convolution processing on the input of the nth neural network and the one-dimensional convolution layer parameters of the one-dimensional convolution layer of the nth neural network model to obtain an nth convolution layer processing result corresponding to the splicing processing result;
performing maximum pooling processing on the nth convolution layer processing result through a maximum pooling layer of the nth neural network model to obtain an nth feature extraction result output by the nth neural network model;
when the value of N is N-1, the N +1 neural network model comprises the one-dimensional convolutional layer, the folding layer and the maximum pooling layer;
the click rate determination module is further configured to:
performing convolution processing on the nth feature extraction result and one-dimensional convolution layer parameters of a one-dimensional convolution layer of the (n + 1) th neural network model to obtain an (n + 1) th convolution layer processing result corresponding to the nth feature extraction result;
performing pairwise alignment addition processing on convolution characteristic values of adjacent dimensions in the n +1 th convolution layer processing result through the folding layer to obtain a folding processing result;
and performing maximum pooling processing on the folding processing result through a maximum pooling layer of the (n + 1) th neural network model to obtain an (n + 1) th feature extraction result output by the (n + 1) th neural network model.
In the foregoing solution, the click rate determining module is further configured to:
performing convolution processing on the feature value of each dimension input by the nth neural network and the one-dimensional convolution layer parameters to obtain a convolution feature value of each dimension;
and splicing the convolution characteristic values of each dimension to obtain an nth convolution layer processing result based on the one-dimensional convolution layer parameters.
In the foregoing solution, the click rate determining module is further configured to:
performing the following processing for the features of each dimension in the nth convolutional layer processing result:
acquiring a plurality of convolution calculation values of the dimensionality, and performing descending sorting processing on the plurality of convolution calculation values;
determining a plurality of convolution calculation values ranked at the top in the descending ranking result as the maximum pooling processing result of the features of the dimensionality;
and splicing the maximum pooling processing result of the features of each dimension to obtain the nth feature extraction result output by the nth neural network model.
In the foregoing solution, the click rate determining module is further configured to:
performing full-connection processing on the feature extraction result of each piece of recommendation information, and performing maximum likelihood processing on the full-connection processing result to obtain a click rate corresponding to each piece of recommendation information;
the recommendation module is further configured to:
and performing descending sorting processing based on click rate on the recommendation information in the recommendation information set, and executing recommendation operation based on a plurality of recommendation information to be sorted in the front.
An embodiment of the present application provides an electronic device, including:
a memory for storing executable instructions;
and the processor is used for realizing the information recommendation method provided by the embodiment of the application when the processor executes the executable instructions stored in the memory.
The embodiment of the application provides a computer-readable storage medium, which stores executable instructions for causing a processor to execute the method for recommending information provided by the embodiment of the application.
The embodiment of the application has the following beneficial effects:
based on the targeted determination of the correlation factors of the same historical information sequence corresponding to different recommendation information, the behavior characteristics representing the user interest are specifically described for different recommendation information, bidirectional targeted characteristic description between the recommendation information and the user is realized, the diversified interest of the user is effectively described, the information recommendation precision of information recommendation based on the click rate predicted by the behavior characteristics is further ensured, meanwhile, invalid recommendation is effectively avoided, and further, the calculation resources related to recommendation logic in the server are saved.
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 application;
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 application;
FIG. 3A is a diagram of an overall model structure of an artificial intelligence-based information recommendation method according to an embodiment of the present application;
FIG. 3B is a schematic structural diagram of an attention module of an artificial intelligence-based information recommendation method according to an embodiment of the present application;
FIG. 3C is a schematic structural diagram of a deep convolutional neural network module of an artificial intelligence-based information recommendation method according to an embodiment of the present application;
4A-4D are schematic flow charts of artificial intelligence-based information recommendation methods provided by embodiments of the present application;
FIG. 5 is an overall architecture diagram of an artificial intelligence-based information recommendation method provided by an embodiment of the present application;
FIG. 6 is a schematic diagram illustrating feature compression of an artificial intelligence-based information recommendation method according to an embodiment of the present application;
FIG. 7 is a cross-feature diagram of an artificial intelligence-based information recommendation method provided by an embodiment of the present application;
fig. 8A-8B are schematic diagrams of attention mechanism models in the related art.
Detailed Description
In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail with reference to the attached drawings, the described embodiments should not be considered as limiting the present application, 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 application.
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.
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 application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.
Before further detailed description of the embodiments of the present application, terms and expressions referred to in the embodiments of the present application will be described, and the terms and expressions referred to in the embodiments of the present application will be used for the following explanation.
1) Feature compression (Embedding): discrete features are mapped to dense vector space, such as feature "man" to [0.3, 0.4, …, 0.9 ].
2) Predicted click-through rate (Pctr): the predicted click probability means that the possible click probability of certain information (advertisement) is predicted before being displayed under a certain condition, the predicted click rate can be applied to a recommendation system, candidate recommendation objects are ranked based on the click rate predicted probability, and the target objects ranked at the front are recommended to a user.
3) Predicted conversion (Pcvr, prediction conversion rate): the conversion rate of the recommended information is predicted directly by constructing a model from the display data, or click data is obtained by using the display data and the conversion data, the conversion rate is predicted by using the click data and the conversion data, and then the predicted click rate and the predicted conversion rate are multiplied to obtain the final predicted conversion rate.
4) Convolutional Neural Networks (CNN) are a type of feed-forward Neural Networks that include convolution calculations and have a deep structure, and are one of the representative algorithms for deep learning, and Convolutional Neural Networks have a characteristic learning capability, and can perform translation invariant classification on input information according to their hierarchical structure, and are also called "translation invariant artificial Neural Networks".
5) Logistic Regression (LR): the logistic regression assumes that the data obeys Bernoulli distribution, and the parameters are solved by gradient descent by a method of maximizing a likelihood function, so that the purpose of classifying the data into two classes is achieved.
6) Factorization model (FM, Factorization Machine): the factorization machine is a machine learning algorithm based on matrix decomposition and is proposed by Steffen, and the most important characteristic is that the factorization machine has good learning capability on sparse data.
7) Click-through rate (CTR): the ratio of the actual number of clicks of the information to the number of displays of the information;
8) deep Neural Networks (DNN, Deep Neural Networks): it is understood that a neural network with many hidden layers, sometimes called a multi-layer perceptron, is divided according to the positions of different layers, and the neural network layers inside the DNN can be divided into three types, an input layer, a hidden layer and an output layer.
9) Attention mechanism (Attention): the attention is focused on important points, and other unimportant factors are ignored, wherein the judgment of the importance degree depends on the application scene, and the attention is divided into spatial attention and temporal attention according to different application scenes, wherein the former is used for image processing, and the latter is used for natural language processing.
10) Thousand show yields (eCPM): the unit of display can be a webpage, and the unit of display refers to the webpage display income of thousands of times in default, which is only a parameter for reflecting the profitability of the website and does not represent the income.
11) Behavior characteristics: the method refers to the characteristics of the historical behaviors of the user, the historical behaviors comprise the interactive behaviors of the user on the historical information, such as purchasing behaviors, clicking behaviors, commenting behaviors and the like, the characteristics of the historical behaviors are vectorized expressions of historical behavior data, and the historical behavior data is data generated by integrating the interactive behaviors of the user on the historical information.
12) Data characteristics of the user: the method refers to vectorization representation of personal attribute data in a user portrait, wherein the personal attribute data comprises age data, gender data, marital data, occupational data and the like of a user, the data are discrete data, and the data characteristics of the user are obtained after vectorization representation of the discrete data.
According to the technical scheme in the related technology, a CTR estimation model is built based on an LR model, an FM model and a DNN model, cross features need to be built manually on the basis of the CTR estimation model of the LR model, the model generalization effect is poor due to the fact that an embedding layer is not introduced, the FM model is compared with the LR model and the embedding layer is introduced, however, the mode that every two of the features are crossed in a forced mode in the FM model can only learn second-order combination features, high-order combination features cannot be learned, the DNN model is compared with the LR model and the high-order combination features can be learned through a multi-layer sensing machine, and the diversified interest of users cannot.
In order to solve the above problems, embodiments of the present application provide an information recommendation method, an apparatus, an electronic device, and a computer-readable storage medium, which can characterize diversified interests of a user through an attention mechanism, learn a high-order combination feature with a characterization capability in a multi-layer pooling manner, implement bidirectional and targeted feature characterization between recommendation information and the user, improve the characterization capability of a model, and effectively characterize diversified interests of the user, thereby ensuring information recommendation accuracy for information recommendation based on a click rate predicted based on the behavior feature.
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, big data and an artificial intelligence platform, and the server may be directly or indirectly connected through wired or wireless communication, which is not limited herein.
An artificial intelligence cloud Service is also commonly referred to as 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) interface, and some of the sophisticated developers can also use the AI framework and the AI infrastructure provided by the platform to deploy and operate and maintain the own dedicated cloud artificial intelligence services.
Referring to fig. 1, fig. 1 is a schematic diagram of an architecture of an artificial intelligence-based information recommendation system provided in an embodiment of the present application, 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 commodities, an application scene for recommending videos, and the like, and according to different application scenes, the information may be news, video articles, pictures and texts, and the like, and may also be information related to products (e.g., real objects such as clothes, virtual articles such as game props). In the process that a user uses a client, a terminal 400 reports collected user interaction behaviors aiming at information to a server 200 as training sample data and user figures and user characteristics corresponding to the user, the training sample data is behavior data of different users reported by each terminal, training of a click rate prediction model is carried out based on the behavior data, the user figures and the user characteristics are fed back by the terminal corresponding to a certain user, the click rate prediction model determines the click rate of the information based on the user characteristics, the information characteristics and the environmental characteristics, all the recalled information is sorted in a descending order based on the click rate, and determining the head information based on the descending sorting result, wherein the head information can be 200 pieces of information sorted at the top in the recommendation information set, and the recommendation operation is executed according to the 200 pieces of information sorted at the top and the corresponding sorting sequence.
The specific architecture of the information recommendation system is described below, in which the terminal 400 is connected to the server 200 through the network 300, the network 300 may be a wide area network or a local area network, or a combination of the two, and the functions of the server 200 may be abstracted into a click-through rate prediction model and model training. The server 200 receives a recommendation information request of the terminal 400, requests to trigger the operation of a click rate prediction model, determines the click rate of each piece of information recalled from the information database 500 based on logs containing information exposure, click rate and other data reported by the terminal 400, and performs descending sorting according to the click rate to return a plurality of pieces of information in the front sorting to the terminal 400 for presentation, and the terminal 400 reports the logs containing the information exposure, click rate and other data to the server 200 in real time as a training sample for generating real-time characteristics of a user and information real-time characteristics to train the click rate prediction model.
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 application, where the server 200 shown in fig. 2 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.
The memory 250 includes volatile memory or nonvolatile memory, and may include both volatile and nonvolatile memory. The nonvolatile memory may be a Read Only Memory (ROM), and the volatile memory may be a Random Access Memory (RAM). The memory 250 described in embodiments herein is intended to comprise any suitable type of memory.
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 in the embodiments of the present application 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, which may be software in the form of programs and plug-ins, and includes the following software modules: the obtaining module 2551, the correlation factor determining module 2552, the behavior feature determining module 2553, the click rate determining module 2554 and the recommending module 2555 are logical modules, and therefore, any combination or further division can be performed according to the implemented functions, and the functions of the respective modules will be described below.
The information recommendation method based on artificial intelligence provided by the embodiment of the present application will be described with reference to exemplary application and implementation of the information recommendation system provided by the embodiment of the present application, where the information recommendation system includes a training phase and an application phase, and first, a model used in the information recommendation method based on artificial intelligence provided by the embodiment of the present application and training performed by each model are described.
The information recommendation system provided by the embodiment of the application relates to a click rate prediction model, wherein the click rate prediction model comprises an attention module and a deep convolutional neural network module.
Referring to fig. 3A, fig. 3A is a structural diagram of an overall click rate prediction model of an artificial intelligence based information recommendation method provided in an embodiment of the present application, where the overall model includes an attention module and a deep convolutional neural network module, and first, for each piece of recommendation information in a recommendation information set, the attention module receives user behavior data, determines a user behavior feature for any piece of recommendation information in the recommendation information set from the user behavior data, performs iterative feature extraction processing on the user behavior feature through the deep convolutional neural network module, and finally outputs a click rate of each piece of recommendation information in the recommendation information set from the click rate prediction model, so as to perform a recommendation operation according to a descending order result of the click rate.
Referring to fig. 3B, fig. 3B is a structural diagram of an attention module of an artificial intelligence based information recommendation method provided in an embodiment of the present application, where the attention module includes a feature concatenation layer, and is configured to perform a concatenation operation on features (a feature of history information, a feature of recommendation information, and a difference feature between the feature of the history information and the feature of the recommendation information) input to the concatenation layer, and the attention module further includes a full connection layer, and is configured to perform mapping processing on the concatenation features obtained by the concatenation operation, so as to obtain a weight of each piece of history information to the recommendation information.
Referring to fig. 3C, fig. 3C is a schematic structural diagram of a deep convolutional neural network module of an artificial intelligence based information recommendation method provided in this embodiment of the present application, where the deep convolutional neural network module includes multiple sets of network structures and output layers, each set of network structure includes a one-dimensional convolutional layer and a pooling layer, the last set of network structure includes a one-dimensional convolutional layer, a folding layer and a pooling layer, the one-dimensional convolutional layer is a wide convolution, convolution is performed on the same embedding dimension of different features, so that an interaction relationship between different features can be learned, on data with local correlation, a convolutional kernel has a feature extraction capability, features on an information side do not have local correlation, but through continuous one-dimensional convolution and pooling processing, features on a finally obtained feature map are equivalent to feature combinations of very high order, and feature combinations in an unusually wide range are sensed, the pooling layer is used for extracting strong (large characteristic value) k characteristics from the characteristics obtained by the corresponding convolutional layers, the largest k values in the characteristics are selected and reserved by the pooling layer, on one hand, the relative position information of the k values is reserved, on the other hand, more important information is extracted at the same time, the output layer is actually a full connection layer, the input characteristics are converted into specific probabilities, and the probabilities are used as the predicted click rate of the corresponding information.
In some embodiments, the click rate is determined by calling a click rate prediction model, the training process of the click rate prediction model includes that training data (a historical information sequence, a user portrait, environmental data and recommendation information) are transmitted in the model in a forward direction, finally, after the features obtained by the pooling layer are transmitted to the output layer, the output layer obtains a click rate prediction value which is a probability value of whether each recommendation information is clicked or not through a maximum likelihood function, a cross entropy loss function of the model can be obtained by maximizing the likelihood probability value of the model, the loss function is minimized to learn parameters of the model, and the training method adopts a random gradient descent method.
Next, an application of the model in the artificial intelligence based information recommendation method provided in the embodiment of the present application is described. An execution subject of the artificial intelligence based information recommendation method provided by the embodiment of the present application is taken as a server for description, referring to fig. 4A, fig. 4A is a flowchart of the artificial intelligence based information recommendation method provided by the embodiment of the present application, and the description will be given with reference to steps 101 and 105 shown in fig. 4A.
In step 101, a history information sequence and a recommendation information set of a user are obtained.
As an example, the history information sequence is an information sequence in which user behaviors are generated, for example, a sequence formed by information clicked by a user within a window time, a sequence formed by information browsed within the window time, when the information is commodity information, a sequence formed by commodities purchased by the user within the window time, the recommendation information set is a set formed by information obtained at a recall stage of the recommendation system, a recall process is performed from millions of information in practical applications, and a recall manner can adopt a collaborative filtering manner to obtain information corresponding to a user portrait to form a recommendation information set, so that a further recommendation process is continuously performed on the recommendation information set by executing step 102 and step 105.
In step 102, a correlation factor of each recommendation information in the recommendation information set corresponding to the history information sequence is determined.
As an example, the correlation factor of the history information sequence corresponding to each recommendation information includes: each piece of historical information in the historical information sequence corresponds to a relevant factor of each piece of recommended information, for example, the historical information sequence is a commodity identifier (skirt, lipstick, milk powder, book, snack) purchased by a user, and the recommended information set comprises the following pieces of recommended information: the reference meanings of the interactive behaviors of five pieces of historical information of a skirt, a lipstick, a milk powder, a book and a snack for the recommendation information 'lipstick' are different in the process of implementing the embodiment of the application, namely for the recommendation information 'lipstick', the influence correlation of the historical behavior generated by the lipstick for the 'lipstick' is larger than the influence correlation of the historical behavior generated by the milk powder for the 'lipstick', so that in order to truly depict the influence of the historical behavior of a user on the behavior for the recommendation information, the relevant factors of each piece of information in the historical information sequence for the recommendation information are required to be determined, for example, the relevant factors of the five pieces of historical information of the skirt, the lipstick, the milk powder, the book and the snack are respectively determined, the relevant factor of each piece of historical information corresponding to the recommendation information forms the relevant factor of the historical information sequence for the recommendation information, and determining the relevant factors of the five pieces of historical information corresponding to the trousers respectively and determining the relevant factors of the five pieces of historical information corresponding to the shoes respectively according to the thought.
Referring to fig. 4B, fig. 4B is a flowchart illustrating an artificial intelligence-based information recommendation method according to an embodiment of the present application, the determining of the correlation factor of each recommendation information in the recommendation information set corresponding to the historical information sequence in step 102 may be implemented in steps 1021 and 10224, and the step 1022 includes steps 10221 and 10224.
In step 1021, a characteristic of each of the sequence of history information is determined.
In step 1022, the following processing is performed for any recommendation information in the recommendation information set and any history information in the history information sequence:
in step 10221, obtaining characteristics of the recommendation information;
as an example, first obtaining related data of the history information, where the data may be an Identifier (ID) of the history information, a category of the history information, and the like, where the feature representation of the data is discrete sparse features, in order to facilitate subsequent processing, it is necessary to map the discrete features into dense features, that is, to obtain the dense features as features corresponding to the recommendation information through an embedding layer (embedding layer), where the sparse features refer to that the number of non-zero values in the feature vector is much smaller than the dimension (length) of the feature vector, where a sparse feature threshold may be set, that is, feature vectors whose number of non-zero values in the feature vector is smaller than the sparse feature threshold are sparse features, relatively speaking, the dense features refer to vectorized representation of the above sparse features, the dense features refer to features whose number of zero values is smaller than the dense feature threshold, and different dimensions in the dense features may have correlation, therefore, the correlation among information can be described based on dense features, and the model has strong generalization capability.
In step 10222, performing bit-wise subtraction processing on the feature of the recommendation information and the feature of the history information to obtain a corresponding difference feature;
in step 10223, the features of the recommendation information, the features of the history information and the corresponding difference features are spliced to obtain the splicing features corresponding to the history information;
in step 10224, the splicing characteristics of the history information are processed by full connection, and the correlation factor of the recommendation information corresponding to the history information is obtained.
As an example, the feature of the recommendation information and the feature of the history information are subjected to position-to-position subtraction to obtain a corresponding difference feature, for example, the feature of the "skirt" and the feature of the "lipstick" are subjected to position-to-position subtraction to obtain a difference feature, and then the feature of the "skirt" and the feature of the "lipstick" are subjected to splicing processing (referred to as merging operation), so as to obtain a splicing feature, and the splicing feature is input into the full-link layer, so as to obtain a correlation factor of the history information "skirt" corresponding to the recommendation information "lipstick".
As an example, comparing and considering the influence of all behavior records, that is, averaging the features of all the user historical behavior objects (features of historical information) by the average pooling layer to form the behavior features of the user, but the interests of each user are various, and the user has various interests, that is, the historical behavior objects can be completely irrelevant, such as "book" and "milk powder" in the above examples, but when making actual recommendation, the user does not need to excessively take into account the preference previously characterized by the historical behavior for "book" when browsing the recommendation information "diaper", that is, for the task of predicting the click rate of "diaper", the historical behavior of the user for "milk powder" is different from the historical behavior of the user for "book", that is, when predicting the click rate for the recommendation information, the attention to the different historical behaviors is different, and the correlation factor of each user behavior to the recommendation information is generated through the above embodiment.
In step 103, behavior characteristics of each piece of recommendation information corresponding to the user are determined according to the correlation factor of each piece of recommendation information in the recommendation information set corresponding to the history information sequence.
Referring to fig. 4C, fig. 4C is a schematic flowchart of an artificial intelligence based information recommendation method provided in the embodiment of the present application, and in step 103, determining behavior characteristics of each piece of recommendation information corresponding to a user according to a correlation factor of each piece of recommendation information in a recommendation information set corresponding to a history information sequence may be implemented in step 1031-1032.
In step 1031, the characteristics of each history information in the history information sequence are determined.
In step 1032, the following processing is performed for each recommendation information in the recommendation information set: and taking the correlation factor of the recommendation information corresponding to the historical information as a weight, and carrying out weighting processing on the characteristics of the plurality of historical information to obtain the behavior characteristics of the representative user for the recommendation information.
As an example, continuing with the above example, regarding the recommendation information "lipstick", the correlation factor of the history information "skirt" corresponding to the recommendation information "lipstick" is g1, the correlation factor of the history information "lipstick" corresponding to the recommendation information "lipstick" is g2, the correlation factor of the history information "milk powder" corresponding to the recommendation information "lipstick" is g3, the correlation factor of the history information "book" corresponding to the recommendation information "lipstick" is g4, the correlation factor of the history information "snack" corresponding to the recommendation information "lipstick" is g5, the feature of the history information "lipstick" is multiplied by the corresponding correlation factor g1, similar processing is performed on other history information, and then a plurality of multiplication results are added, that is, weighting processing is completed, and finally the behavior feature representing the user for the recommendation information is obtained.
In step 104, the behavior feature of each piece of recommendation information is subjected to feature extraction processing iteratively for a plurality of times, and the click rate of each piece of recommendation information is determined based on the feature extraction result of each piece of recommendation information.
Referring to fig. 4D, fig. 4D is a schematic flowchart of the information recommendation method based on artificial intelligence provided in the embodiment of the present application, and the step 104 of iteratively extracting the feature of the behavior of each recommendation information may be implemented by the step 1041-1043.
In step 1041, data characteristics of the user, recommendation environment characteristics of the user, and characteristics of each recommendation information in the recommendation information set are determined.
By way of example, the data features may be derived from a user representation, such as an age feature of the user, an occupation feature of the user, a geographic feature of the user, and so on, the recommendation environment feature of the user may be a network feature of the user, a client type feature served by the recommendation system, and so on, and each recommendation information in the recommendation set may be an Identification (ID) of the recommendation information, a category of the recommendation information, and so on.
In step 1042, the behavior feature, the data feature, the recommendation environment feature, and the feature of the recommendation information are merged.
As an example, the stitching process (concatemate operation) may be a merge concatenation process, merging and concatenating a plurality of feature vectors into a matrix, and as a result of the stitching process, the object of the stitching process may be a behavior feature and other features output by the embedding layer, that is, features not limited to the above-described data feature, recommended environment feature, and recommended information.
In step 1043, an iterative feature extraction process is performed on the result of the stitching process.
In some embodiments, the iterative feature extraction processing on the splicing processing result in step 1043 may be implemented by the following technical solution: performing feature extraction processing on the input of an nth neural network model through the nth neural network model in the N cascaded neural network models, and transmitting an nth feature extraction result output by the nth neural network model to an (N + 1) th neural network model to continue feature extraction processing; wherein N is an integer with the value increasing from 1, the value range of N is more than or equal to 1 and less than or equal to N-1, and N is an integer more than or equal to 2; and when the value of N is 1, the input of the nth neural network model is a splicing processing result, and when the value of N is more than or equal to 2 and less than or equal to N-1, the input of the nth neural network model is a feature extraction result of the nth-1 neural network model.
As an example, a network formed by cascading a plurality of neural network models performs iterative feature extraction processing on a splicing processing result, where an output of a previous neural network model is an input of a current neural network model, and an output of the current neural network model is an input of a next neural network model.
In some embodiments, when N is equal to or greater than 1 and equal to or less than N-1, the nth neural network model comprises a one-dimensional convolutional layer and a maximum pooling layer; the above-mentioned feature extraction processing is performed on the input of the nth neural network model through the nth neural network model in the N cascaded neural network models, and can be implemented by the following technical scheme: performing convolution processing on the input of the nth neural network and the one-dimensional convolution layer parameters of the one-dimensional convolution layer of the nth neural network model to obtain an nth convolution layer processing result corresponding to the splicing processing result; performing maximum pooling processing on the nth convolution layer processing result through a maximum pooling layer of the nth neural network model to obtain an nth feature extraction result output by the nth neural network model; when the value of N is N-1, the N +1 neural network model comprises the one-dimensional convolutional layer, the folding layer and the maximum pooling layer; the above-mentioned transmission of the nth feature extraction result output by the nth neural network model to the (n + 1) th neural network model for continuing the feature extraction processing can be realized by the following technical scheme: performing convolution processing on the nth feature extraction result and one-dimensional convolution layer parameters of a one-dimensional convolution layer of the (n + 1) th neural network model to obtain an (n + 1) th convolution layer processing result corresponding to the nth feature extraction result; carrying out pairwise alignment addition processing on convolution characteristic values of adjacent dimensions in the n +1 th convolution layer processing result through a folding layer to obtain a folding processing result; and performing maximum pooling on the folding processing result through a maximum pooling layer of the (n + 1) th neural network model to obtain an (n + 1) th feature extraction result output by the (n + 1) th neural network model.
As an example, a plurality of feature maps are obtained by multiplying some weight parameters and then adding some bias parameter between the pooled layer and the next convolutional layer, so as to finally obtain a plurality of feature maps, where the plurality of feature maps are obtained by a plurality of convolution kernels (the one-dimensional convolutional layer parameters), the convolution kernels are applied to each line of the feature matrix, i.e. each dimension represented by a vector, operations between different lines are mutually independent, a dependency relationship between two adjacent lines can be realized by a folding operation, the folding operation can be a para-position addition of vectors of two adjacent lines, so that the dimension represented by the vectors is reduced by half, the operation does not increase the number of parameters, but the association between lines in the feature matrix is considered before the last fully-connected layer.
In some embodiments, the convolution processing is performed on the input of the nth neural network and the one-dimensional convolution layer parameters of the one-dimensional convolution layer of the nth neural network model to obtain the nth convolution layer processing result corresponding to the stitching processing result, and the method may be implemented by the following technical solutions: performing convolution processing on the feature value of each dimension input by the nth neural network and the one-dimensional convolution layer parameters to obtain a convolution feature value of each dimension; and splicing the convolution characteristic values of each dimension to obtain an nth convolution layer processing result based on the one-dimensional convolution layer parameters.
As an example, the convolution process is convolution for each dimension, i.e. convolution is performed on each dimension of the feature vector, respectively, that is, instead of performing multidimensional convolution on the whole sentence by using a convolution kernel with the size [ w, 1], the convolution process is performed on one-dimensional convolution by using a convolution kernel with the size [ w, 1] on each dimension, i.e. each convolution kernel can only move laterally to perform convolution on a certain dimension, so that the convolution process is one-dimensional convolution, and different available information can be captured from different dimensions.
As an example, the convolution operation tends to shorten the input length after convolution (L-w +1, L is the input length, and w is the convolution kernel width), and the convolution process in the information recommendation method of the embodiment of the present application is a wide convolution, which increases the input length (L + w-1), because the window of the wide convolution does not need to cover all the input values, and the part without values can be filled with 0 values, so that the edge information is not lost.
In some embodiments, the maximum pooling processing is performed on the nth convolution layer processing result through the maximum pooling layer of the nth neural network model to obtain the nth feature extraction result output by the nth neural network model, and the method can be implemented by the following technical solutions: performing the following processing for the features of each dimension in the nth convolutional layer processing result: acquiring a plurality of convolution calculation values of dimensionality, and performing descending sorting processing on the plurality of convolution calculation values; determining a plurality of convolution calculation values ranked at the top in the descending ranking result as the maximum pooling processing result of the characteristics of the dimensionality; and splicing the maximum pooling processing result of the features of each dimension to obtain the nth feature extraction result output by the nth neural network model.
As an example, the neural network model closest to the output layer includes a convolutional layer, a folding layer, and a pooling layer, and the other neural network models includeThe method comprises a convolutional layer and a pooling layer, wherein parameters of the pooling layer of each neural network model can be the same or different, namely the number k of the numbers of the features selected by the pooling layer in each neural network model can be directly set to be the same, namely the k features with the maximum value are selected for the input of the next layer in each pooling, or in another implementation mode, the parameters of the pooling layer of each neural network model are set to be different, namely the corresponding k value is determined according to the position of the pooling layer, so that the dynamic pooling process is realized, and the formula of the corresponding relation is as follows
Figure BDA0002664157910000181
Where L represents the current number of convolution layers, L represents the total number of convolution layers in the model, ktopAnd the significance of the dynamic k-max pooling treatment is that semantic feature information with corresponding quantity is extracted from sentences with different lengths so as to ensure the uniformity of subsequent convolutional layers.
In some embodiments, the determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information may be implemented by the following technical solutions: and performing full-connection processing on the feature extraction result of each piece of recommendation information, and performing maximum likelihood processing on the full-connection processing result to obtain the click rate corresponding to each piece of recommendation information.
By way of example, operations such as convolutional layer, pooling layer, etc. are to map raw data to hidden layer feature space, and fully-connected layer is to map learned "distributed feature representation" to sample mark space, and in actual use, the fully-connected layer can be implemented by convolution operations: a fully-connected layer that is fully-connected to the previous layer may be converted to a convolution with a convolution kernel of 1x 1; the fully-connected layer of which the front layer is the convolutional layer can be converted into the global convolution with the convolution kernel of hxw, h and w are respectively the height and the width of the convolution result of the front layer, and the maximum likelihood function processing can be to calculate the output result of the fully-connected layer through a softmax function to obtain the predicted click rate of the corresponding recommendation information.
In step 105, a recommendation operation is performed based on the click-through rate of each piece of recommendation information.
In some embodiments, the step 105 of performing the recommendation operation based on the click-through rate of each piece of recommendation information may be implemented by the following technical solutions: and performing descending sorting processing based on the click rate on the recommendation information in the recommendation information set, and executing recommendation operation based on a plurality of recommendation information sorted in the top.
As an example, performing the recommendation operation based on the top-ranked pieces of recommendation information may be directly pushing the pieces of recommendation information to the user terminal for presentation, or continuing to perform a reordering process on the top-ranked pieces of recommendation information and performing the recommendation operation based on the reordering result.
Next, an exemplary application of the information recommendation method provided in the embodiment of the present application in an actual application scenario will be described.
Referring to fig. 5, fig. 5 is an overall architecture diagram of an artificial intelligence based information recommendation method provided in this embodiment of the present application, taking an advertisement recommendation system as an example, where the recommendation system includes a recall (Candidate Generation) stage and a Ranking (Ranking) stage, where the recall process is a coarse Ranking stage for selecting several hundred Candidate recommendation information that may be clicked by a user, and based on a delay consideration, a general recall stage may select a simpler policy for recall, such as using a collaborative filtering, an LR model, a recall policy based on a user device location information service, and the like, and the recall stage is generally used by multiple recall policies, and obtains Candidate recommendation information that may be clicked by the user, and since the recall stage recalls hundreds of recommendation information, a click rate prediction model used in the recall stage is simpler and has no high requirement on accuracy, and commonly used feature sources are a user image and user historical behavior data, the model for recalling is a model based on vector representation learning, sorting (ranking) is a fine ranking stage, top N pieces of recommendation information which can be clicked by a user are selected, and the sorting stage needs to predict the click rate accurately, so that the sorting stage can use abundant user data, the data of the recommendation information, commonly used characteristic sources are user images, historical behavior data of the user and other data, the model for sorting is a model based on feature combination learning, so that a high-accuracy prediction model is obtained, the click rate prediction is performed by using the high-accuracy prediction model, the click rate is the most basic concept in internet advertisements, the advertisement sorting is generally sorted according to ecpm, the ecpm is 1000 CTR click bid, the click bid is determined by an advertiser and cannot be controlled by a platform side, and the platform side needs to make good click rate prediction according to the characteristics of the user and the characteristics of the advertisement, the CTR is used for ordering the advertisements, the ordering is the core of the bidding advertisements, therefore, click rate prediction is also one of core technologies of the bidding advertisements, a click rate prediction model based on a depth-attention-depth convolutional neural network is mainly applied to the fine ordering stage of various recommendation systems, the click rate prediction model is applied to a news recommendation system, diversified interests of users are described by introducing an attention module, and high-order combination characteristics with representation capacity are learned in a multi-layer pooling mode, so that the click rate prediction precision is improved.
In the field of natural language processing, CNN models are used to perform emotion classification of text. The model can extract important semantic information in the sentence through the combination of words, in a certain sense, the process of grammar parsing is shown in figure 7, FIG. 7 is a schematic diagram of feature intersection of an artificial intelligence-based information recommendation method provided in an embodiment of the present application, where a hierarchical feature tree functions like a syntax parse tree, and is a process of modeling sentence semantics by using the model, it can be seen that the bottom layer gradually transfers upwards by combining adjacent word information, the upper layer combines new word group information, therefore, even the words far away from each other in the sentence have interactive behaviors (or certain semantic relation), and intuitively, the model can extract important semantic information in the sentence through the combination of the words, for example, the combination between "cat" and "sit", the hierarchical feature map acts like a parse tree. In the network, the extraction of combination characteristics is completed through a one-dimensional convolutional layer and a k-max pooling layer, and intuitively, the model can extract important semantic information in sentences through the combination of words and phrases, and finally a syntactic parse tree is constructed in a similar way to complete the emotion classification of the sentences.
In the prediction of click rateThe high-order cross combination of the user features and the features of the recommendation information is needed, in view of the excellent performance of the CNN network in terms of natural language processing emotion analysis, a click rate prediction model based on DCNN is provided, then improvement is performed on the basis of DCNN, diversified interests of the user are modeled through an attention module, a click rate prediction model based on a deep-attention-deep convolutional neural network is provided, referring to fig. 3C, the deep convolutional neural network model comprises an input layer, a plurality of network combinations (one-dimensional convolutional layer and pooling layer) and a full-connected layer, referring to fig. 6, fig. 6 is a feature compression diagram of the artificial intelligence-based information recommendation method provided by the embodiment of the application, advertisement click features are almost all discrete features, the features after unique thermal coding are sparse and have very high dimensionality, and a sparse feature space needs to be mapped into a dense feature space through an embedding layer (embedded in g layer), therefore, the problem of feature sparseness is solved, and in an Embedding layer (Embedding layer), the output of each feature after passing through the Embedding layer is as follows: e.g. of the typei＝Embedding(xi) Where xi represents the i-th feature of the input, eiRepresenting the output vector of the ith feature after passing through the embedding layer, and the expression of all features of each sample after being input into the embedding layer is as follows: a is(0)＝[e1,e2,.....em]Wherein a is(0)Representing the input layer of a deep convolutional neural network, eiRepresenting the embedded features of the ith feature, m representing the number of features, eiHas a dimension K of 10.
The convolution layer of the DCNN model is a one-dimensional convolution, assuming wi∈Rw，si∈Rn，ri∈R(n+w-1)，wiIs the parameter of the one-dimensional convolution (convolution kernel window size), siIs a column vector, r, in the same embedding dimension for different features on the embedding layeriIs the result after one-dimensional convolution, and the calculation formula of the one-dimensional convolution is as follows:
Figure BDA0002664157910000211
the convolution used in the model is a one-dimensional wide convolution, with the convolution being performed over the same embedding dimension for different featuresThe interactive relationship among different features can be learned, on the data with local correlation, a convolution kernel has the capability of feature extraction, the advertisement features do not have the local correlation, but the DCNN performs continuous one-dimensional convolution and pooling processing, and finally the features on the feature map are equivalent to high-order feature combinations, so that the feature combinations in a wider range are sensed.
The pooling layer of the DCNN model extracts stronger k features in the features by using a k-max pooling layer, the pooling layer selects the maximum k values in all the feature values, on one hand, the relative position information in the k feature values is kept, on the other hand, a plurality of important information (k values) are simultaneously extracted, the pooling mode can also adapt to the input of different lengths, because only k values are required to be extracted and applied to the output layer finally, after the features obtained by the final pooling layer are transferred to the output layer, the output layer obtains the probability value of whether the information is clicked in each sample, namely the click rate through a maximum likelihood function (Softmax),
Figure BDA0002664157910000212
wherein, yDCNNFor the result obtained by the full join process, the cross entropy loss function of the model can be obtained by maximizing the likelihood probability value of the click rate prediction model, see formula (1):
Figure BDA0002664157910000213
where T is the training data set, yiIs the true class of the ith sample, xiAnd (2) as the characteristic of the ith sample, theta is a parameter of the model, J (theta) is a cross entropy loss function, the model parameter is learned by minimizing the loss function, and the training method adopts a random gradient descent method.
The click rate prediction model based on the DCNN can finely model the interest of a user by learning high-order combination characteristics, but cannot model diversified interests of the user, so that the model is structurally reconstructed on the basis of the DCNN model, an attention module is introduced before a convolutional layer to depict the diversified interests of the user, the click rate prediction model based on a deep-attention-deep convolutional neural network is provided, an attention mechanism is derived from the field of machine translation, for example, by machine translation, see FIG. 8A, FIG. 8A is an attention mechanism model schematic diagram in the related art, a text content (A, B, C) is read in through a cyclic neural network encoder to obtain a text content vector w (the last hidden layer state of the cyclic neural network), and then another cyclic neural network decoder takes the hidden layer state as a starting state, each word of the target is generated in turn (X, Y, Z), and this attention mechanism has the disadvantage that no matter how long the previous text content is and how much information is contained, the text content is finally compressed into a vector of several hundred dimensions, which means that the larger the text content is, the more information is lost in the final state vector, and after the length of the input sentence is increased, the translation result of the final decoder is significantly deteriorated, because the text content is known at the time of input, the applicant finds that the model can obtain better effect by using all the information of the text content in the decoding process when implementing the embodiment of the present application, and not only by using the last state vector, but also the core idea of the attention module is the same, see fig. 8B, which is a schematic diagram of the attention mechanism model in the related art, and firstly, when generating the state (h7 … h9) on the target side, all text content vectors (h1 … h5) are considered as input, and secondly, not all text content has an influence on the generation of the next state, for example, when translating english articles, it is to be noted that "part of the current translation" is not the whole article, and "Attention" means that the appropriate text content is selected and used to generate the next state, and Attention is a weight vector (usually output of soft max) with dimensions equal to the length of the text content, and the larger weight represents the more important the text content at the corresponding position, so that the applicant finds in implementing the embodiment of the present application that the data of the user's historical browsing should play a different role in characterizing the user vector as the item to be recommended is different, and learns the weight vector to characterize the difference in importance.
The embedded features of user's behavioral interest are constructed from all data of the user's history, and usually a summation process or an averaging process is directly used to obtain a representation of the embedded features of a user, such as accumulating all features in a feature class (field) after embedding, but this loses much detail information, the user's interest is diversified because the user's historical behavior includes many aspects, for example, the historical information (browsing history) of a young mother may include cosmetics, women's clothing, milk powder, baby products, etc., and when the information to be recommended is women's clothing, the user's historical click data on apparel products plays a major role in building the user's embedded features, whereas historically browsed data about baby products does not work to build the embedded features of the user, it is considered to model the user's diverse interests through the attention module.
The click rate prediction model based on the depth-attention-depth convolution neural network can capture diversified user interests from rich user historical click data, and is divided into two modules: an attention module to map sparse identifying features to dense embedded feature vector space; the DCNN module is used for modeling a vector output by the embedding layer and learning high-order cross combination features, the number of features contained in each sample in the recommended data set is indefinite, sparse identification class features of a user need to be mapped to embedded feature vectors with fixed lengths before the deep network is used for modeling, and loss of much information is caused by obtaining the fixed embedded feature vectors in an average pooling-based mode.
Because the user data and the recommendation information data exist in the data, the recommendation information must be considered when designing the attention module to learn the user expression, that is, different user vectors can be obtained according to different recommendation information, before learning the expression of the embedded features of the user by using the attention module, the fixed embedded features are used to represent that there is actually strong mathematical assumed intervention, for example, the user U clicks the product a and the product B at the same time, the embedded feature vector of the user U is assumed to be Vu, and the embedded feature vectors of the products a and B are Ua and Ub, because the user clicks a and B at the same time, Vu, Ua > and < Vu, Ub > are relatively large, and under the limitation of Vu, Ua and Ub are very close, that is, when the attention module is not introduced, it is assumed in advance that the embedded feature vector representations of different products clicked by the user are very close, this is an unreasonable mathematical assumption, because the historical clicking behavior of a user is diversified, for example, it is unreasonable that the user clicks a skirt and milk powder at the same time, and the embedded feature vectors of the milk powder and the skirt are restricted to be similar, see fig. 3B, first, the feature of the historical information and the feature of the recommendation information are combined together as input, and then the input is input to the full connection layer to obtain the weight of the corresponding historical information to the corresponding recommendation information, and the behavior feature calculation formula (2) representing the user's behavior interest is as follows:
Figure BDA0002664157910000241
where Vi represents an embedded feature vector of an item i interacted with (e.g., clicked or purchased) by a user, e.g., an item ID clicked by the user, a category ID of the item, Vu is a user feature vector learned by the attention network from historical behavior data of the user, and g (Vi, Va) is a correlation between a representation of a feature a of an item to be recommended and a representation of a feature i of the user, the correlation being learned by the attention neural network. The diversified interests of the user can be taken into consideration through attention learning to obtain the vector representation of the behavior characteristics of the user, the correlation is calculated through an attention network, the behavior characteristics of the user are obtained by performing weighted accumulation on the characteristics of various historical information, and different behavior characteristic representations Vu of the user can be obtained along with different articles Va to be recommended.
The click rate prediction model based on the deep-attention-deep convolutional neural network uses an attention module to model diversified click data of a user to obtain behavior characteristics of the user changing along with an article to be recommended, uses the deep convolutional neural network to model extracted embedded characteristics, and learns high-order cross combination characteristics, and compared with an LR model, an FM model and a deep-average-DCNN model, the information recommendation method provided by the embodiment of the application has the following experiment results of index improvement as shown in the following table (1):
Figure BDA0002664157910000242
(1) model sorting index comparison table
Compared with an FM model, the ranking capability index (AUC) of the model is improved by 1.04 percent, the embedded layer is characterized to be introduced, advertisement data can be better modeled, and the generalization capability of the model is improved, for a multi-hot data type, the DCNN uses an average mode for the embedded layers with the same feature class (field), the AUC is improved by 3.29 percent compared with the FM model, the high-order combined feature learning capability of the DCNN is proved, the precision of the model can be improved, after the attention network is used, the AUC is improved by 4.07 percent compared with the FM model, after the attention mechanism is introduced, the diversity of the historical interest of a user can be considered, and the modeling of data is more detailed.
The click rate prediction model based on the depth-attention-depth convolutional neural network introduces an attention module to depict the diversified interests of users, learns high-order combination characteristics with characterization capability in a multi-layer pooling mode, further improves the characterization capability of the model, can use an LR model on the basis of the depth-attention-depth convolutional neural network, models strong characteristics through the LR model, learns the high-order combination characteristics through the depth-attention-depth convolutional neural network, and can ensure the memory and the generalization of the final model.
Continuing with the exemplary structure of the information recommendation device 255 provided in the embodiments of the present application as software modules, in some embodiments, as shown in fig. 2, the software modules stored in the information recommendation device 255 of the memory 250 may include: an obtaining module 2551, configured to obtain a history information sequence and a recommendation information set of a user; a correlation factor determining module 2552, configured to determine a correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence; the behavior characteristic determining module 2553 is configured to determine, according to the correlation factor of each piece of recommended information in the recommended information set corresponding to the historical information sequence, a behavior characteristic of each piece of recommended information corresponding to the user; the click rate determining module 2554 is configured to perform multiple iterative feature extraction processing on the behavior feature of each piece of recommendation information, and determine the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information; and a recommending module 2555, configured to perform a recommending operation based on the click rate of each piece of recommended information.
In the above scheme, the correlation factor of the history information sequence corresponding to each piece of recommendation information includes: each piece of historical information in the historical information sequence corresponds to the correlation factor of each piece of recommended information; a correlation factor determination module 2552, further configured to: determining the characteristics of each historical information in the historical information sequence; the following processing is executed for any recommendation information in the recommendation information set and any history information in the history information sequence: acquiring the characteristics of the recommendation information; carrying out contraposition subtraction processing on the characteristics of the recommendation information and the characteristics of the historical information to obtain corresponding difference value characteristics; splicing the characteristics of the recommendation information, the characteristics of the historical information and the corresponding difference characteristics to obtain spliced characteristics corresponding to the historical information; and carrying out full connection processing on the splicing characteristics of the historical information to obtain the correlation factor of the recommendation information corresponding to the historical information.
In the above solution, the behavior feature determining module 2553 is further configured to: determining the characteristics of each historical information in the historical information sequence; performing the following processing for each recommendation information in the recommendation information set: and taking the correlation factor of the recommendation information corresponding to the historical information as a weight, and carrying out weighting processing on the characteristics of the plurality of historical information to obtain the behavior characteristics of the representative user for the recommendation information.
In the above solution, the click rate determining module 2554 is further configured to: determining data characteristics of a user, recommendation environment characteristics of the user and characteristics of each recommendation information in a recommendation information set; performing splicing processing on the behavior characteristic, the data characteristic, the recommendation environment characteristic and the characteristic of the recommendation information; and performing iterative feature extraction processing on the splicing processing result.
In the above solution, the click rate determining module 2554 is further configured to: performing feature extraction processing on the input of an nth neural network model through the nth neural network model in the N cascaded neural network models, and transmitting an nth feature extraction result output by the nth neural network model to an (N + 1) th neural network model to continue feature extraction processing; wherein N is an integer with the value increasing from 1, the value range of N is more than or equal to 1 and less than or equal to N-1, and N is an integer more than or equal to 2; and when the value of N is 1, the input of the nth neural network model is a splicing processing result, and when the value of N is more than or equal to 2 and less than or equal to N-1, the input of the nth neural network model is a feature extraction result of the nth-1 neural network model.
In the scheme, when the value of N is more than or equal to 1 and less than or equal to N-1, the nth neural network model comprises a one-dimensional convolutional layer and a maximum pooling layer; click rate determination module 2554, further configured to: performing convolution processing on the input of the nth neural network and the one-dimensional convolution layer parameters of the one-dimensional convolution layer of the nth neural network model to obtain an nth convolution layer processing result corresponding to the splicing processing result; performing maximum pooling processing on the nth convolution layer processing result through a maximum pooling layer of the nth neural network model to obtain an nth feature extraction result output by the nth neural network model; when the value of N is N-1, the N +1 neural network model comprises the one-dimensional convolutional layer, the folding layer and the maximum pooling layer; click rate determination module 2554, further configured to: performing convolution processing on the nth feature extraction result and one-dimensional convolution layer parameters of a one-dimensional convolution layer of the (n + 1) th neural network model to obtain an (n + 1) th convolution layer processing result corresponding to the nth feature extraction result; carrying out pairwise alignment addition processing on convolution characteristic values of adjacent dimensions in the n +1 th convolution layer processing result through a folding layer to obtain a folding processing result; and performing maximum pooling on the folding processing result through a maximum pooling layer of the (n + 1) th neural network model to obtain an (n + 1) th feature extraction result output by the (n + 1) th neural network model.
In the above solution, the click rate determining module 2554 is further configured to: performing convolution processing on the feature value of each dimension input by the nth neural network and the one-dimensional convolution layer parameters to obtain a convolution feature value of each dimension; and splicing the convolution characteristic values of each dimension to obtain an nth convolution layer processing result based on the one-dimensional convolution layer parameters.
In the above solution, the click rate determining module 2554 is further configured to: performing the following processing for the features of each dimension in the nth convolutional layer processing result: acquiring a plurality of convolution calculation values of dimensionality, and performing descending sorting processing on the plurality of convolution calculation values; determining a plurality of convolution calculation values ranked at the top in the descending ranking result as the maximum pooling processing result of the characteristics of the dimensionality; and splicing the maximum pooling processing result of the features of each dimension to obtain the nth feature extraction result output by the nth neural network model.
In the above solution, the click rate determining module 2554 is further configured to: performing full-connection processing on the feature extraction result of each piece of recommendation information, and performing maximum likelihood processing on the full-connection processing result to obtain the click rate corresponding to each piece of recommendation information; a recommendation module 2555, further configured to: and performing descending sorting processing based on the click rate on the recommendation information in the recommendation information set, and executing recommendation operation based on a plurality of recommendation information to be sorted in the front.
Embodiments of the present application provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the information recommendation method in the embodiment of the present application.
Embodiments of the present application provide a computer-readable storage medium storing executable instructions, which when executed by a processor, cause the processor to perform a method provided by embodiments of the present application, for example, an information recommendation method as shown in fig. 4A-4D.
In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash, 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, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) 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 embodiment of the application, based on the targeted determination of the correlation factors of the same historical information sequence corresponding to different pieces of recommendation information, the behavior characteristics representing the interest of the user are targeted for different pieces of recommendation information, so that bidirectional targeted characteristic mapping between the recommendation information and the user is realized, and the diversified interest of the user is effectively mapped. Therefore, the information recommendation precision of information recommendation based on the click rate predicted by the behavior characteristics is guaranteed, meanwhile, invalid recommendation is effectively avoided, and further computing resources related to recommendation logic in the server are saved.
The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims (10)

1. An information recommendation method, comprising:
acquiring a historical information sequence and a recommendation information set of a user;
determining a correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence;
determining behavior characteristics of each piece of recommended information corresponding to the user according to the correlation factor of each piece of recommended information in the recommended information set corresponding to the historical information sequence;
performing multiple iterative feature extraction processing on the behavior feature of each piece of recommendation information, and determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information;
and executing recommendation operation based on the click rate of each piece of recommendation information.
2. The method of claim 1,
the correlation factor of each piece of recommendation information corresponding to the historical information sequence comprises: each piece of historical information in the historical information sequence corresponds to the correlation factor of each piece of recommended information;
the determining the correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence includes:
determining the characteristics of each historical information in the historical information sequence;
for any piece of recommendation information in the recommendation information set and any piece of history information in the history information sequence, executing the following processing:
acquiring the characteristics of the recommendation information;
carrying out opposite subtraction processing on the characteristics of the recommendation information and the characteristics of the historical information to obtain corresponding difference value characteristics;
splicing the features of the recommendation information, the features of the historical information and the corresponding difference features to obtain spliced features corresponding to the historical information;
and carrying out full connection processing on the splicing characteristics of the historical information to obtain the correlation factor of the historical information corresponding to the recommendation information.
3. The method according to claim 2, wherein the determining the behavior characteristic of the user corresponding to each recommendation information in the recommendation information set according to the correlation factor of the history information sequence corresponding to each recommendation information comprises:
determining the characteristics of each historical information in the historical information sequence;
performing the following for each recommendation information in the set of recommendation information:
and weighting the characteristics of the plurality of historical information by taking the correlation factors of the historical information corresponding to the recommendation information as weights to obtain behavior characteristics representing the user aiming at the recommendation information.
4. The method according to claim 1, wherein the performing a plurality of iterative feature extraction processes on the behavior feature of each piece of recommendation information includes:
determining data characteristics of the user, recommendation environment characteristics of the user and characteristics of each recommendation information in the recommendation information set;
splicing the behavior characteristic, the data characteristic, the recommendation environment characteristic and the characteristic of the recommendation information;
and performing iterative feature extraction processing on the splicing processing result.
5. The method of claim 4, wherein the iteratively performing the feature extraction on the stitching result comprises:
performing feature extraction processing on the input of an nth neural network model in N cascaded neural network models, and
transmitting the nth feature extraction result output by the nth neural network model to an (n + 1) th neural network model to continue feature extraction processing;
wherein N is an integer with the value increasing from 1, the value range of N is more than or equal to 1 and less than or equal to N-1, and N is an integer more than or equal to 2; and when the value of N is more than or equal to 2 and less than or equal to N-1, the input of the nth neural network model is the feature extraction result of the nth neural network model.
6. The method of claim 5,
when N is equal to or greater than 1 and equal to or less than N-1, the nth neural network model comprises a one-dimensional convolutional layer and a maximum pooling layer, and the characteristic extraction processing is performed on the input of the nth neural network model through the nth neural network model in the N cascaded neural network models, and the characteristic extraction processing comprises the following steps:
performing convolution processing on the input of the nth neural network and the one-dimensional convolution layer parameters of the one-dimensional convolution layer of the nth neural network model to obtain an nth convolution layer processing result corresponding to the splicing processing result;
performing maximum pooling processing on the nth convolution layer processing result through a maximum pooling layer of the nth neural network model to obtain an nth feature extraction result output by the nth neural network model;
when the value of N is N-1, the (N + 1) th neural network model comprises the one-dimensional convolutional layer, the folding layer and the maximum pooling layer, and the N characteristic extraction result output by the N neural network model is transmitted to the (N + 1) th neural network model to continue the characteristic extraction processing, which comprises the following steps:
performing convolution processing on the nth feature extraction result and one-dimensional convolution layer parameters of a one-dimensional convolution layer of the (n + 1) th neural network model to obtain an (n + 1) th convolution layer processing result corresponding to the nth feature extraction result;
performing pairwise alignment addition processing on convolution characteristic values of adjacent dimensions in the n +1 th convolution layer processing result through the folding layer to obtain a folding processing result;
and performing maximum pooling processing on the folding processing result through a maximum pooling layer of the (n + 1) th neural network model to obtain an (n + 1) th feature extraction result output by the (n + 1) th neural network model.
7. The method of claim 6,
the performing convolution processing on the input of the nth neural network and the one-dimensional convolution layer parameters of the one-dimensional convolution layer of the nth neural network model to obtain an nth convolution layer processing result corresponding to the splicing processing result includes:
performing convolution processing on the feature value of each dimension input by the nth neural network and the one-dimensional convolution layer parameters to obtain a convolution feature value of each dimension;
and splicing the convolution characteristic values of each dimension to obtain an nth convolution layer processing result based on the one-dimensional convolution layer parameters.
8. The method according to claim 6, wherein the performing maximal pooling on the nth convolutional layer processing result through a maximal pooling layer of the nth neural network model to obtain an nth feature extraction result output by the nth neural network model comprises:
performing the following processing for the features of each dimension in the nth convolutional layer processing result:
acquiring a plurality of convolution calculation values of the dimensionality, and performing descending sorting processing on the plurality of convolution calculation values;
determining a plurality of convolution calculation values ranked at the top in the descending ranking result as the maximum pooling processing result of the features of the dimensionality;
and splicing the maximum pooling processing result of the features of each dimension to obtain the nth feature extraction result output by the nth neural network model.
9. The method according to claim 1, wherein the determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information comprises:
performing full-connection processing on the feature extraction result of each piece of recommendation information, and performing maximum likelihood processing on the full-connection processing result to obtain a click rate corresponding to each piece of recommendation information;
the executing recommendation operation based on the click rate of each piece of recommendation information comprises:
and performing descending sorting processing based on click rate on the recommendation information in the recommendation information set, and executing recommendation operation based on a plurality of recommendation information sorted in front.
10. An information recommendation apparatus, comprising:
the acquisition module is used for acquiring a historical information sequence and a recommendation information set of a user;
a correlation factor determining module, configured to determine a correlation factor of each piece of recommendation information in the recommendation information set corresponding to the historical information sequence;
the behavior characteristic determining module is used for determining the behavior characteristic of each piece of recommended information corresponding to the user according to the correlation factor of each piece of recommended information in the recommended information set corresponding to the historical information sequence;
the click rate determining module is used for performing multiple iterative feature extraction processing on the behavior feature of each piece of recommendation information and determining the click rate of each piece of recommendation information based on the feature extraction result of each piece of recommendation information;
and the recommending module is used for executing recommending operation based on the click rate of each piece of recommending information.
CN202010913452.4A 2020-09-03 2020-09-03 Information recommendation method and device Active CN112035747B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010913452.4A CN112035747B (en) 2020-09-03 2020-09-03 Information recommendation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010913452.4A CN112035747B (en) 2020-09-03 2020-09-03 Information recommendation method and device

Publications (2)

Publication Number Publication Date
CN112035747A true CN112035747A (en) 2020-12-04
CN112035747B CN112035747B (en) 2023-09-29

Family

ID=73591333

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010913452.4A Active CN112035747B (en) 2020-09-03 2020-09-03 Information recommendation method and device

Country Status (1)

Country Link
CN (1) CN112035747B (en)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112288042A (en) * 2020-12-18 2021-01-29 蚂蚁智信(杭州)信息技术有限公司 Updating method and device of behavior prediction system, storage medium and computing equipment
CN112434184A (en) * 2020-12-15 2021-03-02 四川长虹电器股份有限公司 Deep interest network sequencing method based on historical movie posters
CN112785390A (en) * 2021-02-02 2021-05-11 微民保险代理有限公司 Recommendation processing method and device, terminal device and storage medium
CN112989182A (en) * 2021-02-01 2021-06-18 腾讯科技（深圳）有限公司 Information processing method, information processing apparatus, information processing device, and storage medium
CN113128510A (en) * 2021-03-26 2021-07-16 武汉光谷信息技术股份有限公司 Semantic segmentation method and system
CN113344662A (en) * 2021-05-31 2021-09-03 联想(北京)有限公司 Product recommendation method, device and equipment
CN113342868A (en) * 2021-08-05 2021-09-03 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment and computer readable storage medium
CN113407854A (en) * 2021-07-19 2021-09-17 广东艾檬电子科技有限公司 Application recommendation method, device and equipment and computer readable storage medium
CN113468434A (en) * 2021-09-06 2021-10-01 北京搜狐新动力信息技术有限公司 Resource recommendation method, device, readable medium and equipment
CN113641900A (en) * 2021-08-05 2021-11-12 维沃移动通信有限公司 Information recommendation method and device
CN113672803A (en) * 2021-08-02 2021-11-19 杭州网易云音乐科技有限公司 Recommendation method and device, computing equipment and storage medium
CN114048392A (en) * 2022-01-13 2022-02-15 北京达佳互联信息技术有限公司 Multimedia resource pushing method and device, electronic equipment and storage medium
CN114936885A (en) * 2022-07-21 2022-08-23 成都薯片科技有限公司 Advertisement information matching pushing method, device, system, equipment and storage medium
CN116188118A (en) * 2023-04-26 2023-05-30 北京龙智数科科技服务有限公司 Target recommendation method and device based on CTR prediction model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110162700A (en) * 2019-04-23 2019-08-23 腾讯科技（深圳）有限公司 The training method of information recommendation and model, device, equipment and storage medium
US20200034431A1 (en) * 2018-07-25 2020-01-30 Baidu Online Network Technology (Bijing ) Co., Ltd. Method, computer device and readable medium for user's intent mining
CN111368210A (en) * 2020-05-27 2020-07-03 腾讯科技（深圳）有限公司 Information recommendation method and device based on artificial intelligence and electronic equipment
CN111475730A (en) * 2020-04-09 2020-07-31 腾讯科技（北京）有限公司 Information recommendation method and device based on artificial intelligence and electronic equipment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200034431A1 (en) * 2018-07-25 2020-01-30 Baidu Online Network Technology (Bijing ) Co., Ltd. Method, computer device and readable medium for user's intent mining
CN110162700A (en) * 2019-04-23 2019-08-23 腾讯科技（深圳）有限公司 The training method of information recommendation and model, device, equipment and storage medium
CN111475730A (en) * 2020-04-09 2020-07-31 腾讯科技（北京）有限公司 Information recommendation method and device based on artificial intelligence and electronic equipment
CN111368210A (en) * 2020-05-27 2020-07-03 腾讯科技（深圳）有限公司 Information recommendation method and device based on artificial intelligence and electronic equipment

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112434184B (en) * 2020-12-15 2022-03-01 四川长虹电器股份有限公司 Deep interest network sequencing method based on historical movie posters
CN112434184A (en) * 2020-12-15 2021-03-02 四川长虹电器股份有限公司 Deep interest network sequencing method based on historical movie posters
CN112288042B (en) * 2020-12-18 2021-04-02 蚂蚁智信(杭州)信息技术有限公司 Updating method and device of behavior prediction system, storage medium and computing equipment
CN112288042A (en) * 2020-12-18 2021-01-29 蚂蚁智信(杭州)信息技术有限公司 Updating method and device of behavior prediction system, storage medium and computing equipment
CN112989182A (en) * 2021-02-01 2021-06-18 腾讯科技（深圳）有限公司 Information processing method, information processing apparatus, information processing device, and storage medium
CN112989182B (en) * 2021-02-01 2023-12-12 腾讯科技（深圳）有限公司 Information processing method, information processing device, information processing apparatus, and storage medium
CN112785390A (en) * 2021-02-02 2021-05-11 微民保险代理有限公司 Recommendation processing method and device, terminal device and storage medium
CN112785390B (en) * 2021-02-02 2024-02-09 微民保险代理有限公司 Recommendation processing method, device, terminal equipment and storage medium
CN113128510A (en) * 2021-03-26 2021-07-16 武汉光谷信息技术股份有限公司 Semantic segmentation method and system
CN113344662A (en) * 2021-05-31 2021-09-03 联想(北京)有限公司 Product recommendation method, device and equipment
CN113407854A (en) * 2021-07-19 2021-09-17 广东艾檬电子科技有限公司 Application recommendation method, device and equipment and computer readable storage medium
WO2023000491A1 (en) * 2021-07-19 2023-01-26 广东艾檬电子科技有限公司 Application recommendation method, apparatus and device, and computer-readable storage medium
CN113672803A (en) * 2021-08-02 2021-11-19 杭州网易云音乐科技有限公司 Recommendation method and device, computing equipment and storage medium
CN113641900A (en) * 2021-08-05 2021-11-12 维沃移动通信有限公司 Information recommendation method and device
CN113342868A (en) * 2021-08-05 2021-09-03 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment and computer readable storage medium
CN113342868B (en) * 2021-08-05 2021-11-02 腾讯科技（深圳）有限公司 Information recommendation method, device, equipment and computer readable storage medium
CN113468434B (en) * 2021-09-06 2021-12-24 北京搜狐新动力信息技术有限公司 Resource recommendation method, device, readable medium and equipment
CN113468434A (en) * 2021-09-06 2021-10-01 北京搜狐新动力信息技术有限公司 Resource recommendation method, device, readable medium and equipment
CN114048392A (en) * 2022-01-13 2022-02-15 北京达佳互联信息技术有限公司 Multimedia resource pushing method and device, electronic equipment and storage medium
CN114936885A (en) * 2022-07-21 2022-08-23 成都薯片科技有限公司 Advertisement information matching pushing method, device, system, equipment and storage medium
CN116188118A (en) * 2023-04-26 2023-05-30 北京龙智数科科技服务有限公司 Target recommendation method and device based on CTR prediction model
CN116188118B (en) * 2023-04-26 2023-08-29 北京龙智数科科技服务有限公司 Target recommendation method and device based on CTR prediction model

Also Published As

Publication number Publication date
CN112035747B (en) 2023-09-29

Similar Documents

Publication Publication Date Title
CN112035747B (en) Information recommendation method and device
CN111581510B (en) Shared content processing method, device, computer equipment and storage medium
CN111368210B (en) Information recommendation method and device based on artificial intelligence and electronic equipment
JP7104244B2 (en) User tag generation method and its devices, computer programs and computer equipment
US20230009814A1 (en) Method for training information recommendation model and related apparatus
CN110162701B (en) Content pushing method, device, computer equipment and storage medium
EP4145308A1 (en) Search recommendation model training method, and search result sorting method and device
CN111444428A (en) Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium
CN111475730A (en) Information recommendation method and device based on artificial intelligence and electronic equipment
CN111241394B (en) Data processing method, data processing device, computer readable storage medium and electronic equipment
US11989488B2 (en) Automatically and intelligently exploring design spaces
CN111914178A (en) Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium
CN111949886B (en) Sample data generation method and related device for information recommendation
US20210326312A1 (en) Automatically improving data quality
CN117836765A (en) Click prediction based on multimodal hypergraph
US20230342833A1 (en) Recommendation method, recommendation network, and related device
WO2024131762A1 (en) Recommendation method and related device
CN111695037A (en) Information recommendation method and device based on artificial intelligence and electronic equipment
WO2024041483A1 (en) Recommendation method and related device
CN112989212A (en) Media content recommendation method, device and equipment and computer storage medium
WO2024002167A1 (en) Operation prediction method and related apparatus
CN112749330A (en) Information pushing method and device, computer equipment and storage medium
CN113254679A (en) Multimedia resource recommendation method and device, electronic equipment and storage medium
WO2024067779A1 (en) Data processing method and related apparatus
CN114817692A (en) Method, device and equipment for determining recommended object and computer storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant