Information recommendation method and device, equipment and medium

Abstract

The invention provides an information recommendation method and device, electronic equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining a plurality of domain features of a to-be-recommended task, wherein the plurality of domain features comprise at least one article feature of to-be-recommended information and at least one object feature of a target object; performing feature domain-based layer construction processing on the plurality of domain features through each layer of constructor in the plurality of layers of constructors to obtain cross features of each layer of constructor; performing weighted aggregation processing on the cross features corresponding to the multiple layers of constructors to obtain aggregation features of the to-be-recommended tasks; performing index prediction processing on the aggregation features of the to-be-recommended task to obtain a recommendation index of a target object corresponding to the to-be-recommended information; and executing recommendation operation based on the recommendation index of the target object corresponding to the to-be-recommended information. According to the invention, the accuracy of information recommendation can be improved.

Description

Technical Field

[0001] The present application relates to artificial intelligence technology, and in particular to an information recommendation method, device, electronic device, computer-readable storage medium, and computer program product. Background Art

[0002] Artificial Intelligence (AI) is a comprehensive technology in computer science. By studying the design principles and implementation methods of various intelligent machines, it enables machines to have the functions of perception, reasoning and decision-making. Artificial Intelligence technology is a comprehensive discipline that covers a wide range of fields, such as natural language processing technology and machine learning / deep learning. With the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0003] Recommendation system is one of the important applications in the field of artificial intelligence. It can help users find information that may interest them in an environment with information overload and push the information to users who are interested in it.

[0004] The recommendation system in the related art can determine the information that the user may be interested in from a large amount of information to be recommended, and recommend to the user the information that the user may be interested in. However, the accuracy of information recommendation in the related art needs to be improved. Summary of the invention

[0005] The embodiments of the present application provide an information recommendation method, device, electronic device, computer-readable storage medium, and computer program product, which can improve the accuracy of information recommendation.

[0006] The technical solution of the embodiment of the present application is implemented as follows:

[0007] The present application provides an information recommendation method, including:

[0008] Acquire multiple domain features of the task to be recommended, where the multiple domain features include at least one item feature of the information to be recommended and at least one object feature of the target object;

[0009] By using each layer constructor in the multi-layer constructor, the plurality of domain features are subjected to layer construction processing based on the feature domain, so as to obtain cross features of each layer constructor;

[0010] Performing weighted aggregation processing on the cross features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended;

[0011] Performing index prediction processing on the aggregated features of the task to be recommended to obtain a recommendation index corresponding to the target object of the information to be recommended;

[0012] A recommendation operation is performed based on the recommendation index of the target object corresponding to the information to be recommended.

[0013] The present application provides an information recommendation device, including:

[0014] An acquisition module, configured to acquire a plurality of domain features of the task to be recommended, wherein the plurality of domain features include at least one item feature of the information to be recommended and at least one object feature of the target object;

[0015] A layer construction module, used for performing layer construction processing based on feature domains on the plurality of domain features through each layer constructor in the plurality of layer constructors, so as to obtain cross features of each layer constructor;

[0016] An aggregation module, used for performing weighted aggregation processing on the cross features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended;

[0017] A prediction module, used to perform index prediction processing on the aggregated features of the task to be recommended, to obtain a recommendation index corresponding to the target object of the information to be recommended;

[0018] The recommendation module is used to perform a recommendation operation based on the recommendation index of the target object corresponding to the information to be recommended.

[0019] An embodiment of the present application provides an electronic device for information recommendation, the electronic device comprising:

[0020] Memory for storing computer programs or computer executable instructions;

[0021] The processor is used to implement the information recommendation method provided in the embodiment of the present application when executing the computer program or computer executable instructions stored in the memory.

[0022] An embodiment of the present application provides a computer-readable storage medium storing a computer program or computer-executable instructions, which, when executed by a processor, can implement the information recommendation method provided in the embodiment of the present application.

[0023] An embodiment of the present application provides a computer program product, including a computer program or computer executable instructions, which, when executed by a processor, can implement the information recommendation method provided by the embodiment of the present application.

[0024] The embodiments of the present application have the following beneficial effects:

[0025] Through each layer constructor in the multi-layer constructor, multiple domain features are subjected to layer construction processing based on the feature domain to obtain the cross features of each layer constructor, and the cross features corresponding to the multi-layer constructors are subjected to weighted aggregation processing to obtain the aggregated features of the task to be recommended, thereby fully integrating multiple domain features to improve the accuracy and diversity of the aggregated features, and then performing indicator prediction based on the aggregated features to improve the accuracy of the recommendation indicators and improve the accuracy of the recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of an application scenario of the recommendation system provided in an embodiment of the present application;

[0027] Figure 2 is a schematic diagram of the structure of an electronic device for information recommendation provided in an embodiment of the present application;

[0028] Figure 3A-3D It is a flowchart of the information recommendation method provided in the embodiment of the present application;

[0029] Figure 4 is a schematic diagram of an interface for advertisement recommendation provided in an embodiment of the present application;

[0030] Figure 5 It is a flowchart of the information recommendation method provided in the embodiment of the present application;

[0031] Figure 6 It is a flow chart of an explicit high-order feature cross-model (also called information recommendation model) provided in an embodiment of the present application;

[0032] Figure 7A-7D is a schematic diagram of a characteristic crossover function provided in an embodiment of the present application;

[0033] Figure 8 is a schematic diagram of a layer constructor provided in an embodiment of the present application;

[0034] Figure 9A-9C This is a schematic diagram of a layer aggregator provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0036] In the following description, the terms "first\second" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0037] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be 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.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0039] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0040] 1) Target object: the user currently using the recommendation system, that is, the current user. For example, if user A is using the text recommendation system to read news, then user A is the target object.

[0041] 2) Object feature set: also known as user role, as an effective tool to outline the target object and connect user demands with design direction. Object feature set has been widely used in various fields. In the actual operation process, it often uses the most simple and life-like words to connect the user's attributes, behaviors and expectations as a virtual representative of the actual user.

[0042] 3) Recommendation indicators: indicators used to guide the recommendation system to make recommendations, such as whether the target object will click on the recommended information, whether the target object is interested in the recommended information, whether the target object will convert based on the recommended information, whether the target object will evaluate the recommended information, etc.

[0043] 4) Tasks to be recommended: tasks corresponding to recommendation indicators, that is, tasks that the target object needs to perform on the recommended information based on the recommendation indicators. For example, the target object determines whether the information to be recommended needs to be recommended based on the recommendation indicators, and the target user determines whether the recommendation position of the information to be recommended needs to be adjusted based on the recommendation indicators.

[0044] 5) Click-through rate prediction: It refers to predicting the clicks on each recommended information based on the given recommended information (such as advertisements), users, and context.

[0045] 6) Deep Neural Network (DNN): It is a type of feedforward neural network with a deep structure. It is a technology in the field of machine learning (ML) that can represent complex functions with fewer parameters.

[0046] 7) Embedding method: a method of converting discrete variables into continuous vector representations. Neural network models generally convert discrete features into embedding vectors, concatenate or perform other processing before using them as the input layer of the neural network model.

[0047] 8) Representation: The embedding vector obtained by combining all features in the neural network model, that is, the representation is a high-dimensional vector.

[0048] 9) Hadamard Product: A matrix multiplication operation. The Hadamard product of matrices A and B is a new matrix whose elements are defined as the product of the corresponding elements of matrices A and B.

[0049] The embodiments of the present application provide an information recommendation method, device, electronic device, computer-readable storage medium, and computer program product, which can improve the accuracy of recommendations.

[0050] The information recommendation method provided in the embodiment of the present application can be implemented by the terminal alone; it can also be implemented by the terminal and the server in collaboration. For example, the terminal alone undertakes the information recommendation method described below, or the terminal sends an information recommendation request for the target object to the server, and the server executes the information recommendation method according to the received information recommendation request for the target object, determines the recommendation index of the target object corresponding to the information to be recommended, and performs a recommendation operation based on the recommendation index of the target object corresponding to the information to be recommended, in response to the information recommendation request for the target object.

[0051] The following describes exemplary applications of the electronic device provided in the embodiments of the present application. The electronic device provided in the embodiments of the present application can be implemented as various types of user terminals such as laptops, tablet computers, desktop computers, set-top boxes, mobile devices (for example, mobile phones, portable music players, personal digital assistants, dedicated messaging devices, portable gaming devices, vehicle-mounted devices), smart phones, smart speakers, smart watches, smart televisions, and vehicle-mounted terminals; the electronic device can be implemented as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, among which the cloud service can be an information recommendation service for terminals to call.

[0052] Taking the server as an example, it can be a server cluster deployed in the cloud, opening artificial intelligence cloud services (AI as a Service, AIaaS) to users. The AIaaS platform will split several common AI services and provide independent or packaged services in the cloud. This service model is similar to an AI theme mall. All users can access one or more artificial intelligence services provided by the AIaaS platform through an application programming interface.

[0053] For example, one of the artificial intelligence cloud services can be an information recommendation service, that is, the server in the cloud is encapsulated with the information recommendation program provided by the embodiment of the present application. The user calls the information recommendation service in the cloud service through a terminal (running a client, such as a news client, a video client, etc.), so that the server deployed in the cloud calls the encapsulated information recommendation program, and through each layer constructor in the multi-layer constructor, multiple domain features are subjected to layer construction processing based on the feature domain to obtain the cross-features of each layer constructor; the cross-features corresponding to the multi-layer constructors are subjected to weighted aggregation processing to obtain the aggregated features of the task to be recommended; the aggregated features of the task to be recommended are subjected to index prediction processing to obtain the recommendation index of the target object corresponding to the information to be recommended; the recommendation operation is performed based on the recommendation index of the target object corresponding to the information to be recommended, so that the information to be recommended is distributed to users who meet their interest requirements, thereby improving the effect of information recommendation.

[0054] See also Figure 1 , Figure 1It is a schematic diagram of an application scenario of the recommendation system 10 provided in an embodiment of the present application. The terminals (exemplarily showing the terminal 200-1, the terminal 200-2 and the terminal 200-3) are connected to the server 100 via the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0055] The terminal (running a client, such as a news client, a video client, etc.) can be used to obtain information recommendation requests for the target object. For example, when the target object opens the video client running on the terminal, the terminal automatically obtains a video recommendation request for the target object.

[0056] In some embodiments, after the terminal obtains an information recommendation request for the target object, the terminal calls the information recommendation interface of the server 100 (which can be provided in the form of a cloud service, i.e., an information recommendation service). Based on the information recommendation request for the target object, the server 100 performs layer construction processing based on the feature domain on multiple domain features through each layer constructor in the multi-layer constructor to obtain the cross-features of each layer constructor; performs weighted aggregation processing on the cross-features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended; performs indicator prediction processing on the aggregated features of the task to be recommended to obtain the recommendation index of the target object corresponding to the information to be recommended; performs recommendation operations based on the recommendation index of the target object corresponding to the information to be recommended, thereby distributing the information to be recommended to users who meet their interest requirements, thereby improving the effect of information recommendation.

[0057] As an application example, for video applications, when multiple target objects open the video client running on the terminal, the terminal automatically obtains a video recommendation request for the target object, calls the information recommendation interface of the server 100, and the server 100 executes an information recommendation method based on artificial intelligence to determine the recommendation indicators (such as whether to click) of the videos to be recommended corresponding to the multiple target objects. Then, based on the recommendation indicators of the videos to be recommended corresponding to the multiple target objects, the click-through rate of the video to be recommended is estimated. For example, among 100 target objects, 80 target objects may click on the video to be recommended, and the click-through rate of the video to be recommended is estimated to be 80%. The recommendation position of the video to be recommended is adjusted so that the video to be recommended is located in a position with better exposure, thereby improving the effect of information recommendation in response to the information recommendation request for the target object.

[0058] In some embodiments, an information recommendation plug-in may be implanted in the client running in the terminal to implement an information recommendation method based on artificial intelligence locally on the client. For example, after the terminal obtains an information recommendation request for a target object, the information recommendation plug-in is called to implement the information recommendation method, and multiple domain features are subjected to layer construction processing based on the feature domain through each layer constructor in the multi-layer constructor to obtain the cross-features of each layer constructor; weighted aggregation processing is performed on the cross-features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended; the index prediction processing is performed on the aggregated features of the task to be recommended to obtain the recommendation index of the target object corresponding to the information to be recommended; the recommendation operation is performed based on the recommendation index of the target object corresponding to the information to be recommended, so as to distribute the information to be recommended to users who meet their interests and requirements, thereby improving the effect of information recommendation.

[0059] As an application example, for a news application, when the target object opens the news client running on the terminal, the terminal automatically obtains a news recommendation request for the target object, calls an information recommendation plug-in, executes an information recommendation method based on multiple recommendation features of the task to be recommended, and determines the recommendation index (for example, whether to click) of the target object corresponding to the news to be recommended. When the recommendation index of the target object corresponding to the news to be recommended indicates that the target object will click on the news to be recommended, the task to be recommended is executed, thereby improving the effect of information recommendation in response to the information recommendation request for the target object.

[0060] In some embodiments, the terminal or server can implement the information recommendation method provided in the embodiments of the present application by running various computer executable instructions or computer programs. For example, computer executable instructions can be microprogram-level commands, machine instructions or software instructions. The computer program can be a native program or software module in the operating system; it can be a local (Native) application (APP, Application), that is, a program that needs to be installed in the operating system to run, such as a live broadcast application or an instant messaging application; it can also be a small program embedded in any APP, that is, a program that can be run only by downloading it to a browser environment. In short, the above-mentioned computer executable instructions can be instructions in any form, and the above-mentioned computer program can be an application, module or plug-in in any form.

[0061] In some embodiments, multiple servers may form a blockchain, and server 100 is a node on the blockchain. Information connections may exist between each node in the blockchain, and information may be transmitted between nodes through the above information connections. Among them, data related to the information recommendation method provided in the embodiment of the present application (such as the logic of information recommendation and recommendation indicators) may be stored on the blockchain.

[0062] The structure of the electronic device for information recommendation provided by the embodiment of the present application is described below. Figure 2 , Figure 2 is a schematic diagram of the structure of an electronic device 500 for information recommendation provided in an embodiment of the present application. Taking the electronic device 500 as a terminal as an example, Figure 2 The electronic device 500 for information recommendation shown includes: at least one processor 510, a memory 550, at least one network interface 520 and a user interface 530. The various components in the electronic device 500 are coupled together through a bus system 540. It can be understood that the bus system 540 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 540 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 540 is not shown in FIG. Figure 2 Various buses are labeled as bus system 540 .

[0063] The processor 510 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0064] The memory 550 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 550 described in the embodiment of the present application is intended to include any suitable type of memory. The memory 550 optionally includes one or more storage devices that are physically far away from the processor 510.

[0065] In some embodiments, the memory 550 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0066] Operating system 551, including system programs for processing various basic system services and performing hardware-related tasks, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0067] A network communication module 552, used to reach other electronic devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 include: Bluetooth, Wireless Compatibility Authentication (WiFi), and Universal Serial Bus (USB), etc.;

[0068] In some embodiments, the information recommendation device provided in the embodiments of the present application can be implemented in software. The information recommendation device provided in the embodiments of the present application can be provided as various software embodiments, including various forms including application programs, software, software modules, scripts or codes.

[0069] Figure 2 An information recommendation device 555 stored in a memory 550 is shown, which may be software in the form of a program or plug-in, and includes a series of modules, including an acquisition module 5551, a layer construction module 5552, an aggregation module 5553, a prediction module 5554, and a recommendation module 5555. These modules are logical, and therefore may be arbitrarily combined or further split according to the functions implemented. The functions of each module will be described below.

[0070] As mentioned above, the information recommendation method provided in the embodiment of the present application can be implemented by various types of electronic devices, such as a terminal, a server, or a combination of the two. Therefore, the execution subject of each step will not be repeatedly described below. Figure 3A , Figure 3A is a flowchart of the information recommendation method provided in the embodiment of the present application, combined with Figure 3A The steps shown are explained.

[0071] In the following steps, the information to be recommended can be data such as text, images, graphics, videos, etc. Recommendation indicators are indicators used to guide the recommendation system to make recommendations, for example, whether the target object will click on the information to be recommended, whether the target object is interested in the information to be recommended, whether the target object will convert based on the information to be recommended, whether the target object will evaluate the information to be recommended, etc.

[0072] In step 101, a plurality of domain features of a task to be recommended are obtained, wherein the plurality of domain features include at least one item feature of information to be recommended and at least one object feature of a target object.

[0073] For example, feature extraction is performed on the information to be recommended through offline or online means to obtain item features (such as advertising features) of the information to be recommended. For example, item features are used to characterize the features of the information to be recommended, such as item features including the category and price of the information to be recommended.

[0074] For example, feature extraction processing is performed on the object feature set of the target object through offline means or online means to obtain object features of the target object (such as user-side features). For example, the object features are used to characterize the features of the target object, such as the object features including the age, preference, gender, etc. of the target user.

[0075] In some embodiments, the multiple domain features also include contextual features of the task to be recommended, and the contextual features are used to characterize real-time features of the task to be recommended, such as the current time, information about the electronic device used by the target object, the geographic location of the target object, etc.

[0076] As an example of obtaining domain features, after the domain features of the task to be recommended are pre-extracted through offline means, after the target object opens the client running on the terminal, the terminal automatically obtains an information recommendation request for the target object (including the identifier of the target object), and based on the information recommendation request for the target object, obtains the pre-extracted domain features of the task to be recommended, so as to subsequently perform layer construction operations based on the domain features of the task to be recommended.

[0077] In step 102, each layer constructor in the multi-layer constructor performs layer construction processing based on the feature domain on multiple domain features to obtain cross features of each layer constructor.

[0078] Here, the domain features are interacted with by each layer of constructors to obtain the cross features of each layer of constructors (also called cross terms (i.e., terms processed by the feature cross function)), and then aggregation processing is performed based on the cross features of each layer of constructors, where the terms in the first layer of constructors refer to each domain feature, and the terms in other layer constructors refer to the terms constructed by the layer constructors.

[0079] It should be noted that the layer constructor can also be a global layer constructor. Through each layer of global layer constructor, multiple domain features can be globally constructed to obtain the cross features of each layer constructor, where the i-th layer global layer constructor (Layer Assembler with Global-wise Terms, AGT) contains a total of K items (K is equal to the number of domain features), where the k-th item in the i-th layer is equal to the result of summing up the cross features of all cross features of the first layer and each cross feature of the i-1-th layer, that is, the k-th item contains the i-order cross terms of all domains.

[0080] See also Figure 3B , Figure 3B is a flowchart of the information recommendation method provided in the embodiment of the present application, Figure 3B Show Figure 3AStep 102 in can be implemented through steps 1021-1022: in step 1021, the i-th layer constructor in the multi-layer constructor performs the following processing: determining the cross-features of the i-1-th layer constructor, wherein when the i-1-th layer constructor is the first layer constructor, the cross-features of the i-1-th layer constructor are multiple domain features; in step 1022, performing feature cross-feature processing on the multiple domain features and the cross-features of the i-1-th layer constructor to obtain the cross-features of the i-th layer constructor; wherein i is a successively increasing positive integer, 1<i≤I, and I is the number of layers of the multi-layer constructor.

[0081] As an example, the cross-features of the first-layer constructor are multiple domain features. Through the second-layer constructor, feature cross-processing is performed on the multiple domain features and the cross-features of the first-layer constructor to obtain the cross-features of the second-layer constructor; through the third-layer constructor, feature cross-processing is performed on the multiple domain features and the cross-features of the second-layer constructor to obtain the cross-features of the third-layer constructor, and so on, to determine the cross-features of each layer of constructor.

[0082] See also Figure 3C , Figure 3C is a flowchart of the information recommendation method provided in the embodiment of the present application, Figure 3C Show Figure 3B Step 1022 in can be implemented through steps 10221-10222: in step 10221, the following processing is performed on the j-th domain feature among the multiple domain features: feature cross processing is performed on the j-th domain feature and each cross feature in the cross features of the i-1-th layer constructor to obtain multiple cross sub-features; in step 10222, the sum of the multiple cross sub-features is used as the j-th cross feature of the i-th layer constructor; wherein j is a positive integer, 1≤j≤J, and J is the number of the multiple domain features.

[0083] Here, the layer constructor can be a layer constructor based on feature domain (Layer Assembler with Field-wise Terms, AFT). Through the layer constructor based on feature domain, multiple domain features can be processed by feature domain-based layer construction to obtain the cross features of each layer constructor, where the i-th layer AFT contains a total of J items (K is equal to the number of domain features), where the j-th item in the i-th layer (i.e., the j-th cross feature) is equal to the j-th item of the first layer AFT and each cross feature of the i-1-th layer after feature crossover and then the sum of the calculation results, that is, the j-th item contains all the i-order cross terms related to the j-th domain.

[0084] In some embodiments, step 10221 can be implemented in the following manner: perform the following processing on the kth cross feature in the cross features of the i-1th layer constructor: perform Hadamard product processing on the jth domain feature and the kth cross feature to obtain the kth cross sub-feature; wherein k is a positive integer, 1≤k≤J.

[0085] For example, the embodiment of the present application may use the feature crossover function of Hadamard product (Naive Hadamard Product, N-HP) to directly calculate the Hadamard product between two items (i.e., the jth domain feature and the kth crossover feature) to obtain the kth crossover sub-feature. Fig. 7A As shown, the embodiment of the present application can convert the calculation process of the Hadamard product into mapping two items (ie, the j-th domain feature and the k-th cross feature) through the N-HP matrix.

[0086] See also Figure 3D , Figure 3D is a flowchart of the information recommendation method provided in the embodiment of the present application, Figure 3D Show Figure 3C Step 10221 in the figure can be implemented through steps 102211-102212: in step 102211, the following processing is performed on the kth cross feature in the cross features of the i-1th layer constructor: the jth domain feature and the kth cross feature are processed with Hadamard product to obtain the kth Hadamard product result; in step 102212, the kth Hadamard product result is mapped to obtain the kth cross sub-feature; wherein k is a positive integer, 1≤k≤J.

[0087] In some embodiments, step 102212 can be implemented by: determining the domain pair scaling weight corresponding to each element in the kth Hadamard product result; based on the domain pair scaling weight, weighting each element in the kth Hadamard product result to obtain the kth cross sub-feature.

[0088] For example, the embodiment of the present application may use the pair-wise weighted scaled Hadamard product (W-HP) as a feature crossover function to calculate the Hadamard product between two items (i.e., the jth domain feature and the kth crossover feature), and then multiply each element in the calculation result of the Hadamard product (i.e., the kth Hadamard product result) by the same pair-wise scaling weight to obtain the kth crossover sub-feature. The pair-wise scaling weight means that each pair of domains (also called items) has a learnable weight coefficient to capture the importance of the domain pair. Figure 7B As shown, the embodiment of the present application can convert the calculation process of the domain-pair weighted scaled Hadamard product into mapping two items (ie, the j-th domain feature and the k-th cross feature) through the W-HP matrix.

[0089] In some embodiments, step 102212 can be implemented by: determining the domain pair scaling vector corresponding to the k-th Hadamard product result; and taking the product of the domain pair scaling vector and the k-th Hadamard product result as the k-th cross sub-feature.

[0090] For example, the embodiment of the present application may use the domain pair vector scaled Hadamard product (Pair-wise Vector scaled Hadamard product, V-HP) as a feature crossover function to calculate the Hadamard product between two items (i.e., the j-th domain feature and the k-th crossover feature), and then multiply the calculation result of the Hadamard product (i.e., the k-th Hadamard product result) by a domain pair scaling vector to obtain the k-th crossover sub-feature. The domain pair scaling vector means that each pair of domains has a learnable vector (with a dimension of K, where K represents the dimension of the embedding vector of the domain feature), which is used to convert the two items when they interact. Figure 7C As shown, the embodiment of the present application can convert the calculation process of the domain pair vector scaled Hadamard product into mapping two items (ie, the j-th domain feature and the k-th cross feature) through the V-HP matrix.

[0091] In some embodiments, step 10221 can be implemented in the following manner: perform the following processing on the kth cross feature in the cross features of the i-1th layer constructor: determine the domain pair projection matrix corresponding to the jth domain feature; based on the domain pair projection matrix, perform matrix transformation processing on the jth domain feature to obtain the transformed jth domain feature; perform Hadamard product processing on the transformed jth domain feature and the kth cross feature to obtain the kth cross sub-feature; wherein k is a positive integer, 1≤k≤J.

[0092] For example, the embodiment of the present application may use the pair-wise matrix projected Hadamard product (M-HP) as a feature crossover function to first transform one of the two items in the feature crossover (i.e., the j-th domain feature and the k-th crossover feature) through a pair-wise projection matrix, and then calculate the Hadamard product between the transformed one item and the other item to obtain the k-th crossover sub-feature. The pair-wise projection matrix means that the two items have a learnable matrix (with a dimension of K*K, where K represents the dimension of the embedding vector of the domain feature), which is used to convert the feature embeddings of the two items when they interact. Fig.7D As shown, the embodiment of the present application can convert the calculation process of the domain-to-matrix projection Hadamard product into mapping two items (ie, the j-th domain feature and the k-th cross feature) through the M-HP matrix.

[0093] Following step 102, in step 103, weighted aggregation processing is performed on the cross features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended.

[0094] Here, through the layer aggregator, the cross features corresponding to the multiple layer constructors are weightedly aggregated to obtain the aggregated features of the task to be recommended, so that the aggregated features can fully and effectively aggregate the domain features, so that the index prediction can be performed based on the accurate aggregated features to obtain accurate recommendation indexes.

[0095] In some embodiments, step 103 can be implemented in the following ways: determining the hierarchical weight of each layer constructor; based on the hierarchical weight of each layer constructor, weighted processing is performed on the cross features corresponding to the multiple layers of layer constructors to obtain the weighted cross features corresponding to the multiple layers of layer constructors; and concatenation processing is performed on the weighted cross features corresponding to the multiple layers of layer constructors to obtain the aggregated features of the task to be recommended.

[0096] For example, the embodiment of the present application may use an order-wise weighted layer aggregator (Aggregator with Order-wise weight, Agg-O) to multiply the cross features output by the layer constructors of all layers by a weight related to the order (i.e., the layer weight) to obtain weighted cross features corresponding to multiple layers of layer constructors, and then concatenate the weighted cross features corresponding to multiple layers of layer constructors to obtain the aggregated features (also called representation) of the task to be recommended. The weight related to the order is a learnable parameter. Fig. 9A As shown, the layer aggregator based on order weight in the embodiment of the present application multiplies the embedding vectors of the cross features output by the layer constructor by a weight related to the order.

[0097] In some embodiments, step 103 can be implemented in the following ways: determining the item weight of each cross-feature of each layer constructor; based on the item weight of each cross-feature of each layer constructor, weighting each cross-feature of each layer constructor to obtain the weighted cross-feature of each layer constructor; splicing the weighted cross-features corresponding to multiple layers of layer constructors to obtain the aggregated features of the task to be recommended.

[0098] For example, the embodiment of the present application may use a layer aggregator based on term-wiseweight (Agg-T), which multiplies each item in the cross-features output by the layer constructors of all layers by a weight about the item (i.e., the item weight), obtains the weighted cross-features of each layer constructor, and then concatenates the weighted cross-features corresponding to multiple layers of layer constructors to obtain the aggregated features (also called representation) of the task to be recommended. The weights about the items are learnable parameters. Fig. 9B As shown, the item weight-based layer aggregator in the embodiment of the present application multiplies each item in the embedding vector of the cross-feature output by the layer constructor by a weight related to the item.

[0099] In some embodiments, step 103 can be implemented in the following ways: determining the element weight of each element in each cross-feature of each layer constructor; based on the element weight of each element in each cross-feature of each layer constructor, weighting each element in each cross-feature of each layer constructor to obtain the weighted cross-feature of each layer constructor; splicing the weighted cross-features corresponding to multiple layers of layer constructors to obtain the aggregated features of the task to be recommended.

[0100] For example, the embodiment of the present application may use an element-wise weighted layer aggregator (Agg-E), which multiplies each element of each item in the cross-features output by the layer constructors of all layers by a weight about the element (i.e., the element weight), obtains the weighted cross-features of each layer constructor, and then concatenates the weighted cross-features corresponding to multiple layers of layer constructors to obtain the aggregated features (also called representation) of the task to be recommended. The weights about the elements are learnable parameters. Fig. 9C As shown, the element-weighted layer aggregator in the embodiment of the present application multiplies each element in the embedding vector of the cross-feature output by the layer constructor by a weight related to the element.

[0101] In step 104, the aggregated features of the task to be recommended are processed for index prediction to obtain a recommendation index for the target object corresponding to the information to be recommended.

[0102] For example, after obtaining the aggregate features of the task to be recommended, the classifier is used to perform indicator prediction processing on the aggregate features of the task to be recommended to obtain the recommendation indicator of the target object corresponding to the information to be recommended, such as whether the target object will click on the information to be recommended, whether the target object is interested in the information to be recommended, whether the target object will convert based on the information to be recommended, whether the target object will evaluate the information to be recommended, etc., thereby executing the task to be recommended based on the recommendation indicator of the target object corresponding to the information to be recommended, thereby improving the accuracy of information recommendation.

[0103] In step 105, a recommendation operation is performed based on the recommendation index of the target object corresponding to the information to be recommended.

[0104] For example, after obtaining the recommendation index of the target object corresponding to the information to be recommended, the recommended task is performed based on the recommendation index of the target object corresponding to the information to be recommended. For example, when the recommendation index of the target object corresponding to the information to be recommended indicates that the target object may click on the information to be recommended, the recommendation operation is performed to display the information to be recommended to the target object, thereby improving the effect of information recommendation.

[0105] The following is an explanation of an exemplary application of the embodiments of the present application in a practical application scenario.

[0106] The embodiments of the present application can be applied to personalized recommendations including e-commerce shopping, video (or music) recommendations, news information flow recommendations, life service scenarios, etc. Among them, online advertising is the most direct and transparent way to realize traffic for most Internet companies. Take Moments advertising as an example. Figure 4 As shown in the figure, when a user opens the Moments and refreshes the list, the advertising system selects suitable ads from the advertising library for recommendation and display. After the ad is displayed to the user, if the user clicks on the ad or even activates the app, places an order, etc., the system will automatically deduct fees from the advertiser, thereby realizing traffic monetization.

[0107] Accurate advertising recommendations can help advertisers tap into potential users faster, thereby improving the efficiency of advertising delivery, maximizing resource utilization, and improving user experience, achieving a win-win situation for advertisers, advertising platforms, and users. In this scenario, an embodiment of the present application proposes a unified explicit high-order feature crossover model (referred to as the model), the representation part of which includes three components, namely a feature crossover function, a layer constructor, and a layer aggregator, and then the output of the representation part is passed through a classifier to obtain a prediction result. The model mainly uses a feature crossover method to model the relationship between features in different domains on the user side, the advertising side, and the context side. Using a more efficient explicit high-order feature crossover model can control the spatial and temporal complexity of the model, while improving the prediction accuracy of the model, thereby improving the effectiveness of advertisers.

[0108] like Figure 5As shown in the figure, in the online advertising business scenario, the advertising platform receives the advertising request from the content server, and then selects the most suitable advertisement from the advertising library and returns it for display. All the advertising exposure, click and conversion data generated in this process will be stored and used for scheduled model training. Each retrained model will be loaded into the advertising platform server for real-time advertising click-through rate and conversion rate prediction.

[0109] The construction of training samples and features is described below.

[0110] For the click-through rate (CTR) model, the CTR model predicts the probability of a user clicking on an ad after it is displayed. Therefore, the training sample for training the CTR model is the user's single ad exposure record, and the label is determined by whether the user clicks on the ad. The clicked ad sample is marked as a positive sample (y=1), and the unclicked ad sample is marked as a negative sample (y=0).

[0111] For the conversion rate (CVR) model, the CVR model predicts the probability that a user will convert after clicking an ad. Therefore, the training sample for training the CVR model is a single ad click record of the user, and the label is determined by whether the user converts. For each conversion of the user, if a click operation within a given window before this conversion can be found, then if a click operation can be found, the ad sample is marked as a positive sample, otherwise the ad sample is marked as a negative sample.

[0112] It should be noted that the features of each sample, including user-side features, advertising-side features, and contextual features, are all discrete. For features that were originally continuous values, they are discretized into discrete features.

[0113] like Figure 6 As shown, the use of an explicit high-order feature cross model to predict advertising click-through rate is used as an example. The explicit high-order feature cross model includes 1) a feature cross function, 2) a layer constructor, and 3) a layer aggregator. The feature cross function, layer constructor, and layer aggregator are described below respectively.

[0114] It should be noted that before being processed by the feature crossover function and the layer constructor, the user-side features, the advertising-side features and the context features can be subjected to sparse feature preprocessing to convert the user-side features, the advertising-side features and the context features into an embedding vector of a fixed dimension to obtain sparse features.

[0115] 1) Feature crossover function: For each item in a layer (the items in the first layer refer to the embedding vector of each feature, and the items in other layers refer to the items constructed by the layer constructor), a feature crossover function is used to calculate the feature crossover between the two items. The feature crossover functions provided in the embodiments of the present application include the following four types, namely:

[0116] a. Naive Hadamard Product (N-HP): Directly calculate the Hadamard product between two terms.

[0117] like Fig. 7A As shown, the embodiment of the present application can convert the calculation process of the Hadamard product into mapping two items (item A and item B) through the N-HP matrix.

[0118] b. Pair-wise Weight Scaled Hadamard Product (W-HP): After calculating the Hadamard product between two terms, each element in the calculation result of the Hadamard product is multiplied by the same pair-wise scaling weight. The pair-wise scaling weight means that each pair of domains (also known as terms) has a learnable weight coefficient to capture the importance of the domain pair.

[0119] like Figure 7B As shown, the embodiment of the present application can convert the calculation process of the domain-pair weighted scaled Hadamard product into mapping two items (item A and item B) through the W-HP matrix.

[0120] c. Pair-wise Vector scaled Hadamard product (V-HP): After calculating the Hadamard product between two items, multiply the result of the Hadamard product by a pair-wise scaled vector. The pair-wise scaled vector means that each pair of domains has a learnable vector (with a dimension of K, where K represents the dimension of the embedding vector) that is used to transform the feature embeddings of a pair of domains when they interact.

[0121] like Figure 7C As shown, the embodiment of the present application can convert the calculation process of the domain-pair vector scaled Hadamard product into mapping two items (item A and item B) through the V-HP matrix.

[0122] d. Pair-wise Matrix projected Hadamard product (M-HP): After one of the two items in the feature crossover is transformed by a pair-wise projection matrix, the Hadamard product between the transformed one and the other is calculated. The pair-wise projection matrix means that each pair of domains has a learnable matrix (with a dimension of K*K, where K represents the dimension of the embedding vector), which is used to transform the feature embeddings of a pair of domains when they interact.

[0123] like Fig.7D As shown, the embodiment of the present application can convert the calculation process of the domain pair matrix projection Hadamard product into mapping two items (A item and B item) through the M-HP matrix.

[0124] 2) Layer constructor: Use the cross terms in the previous layer (i.e., the terms processed by the feature cross function) and the feature embedding of the first layer to construct the cross terms in the next layer. The layer constructor provided in the embodiment of the present application includes the following:

[0125] a. Layer Assembler with Field-wise Terms (AFT): The i-th layer contains K items (K is equal to the number of fields), among which the k-th item in the i-th layer is equal to the embedding of the k-th field in the first layer and the sum of the feature crossover of each item in the i-1-th layer, that is, the k-th item contains all the i-order cross terms related to the k-th field.

[0126] like Figure 8 As shown, when the i-th layer contains 3 items (i.e., v 1 、v 2 、v 3 ), then the first item in the i-th layer is equal to the embedding of the first domain of the first layer and the sum of the features of each item in the i-1-th layer after cross-feature calculation.

[0127] 3) Layer aggregator: The layer aggregator takes the embedding vectors output by the layer constructors of all layers as input, aggregates the embedding vectors output by the layer constructors of all layers into a representation, and uses the representation as the input of the classifier. The layer constructors provided in the embodiments of the present application include the following three types, namely:

[0128] a. Aggregator with Order-wise weight (Agg-O): The embedding vectors output by the layer constructors of all layers are multiplied by a weight related to the order, and then concatenated to obtain the representation.

[0129] like Fig. 9A As shown, the order-weighted layer aggregator in the embodiment of the present application multiplies the embedding vectors output by the layer constructor by a weight related to the order.

[0130] b. Aggregator with Term-wise weight (Agg-T): Multiply each item in the embedding vector output by the layer constructors of all layers by a weight related to the item and then concatenate them to get the representation.

[0131] like Fig. 9B As shown, the item weight-based layer aggregator in the embodiment of the present application multiplies each item in the embedding vector output by the layer constructor by a weight related to the item.

[0132] c. Aggregator with Element-wise weight (Agg-E): Multiply each element in the embedding vector output by the layer constructors of all layers by a weight related to the element and then concatenate them to get the representation.

[0133] like Fig. 9C As shown, the element-weighted layer aggregator in the embodiment of the present application multiplies each element in the embedding vector output by the layer constructor by a weight related to the element.

[0134] Finally, the CTR model constructs a multi-layer fully connected deep neural network as a classifier, and uses the representation output by the layer aggregator as the input of the classifier. The classifier obtains the user's CTR estimate for the advertisement through the softmax function.

[0135] It should be noted that the model provided in the embodiment of the present application is an end-to-end model, and all parameters of the model can be updated using a gradient algorithm. Therefore, the method provided in the embodiment of the present application is applicable to any deep learning algorithm framework.

[0136] As an example of online application, a new model is trained every hour and pushed online for online prediction. The specific online prediction process is as follows: Step 1: The requester sends an ad request. After the recall and coarse ranking models preliminarily screen the ads, an ad set is returned to the fine ranking system. Step 2: The fine ranking system queries the user-side features and ad-side features, and after preprocessing, the features are input. Figure 6 The network structure shown is used to calculate the estimated CTR (pCTR) / estimated CVR (pCVR); Step 3, calculate the effective cost per thousand impressions (eCPM) using the pCTR / pCVR calculated in Step 2, sort all the ads in the ad set, and select the top K ads for exposure.

[0137] In order to verify the effectiveness of this solution, the embodiment of the present application conducted some offline experiments on two public datasets, Criteo and Avazu, and one synthetic dataset, as shown in Table 1.

[0138] Table 1

[0139]

[0140]

[0141] As can be seen from Table 1, the area (AUC) and logarithmic function (Logloss) under the ROC curve on the two public data sets prove that the model provided by the embodiment of the present application has achieved good results. When the cross function and layer constructor are kept the same, different layer aggregators achieve similar results, and the layer aggregator based on the order weight is relatively good.

[0142] The embodiment of the present application plots the weights and feature cross strengths learned by different feature cross functions on the Avazu dataset and compares them with the mutual information between domain pairs. It is found that as the complexity of the feature cross function increases and the number of learnable parameters increases, the learned weights and feature cross strengths are closer to the mutual information.

[0143] The embodiments of the present application respectively generate synthetic data sets with data order (L) equal to 4 and 5, and compare the sensitivity of the model provided by the embodiments of the present application to the model order. It is found that when the model order gradually increases and is greater than the data order, the mean square error (RMSE) of the model provided by the embodiments of the present application remains very stable, proving that the layer aggregator based on order weights can capture the importance of data order.

[0144] So far, the information recommendation method provided by the embodiment of the present application has been described in combination with the exemplary application and implementation of the electronic device provided by the embodiment of the present application. The following will continue to describe the cooperation of various modules in the information recommendation device 555 provided by the embodiment of the present application to implement the information recommendation solution.

[0145] An acquisition module 5551 is used to acquire multiple domain features of the task to be recommended, wherein the multiple domain features include at least one item feature of the information to be recommended and at least one object feature of the target object; a layer construction module 5552 is used to perform layer construction processing based on the feature domain on the multiple domain features through each layer constructor in the multi-layer constructor to obtain the cross-features of each layer constructor; an aggregation module 5553 is used to perform weighted aggregation processing on the cross-features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended; a prediction module 5554 is used to perform indicator prediction processing on the aggregated features of the task to be recommended to obtain the recommendation indicator of the target object corresponding to the information to be recommended; a recommendation module 5555 is used to perform a recommendation operation based on the recommendation indicator of the target object corresponding to the information to be recommended.

[0146] In some embodiments, the layer construction module 5552 is also used to perform the following processing through the i-th layer constructor in the multi-layer constructor: determine the cross-features of the i-1-th layer constructor, wherein, when the i-1-th layer constructor is the first layer constructor, the cross-features of the i-1-th layer constructor are the multiple domain features; perform feature cross-processing on the multiple domain features and the cross-features of the i-1-th layer constructor to obtain the cross-features of the i-th layer constructor; wherein i is a successively increasing positive integer, 1<i≤I, and I is the number of layers of the multi-layer constructor.

[0147] In some embodiments, the layer construction module 5552 is also used to perform the following processing on the j-th domain feature among the multiple domain features: perform feature cross-processing on the j-th domain feature and each cross-feature in the cross-features of the i-1-th layer constructor to obtain multiple cross-sub-features; and use the sum of the multiple cross-sub-features as the j-th cross-feature of the i-th layer constructor; wherein j is a positive integer, 1≤j≤J, and J is the number of the multiple domain features.

[0148] In some embodiments, the layer construction module 5552 is also used to perform the following processing on the kth cross feature in the cross features of the i-1th layer constructor: perform Hadamard product processing on the jth domain feature and the kth cross feature to obtain the kth cross sub-feature; wherein k is a positive integer, 1≤k≤J.

[0149] In some embodiments, the layer construction module 5552 is also used to perform the following processing on the kth cross feature in the cross features of the i-1th layer constructor: perform Hadamard product processing on the jth domain feature and the kth cross feature to obtain the kth Hadamard product result; perform mapping processing on the kth Hadamard product result to obtain the kth cross sub-feature; wherein k is a positive integer, 1≤k≤J.

[0150] In some embodiments, the layer construction module 5552 is also used to determine the domain pair scaling weight corresponding to each element in the kth Hadamard product result; based on the domain pair scaling weight, weighted processing is performed on each element in the kth Hadamard product result to obtain the kth cross sub-feature.

[0151] In some embodiments, the layer construction module 5552 is further used to determine a domain pair scaling vector corresponding to the k-th Hadamard product result; and use the product of the domain pair scaling vector and the k-th Hadamard product result as the k-th cross sub-feature.

[0152] In some embodiments, the layer construction module 5552 is also used to perform the following processing on the kth cross feature in the cross features of the i-1th layer constructor: determine the domain pair projection matrix corresponding to the jth domain feature; based on the domain pair projection matrix, perform matrix transformation processing on the jth domain feature to obtain the transformed jth domain feature; perform Hadamard product processing on the transformed jth domain feature and the kth cross feature to obtain the kth cross sub-feature; wherein k is a positive integer, 1≤k≤J.

[0153] In some embodiments, the aggregation module 5553 is also used to determine the hierarchical weight of each of the layer constructors; based on the hierarchical weight of each of the layer constructors, weighted processing is performed on the cross features corresponding to the multi-layer constructors to obtain the weighted cross features corresponding to the multi-layer constructors; the weighted cross features corresponding to the multi-layer constructors are concatenated to obtain the aggregated features of the task to be recommended.

[0154] In some embodiments, the aggregation module 5553 is also used to determine the item weight of each cross-feature of each of the layer constructors; based on the item weight of each cross-feature of each of the layer constructors, weighted processing is performed on each cross-feature of each of the layer constructors to obtain the weighted cross-feature of each of the layer constructors; and the weighted cross-features corresponding to the multiple layers of layer constructors are concatenated to obtain the aggregated features of the task to be recommended.

[0155] In some embodiments, the aggregation module 5553 is also used to determine the element weight of each element in each cross-feature of each of the layer constructors; based on the element weight of each element in each cross-feature of each of the layer constructors, weighted processing is performed on each element in each cross-feature of each of the layer constructors to obtain the weighted cross-feature of each of the layer constructors; and the weighted cross-features corresponding to the multiple layers of layer constructors are concatenated to obtain the aggregated features of the task to be recommended.

[0156] The embodiment of the present application provides a computer program product, which includes a computer program or a computer executable instruction, and the computer program or the computer executable instruction is stored in a computer-readable storage medium. The processor of the electronic device reads the computer program or the computer executable instruction from the computer-readable storage medium, and the processor executes the computer program or the computer executable instruction, so that the electronic device performs the information recommendation method described in the embodiment of the present application.

[0157] The present application embodiment provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions or computer programs are stored. When the computer-executable instructions or computer programs are executed by a processor, the processor will be caused to execute the information recommendation method provided by the present application embodiment, for example, Figure 3A The information recommendation method is shown.

[0158] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0159] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), 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.

[0160] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing 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 storing one or more modules, subroutines, or code portions).

[0161] As an example, computer executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0162] It is understandable that in the embodiments of the present application, related data such as user information is involved. When the embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0163] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. An information recommendation method, characterized in that: The method comprises: Acquire multiple domain features of the task to be recommended, where the multiple domain features include at least one item feature of the information to be recommended and at least one object feature of the target object; By using each layer constructor in the multi-layer constructor, the plurality of domain features are subjected to layer construction processing based on the feature domain, so as to obtain cross features of each layer constructor; Performing weighted aggregation processing on the cross features corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended; Performing index prediction processing on the aggregated features of the task to be recommended to obtain a recommendation index corresponding to the target object of the information to be recommended; A recommendation operation is performed based on the recommendation index of the target object corresponding to the information to be recommended.

2. The method according to claim 1, characterized in that: The method of performing layer construction processing based on feature domains on the multiple domain features by each layer constructor in the multi-layer constructor to obtain cross features of each layer constructor includes: The following processing is performed by the i-th layer constructor in the multi-layer constructor: Determine a cross feature of an i-1th layer constructor, wherein when the i-1th layer constructor is a first layer constructor, the cross feature of the i-1th layer constructor is the plurality of domain features; Performing feature cross processing on the multiple domain features and the cross features of the i-1th layer constructor to obtain the cross features of the i-th layer constructor; Wherein, i is a positive integer that increases successively, 1<i≤I, and I is the number of layers of the multi-layer constructor.

3. The method according to claim 2, characterized in that The performing feature cross processing on the multiple domain features and the cross features of the i-1th layer constructor to obtain the cross features of the i-th layer constructor includes: The following processing is performed for the j-th domain feature among the multiple domain features: Perform feature cross processing on the j-th domain feature and each cross feature in the cross features of the i-1-th layer constructor to obtain a plurality of cross sub-features; The sum of the multiple cross sub-features is used as the j-th cross feature of the i-th layer constructor; Wherein, j is a positive integer, 1≤j≤J, and J is the number of the multiple domain features.

4. The method according to claim 3, characterized in that: The j-th domain feature is subjected to feature cross processing with each cross feature of the i-1-th layer constructor to obtain a plurality of cross sub-features, including: The following processing is performed on the kth cross feature in the cross features of the i-1th layer constructor: Performing Hadamard product processing on the j-th domain feature and the k-th cross feature to obtain the k-th cross sub-feature; Wherein, k is a positive integer, 1≤k≤J.

5. The method according to claim 3, characterized in that: The j-th domain feature is subjected to feature cross processing with each cross feature of the i-1-th layer constructor to obtain a plurality of cross sub-features, including: The following processing is performed on the kth cross feature in the cross features of the i-1th layer constructor: Performing Hadamard product processing on the j-th domain feature and the k-th cross feature to obtain a k-th Hadamard product result; Performing mapping processing on the k-th Hadamard product result to obtain the k-th cross sub-feature; Wherein, k is a positive integer, 1≤k≤J.

6. The method according to claim 5, characterized in that The mapping process is performed on the k-th Hadamard product result to obtain the k-th intersection sub-feature, including: Determine a domain pair scaling weight corresponding to each element in the k-th Hadamard product result; Based on the domain pair scaling weights, weighted processing is performed on each element in the k-th Hadamard product result to obtain the k-th cross sub-feature.

7. The method according to claim 5, characterized in that The mapping process is performed on the k-th Hadamard product result to obtain the k-th intersection sub-feature, including: Determine a domain pair scaling vector corresponding to the k-th Hadamard product result; The product of the domain pair scaling vector and the k-th Hadamard product result is used as the k-th cross sub-feature.

8. The method according to claim 3, characterized in that The j-th domain feature is subjected to feature cross processing with each cross feature of the i-1-th layer constructor to obtain a plurality of cross sub-features, including: The following processing is performed on the kth cross feature in the cross features of the i-1th layer constructor: Determine a domain pair projection matrix corresponding to the j-th domain feature; Based on the domain pair projection matrix, performing matrix transformation processing on the j-th domain feature to obtain a transformed j-th domain feature; Performing Hadamard product processing on the transformed j-th domain feature and the k-th cross feature to obtain the k-th cross sub-feature; Wherein, k is a positive integer, 1≤k≤J.

9. The method according to claim 1, characterized in that: The step of performing weighted aggregation processing on the cross features respectively corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended includes: determining a level weight for each of said level constructors; Based on the level weight of each of the layer constructors, weighted processing is performed on the cross features respectively corresponding to the multiple layer constructors to obtain weighted cross features respectively corresponding to the multiple layer constructors; The weighted cross features corresponding to the multi-layer constructors are concatenated to obtain the aggregated features of the task to be recommended.

10. The method according to claim 1, characterized in that The step of performing weighted aggregation processing on the cross features respectively corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended includes: determining a term weight for each cross feature of each of said layer constructors; Based on the item weight of each cross feature of each layer constructor, weighting processing is performed on each cross feature of each layer constructor to obtain a weighted cross feature of each layer constructor; The weighted cross features corresponding to the multi-layer constructors are concatenated to obtain the aggregated features of the task to be recommended.

11. The method according to claim 1, characterized in that The step of performing weighted aggregation processing on the cross features respectively corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended includes: determining an element weight for each element in each intersection feature of each of said layer constructors; Based on the element weight of each element in each cross feature of each layer constructor, weighting processing is performed on each element in each cross feature of each layer constructor to obtain a weighted cross feature of each layer constructor; The weighted cross features corresponding to the multi-layer constructors are concatenated to obtain the aggregated features of the task to be recommended.

12. An information recommendation device, characterized in that: The device comprises: An acquisition module, configured to acquire a plurality of domain features of the task to be recommended, wherein the plurality of domain features include at least one item feature of the information to be recommended and at least one object feature of the target object; A layer construction module, used for performing layer construction processing based on feature domains on the plurality of domain features through each layer constructor in the plurality of layer constructors, so as to obtain cross features of each layer constructor; An aggregation module, used for performing weighted aggregation processing on the cross features respectively corresponding to the multi-layer constructors to obtain the aggregated features of the task to be recommended; A prediction module, used to perform index prediction processing on the aggregated features of the task to be recommended, to obtain a recommendation index corresponding to the target object of the information to be recommended; The recommendation module is used to perform a recommendation operation based on the recommendation index of the target object corresponding to the information to be recommended.

13. An electronic device, characterized in that: The electronic device comprises: Memory for storing computer programs or computer executable instructions; A processor, configured to implement the information recommendation method according to any one of claims 1 to 11 when executing the computer program or computer executable instructions stored in the memory.

14. A computer-readable storage medium, characterized in that: A computer program or a computer executable instruction is stored, and when the computer program or the computer executable instruction is executed by a processor, the information recommendation method according to any one of claims 1 to 11 is implemented.

15. A computer program product comprising a computer program or computer executable instructions, characterized in that: When the computer program or computer executable instructions are executed by a processor, the information recommendation method according to any one of claims 1 to 11 is implemented.

Images