Information recommendation methods, devices, equipment, and storage media based on artificial intelligence

Abstract

This application provides an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence. The method includes: acquiring multiple recommendation features for a task to be recommended, the multiple recommendation features including at least one item feature of the information to be recommended and at least one user feature of the target user; performing feature projection processing on the multiple recommendation features based on the projection space corresponding to each recommendation feature to obtain projected features of the projection space; performing feature interaction processing based on each recommendation feature and the projected features of the corresponding projection space to obtain a feature domain corresponding to each recommendation feature; performing index prediction processing based on the feature domains corresponding to the multiple recommendation features to obtain a recommendation index for the target user corresponding to the information to be recommended; and executing the recommendation task based on the recommendation index for the target user corresponding to the information to be recommended. This application can improve the accuracy of content or information recommendation.

Description

Technical Field

[0001] This application relates to artificial intelligence technology, and more particularly to an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence. Background Technology

[0002] Artificial Intelligence (AI) is a comprehensive technology within computer science that studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. AI technology is a multidisciplinary field, encompassing a wide range of areas, including natural language processing and machine learning / deep learning. With technological advancements, AI will be applied in more fields and play an increasingly important role.

[0003] Recommender systems are one of the important applications of artificial intelligence. They can help users discover information that may be of interest to them in an information-overloaded environment and push the information to users who are interested in it.

[0004] Although recommendation systems in related technologies can identify information that a user might be interested in from a large amount of information based on recommendation metrics, and then recommend such information to the user, the accuracy of these recommendation metrics needs improvement. Summary of the Invention

[0005] This application provides an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence, which can improve the accuracy of information recommendation.

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

[0007] This application provides an information recommendation method based on artificial intelligence, including:

[0008] Obtain multiple recommendation features for the task to be recommended, wherein the multiple recommendation features include at least one item feature of the information to be recommended and at least one user feature of the target user;

[0009] Based on the projection space corresponding to each of the recommended features, feature projection processing is performed on the multiple recommended features to obtain the projected features of the projection space;

[0010] Based on each of the recommended features and the corresponding projection features of the projection space, feature interaction processing is performed to obtain the feature domain corresponding to each of the recommended features;

[0011] Based on the feature domains corresponding to the multiple recommendation features, index prediction processing is performed to obtain the recommendation index of the target user corresponding to the information to be recommended.

[0012] The recommendation task is executed based on the recommendation metrics of the target user corresponding to the information to be recommended.

[0013] In the above technical solution, the indicator prediction processing is achieved through a first prediction model and a second prediction model;

[0014] The step of performing index prediction processing based on the feature domains corresponding to the multiple recommendation features to obtain the recommendation index for the target user corresponding to the information to be recommended includes:

[0015] The recommendation index of the first prediction model is obtained by performing index prediction processing on the feature domains corresponding to the multiple recommendation features through the first prediction model.

[0016] The recommendation index of the second prediction model is obtained by performing index prediction processing on multiple recommendation features of the task to be recommended through the second prediction model.

[0017] Based on the recommendation metrics of the first prediction model and the recommendation metrics of the second prediction model, the recommendation metrics for the target user corresponding to the information to be recommended are obtained.

[0018] In the above technical solution, obtaining the recommendation index for the target user corresponding to the information to be recommended based on the recommendation index of the first prediction model and the recommendation index of the second prediction model includes:

[0019] The recommendation indicators of the first prediction model and the recommendation indicators of the second prediction model are summed to obtain the summed recommendation indicators.

[0020] The summed recommendation metrics are normalized to obtain the recommendation metrics for the target user corresponding to the information to be recommended.

[0021] This application provides an information recommendation device based on artificial intelligence, including:

[0022] The acquisition module is used to acquire multiple recommendation features of the task to be recommended, wherein the multiple recommendation features include at least one item feature of the information to be recommended and at least one user feature of the target user;

[0023] The feature interaction module is used to perform feature projection processing on the multiple recommended features based on the projection space corresponding to each of the recommended features, so as to obtain the projected features of the projection space.

[0024] Based on each of the recommended features and the corresponding projection features of the projection space, feature interaction processing is performed to obtain the feature domain corresponding to each of the recommended features;

[0025] The prediction module is used to perform indicator prediction processing based on the feature domains corresponding to the multiple recommendation features to obtain the recommendation indicators of the target user corresponding to the information to be recommended.

[0026] The recommendation module is used to execute the recommendation task based on the recommendation index of the target user corresponding to the information to be recommended.

[0027] In the above technical solution, the feature interaction module is further used to perform nonlinear feature extraction processing on the multiple recommendation features based on the weight of the projection space corresponding to each recommendation feature, so as to obtain the extracted features of the projection space.

[0028] The extracted features of the projection space are multiplied with the projection matrix of the projection space to obtain the projection features of the projection space.

[0029] In the above technical solution, the feature interaction module is further used to scale the plurality of recommendation features based on the weight of the projection space corresponding to each recommendation feature, so as to obtain the scaled plurality of recommendation features;

[0030] The scaled multiple recommendation features are then concatenated to obtain the extracted features of the projection space.

[0031] In the above technical solution, the feature interaction module is further used to multiply each of the recommended features and the corresponding projection features of the projection space to obtain the feature vector of each of the recommended features;

[0032] The feature vector of each recommended feature is mapped to obtain the feature domain corresponding to each recommended feature.

[0033] In the above technical solution, the projection space corresponding to each of the recommended features includes multiple channels; the feature interaction module is further used to perform feature projection processing on the multiple recommended features based on any one of the channels of the projection space corresponding to each of the recommended features, to obtain the projection feature of any one of the channels;

[0034] Based on feature interaction processing of each of the recommended features and the projection features corresponding to any of the channels, the feature domain corresponding to any of the channels is obtained;

[0035] The feature domains corresponding to multiple channels in the projection space are fused to obtain the feature domain corresponding to each recommended feature.

[0036] In the above technical solution, the feature interaction module is further configured to sum the feature domains corresponding to the multiple channels of the projection space, and use the result of the summation as the feature domain corresponding to each of the recommended features; or,

[0037] The feature domains corresponding to the multiple channels of the projection space are averaged, and the result of the averaging is used as the feature domain corresponding to each of the recommended features.

[0038] In the above technical solution, the feature interaction network for feature interaction includes multiple cascaded feature interaction layers; the feature interaction module is further configured to perform the following processing through any of the feature interaction layers in the feature interaction network:

[0039] Based on the projection space corresponding to each of the recommended features, feature projection processing is performed on multiple recommended features input to any of the feature interaction layers to obtain the projected features of any of the feature interaction layers.

[0040] Among them, the multiple recommendation features input to the first feature interaction layer are the multiple recommendation features included in the task to be recommended, the multiple recommendation features input to the subsequent feature interaction layer are the multiple recommendation features output by the previous feature interaction layer of the subsequent feature interaction layer, and the subsequent feature interaction layer is the feature interaction layer other than the first feature interaction layer among the multiple cascaded feature interaction layers.

[0041] The following processing is performed through any of the feature interaction layers in the feature interaction network:

[0042] Based on the multiple recommended features input to any of the feature interaction layers and the corresponding projected features of any of the feature interaction layers, feature interaction processing is performed to obtain the feature domain of any of the feature interaction layers, and the multiple recommended features included in the feature domain are output.

[0043] The feature domain of the last feature interaction layer is determined to be the feature domain corresponding to each of the recommended features.

[0044] In the above technical solution, the feature interaction module is further used to perform feature interaction processing based on the multiple recommended features input to any feature interaction layer and the projected features corresponding to any feature interaction layer, to obtain the latent features of any feature interaction layer.

[0045] The latent features of any of the feature interaction layers are summed with the plurality of recommended features input to any of the feature interaction layers to obtain the feature domain of any of the feature interaction layers.

[0046] In the above technical solution, the indicator prediction processing is implemented through a first prediction model and a second prediction model; the prediction module is also used to perform indicator prediction processing on the feature domains corresponding to the multiple recommendation features through the first prediction model to obtain the recommendation indicators of the first prediction model.

[0047] The recommendation index of the second prediction model is obtained by performing index prediction processing on multiple recommendation features of the task to be recommended through the second prediction model.

[0048] Based on the recommendation metrics of the first prediction model and the recommendation metrics of the second prediction model, the recommendation metrics for the target user corresponding to the information to be recommended are obtained.

[0049] In the above technical solution, the prediction module is further used to concatenate the feature domains corresponding to the multiple recommended features through the first prediction model to obtain the concatenated feature domains.

[0050] The concatenated feature domain is weighted based on the weights of the first prediction model to obtain a weighted feature domain.

[0051] The bias of the first prediction model is summed with the weighted feature domain to obtain the recommendation index of the first prediction model.

[0052] In the above technical solution, the prediction module is further used to sum the recommendation index of the first prediction model and the recommendation index of the second prediction model to obtain the summed recommendation index.

[0053] The summed recommendation metrics are normalized to obtain the recommendation metrics for the target user corresponding to the information to be recommended.

[0054] In the above technical solution, the second prediction model includes multiple cascaded hidden layers; the prediction module is further used to map multiple recommendation features of the task to be recommended through the first hidden layer of the multiple cascaded hidden layers.

[0055] The mapping result of the first hidden layer is output to the subsequent cascaded hidden layers, so that mapping processing and mapping result output can continue in the subsequent cascaded hidden layers until the last hidden layer is output.

[0056] The mapping result output from the last hidden layer is used as the recommendation metric for the second prediction model.

[0057] In the above technical solution, the prediction module is further configured to perform the following processing through the j-th hidden layer of the plurality of cascaded hidden layers:

[0058] The mapping results of the (j-1)th hidden layer are weighted based on the weights of the j-th hidden layer to obtain the weighted mapping results;

[0059] The bias of the j-th hidden layer is summed with the weighted mapping result to obtain the mapping result of the j-th hidden layer, and the mapping result of the j-th hidden layer is output to the (j+1)-th hidden layer;

[0060] Where j is an increasing natural number and its value ranges from 2 ≤ j ≤ N-1, and N is the number of the multiple cascaded hidden layers.

[0061] This application provides an electronic device for information recommendation, the electronic device comprising:

[0062] Memory, used to store executable instructions;

[0063] When the processor executes the executable instructions stored in the memory, it implements the artificial intelligence-based information recommendation method provided in the embodiments of this application.

[0064] This application provides a computer-readable storage medium storing executable instructions for inducing a processor to execute and implement the artificial intelligence-based information recommendation method provided in this application.

[0065] This application provides a computer program that, when executed by a processor, implements the artificial intelligence-based information recommendation method provided in this application.

[0066] The embodiments of this application have the following beneficial effects:

[0067] By projecting multiple recommendation features into different projection spaces and performing feature interaction on the projected features in each projection space, the diversity of the feature domain is improved; based on the diverse feature domains, the accuracy of the recommendation indicators is improved, thereby enhancing the accuracy of the recommendations. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating an application scenario of the recommendation system provided in the embodiments of this application;

[0069] Figure 2 This is a schematic diagram of the structure of an electronic device for information recommendation provided in an embodiment of this application;

[0070] Figures 3-5 This is a flowchart illustrating the information recommendation method based on artificial intelligence provided in an embodiment of this application;

[0071] Figure 6 This is a schematic diagram of the structure of the feature interaction network provided in the embodiments of this application;

[0072] Figure 7 This is a schematic diagram of the multi-channel structure of the feature interaction network provided in the embodiments of this application;

[0073] Figure 8 This is a schematic diagram of the structure of the feature interaction layer provided in the embodiments of this application;

[0074] Figure 9 This is a flowchart illustrating the information recommendation method based on artificial intelligence provided in an embodiment of this application;

[0075] Figure 10 This is a schematic diagram of the feature domain perception interaction module provided in the embodiments of this application. Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0077] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0078] Unless otherwise defined, 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 this application only and is not intended to limit this application.

[0079] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0080] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0081] 1) Target user: The user currently using the recommendation system, i.e., the current user. For example, if user A is using a text recommendation system to read news, then user A is the target user.

[0082] 2) User persona: Also known as user role, it is an effective tool for outlining target users and connecting user needs with design direction. User personas have been widely used in various fields. In practice, they are often used in the simplest and most relatable language to connect user attributes, behaviors, and expectations, serving as virtual representatives of actual users.

[0083] 3) Recommendation metrics: Metrics used to guide the recommendation system in making recommendations, such as whether the target user will click on the information to be recommended, whether the target user is interested in the information to be recommended, whether the target user will convert based on the information to be recommended, and whether the target user will rate the information to be recommended.

[0084] 4) Tasks to be recommended: Tasks corresponding to recommendation metrics, that is, tasks that the target user needs to perform based on the recommendation metrics for the information to be recommended. For example, the target user determines whether the information to be recommended needs to be recommended based on the recommendation metrics, or the target user determines whether the recommendation position of the information to be recommended needs to be adjusted based on the recommendation metrics.

[0085] 5) Click-through rate prediction: This refers to predicting the click-through rate of each piece of information to be recommended based on given information such as the information to be recommended (e.g., an advertisement), users, and context.

[0086] 6) Factorization Machines (FM): By introducing latent vectors, the training complexity is reduced. It is a machine learning algorithm based on matrix factorization and has a good learning ability for sparse data.

[0087] 7) Deep Neural Network (DNN): A type of feedforward neural network with a deep structure. It is a technique in the field of machine learning (ML) that can represent complex functions with fewer parameters.

[0088] This application provides an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence, which can improve the accuracy of recommendations.

[0089] The AI-based information recommendation method provided in this application can be implemented by a terminal / server alone; or it can be implemented collaboratively by a terminal and a server. For example, the terminal can independently undertake the AI-based information recommendation method described below, or the terminal can send an information recommendation request for a target user to the server, and the server can execute the AI-based information recommendation method according to the received information recommendation request for the target user, determine the recommendation index of the target user corresponding to the information to be recommended, and execute the recommendation task based on the recommendation index of the target user corresponding to the information to be recommended, in response to the information recommendation request for the target user.

[0090] The electronic device for information recommendation provided in this application can be various types of terminal devices or servers. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, smart vehicle, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0091] Taking servers as an example, such as server clusters deployed in the cloud, AI as a Service (AIaaS) is offered to users. The AIaaS platform breaks down several common AI services and provides them as independent or packaged services in the cloud. This service model is similar to an AI-themed marketplace, where all users can access and use one or more AI services provided by the AIaaS platform through application programming interfaces.

[0092] For example, one type of AI cloud service could be an information recommendation service, where a cloud server encapsulates the information recommendation program provided in this application embodiment. Users invoke the information recommendation service in the cloud service through a terminal (running a client, such as a news client, video client, etc.), causing the cloud-deployed server to invoke the encapsulated information recommendation program. This program projects features based on the projection space corresponding to each recommendation feature of the task to be recommended, and performs feature interaction based on each recommendation feature and the corresponding projection features of the projection space (i.e., projection representation). It also makes predictions based on the feature domains corresponding to multiple recommendation features (i.e., feature domain-aware feature representations), and executes the task to be recommended based on recommendation metrics. This distributes the information to be recommended to users who match their interests, quickly obtains user behavior data, and improves the effectiveness of information recommendation.

[0093] See Figure 1 , Figure 1 This is a schematic diagram of the application scenario of the recommendation system 10 provided in the embodiments of this application. The terminal (exemplarily shown as terminal 200-1, terminal 200-2 and terminal 200-3) connects to the server 100 through the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0094] A terminal (running a client, such as a news client or a video client) can be used to obtain information recommendation requests for a target user. For example, when a target user opens a video client running on the terminal, the terminal automatically obtains video recommendation requests for the target user.

[0095] In some embodiments, after the terminal obtains an information recommendation request for a target user, it calls the information recommendation interface of the server 100 (which can be provided as a cloud service, i.e., an information recommendation service). Based on the information recommendation request for the target user, the server 100 performs feature projection based on the projection space corresponding to each recommendation feature of the task to be recommended, and performs feature interaction based on each recommendation feature and the projected features of the corresponding projection space. It also performs prediction based on the feature domains corresponding to multiple recommendation features, and executes the task to be recommended based on recommendation metrics. This distributes the information to be recommended to users who meet their interests, quickly obtains user behavior data, improves the effectiveness of information recommendation, and responds to the information recommendation request for the target user.

[0096] As an application example, in a video application, when multiple target users open the video client running on their terminals, the terminals automatically obtain video recommendation requests for the target users and call the information recommendation interface of server 100. Server 100 executes an AI-based information recommendation method based on multiple recommendation features of the task to be recommended, determines the recommendation indicators (e.g., whether it has been clicked) for each of the target users corresponding to the video to be recommended, and then estimates the click-through rate of the video to be recommended based on the recommendation indicators for each of the target users. For example, if 80 out of 100 target users are likely to click on the video to be recommended, then the estimated click-through rate of the video to be recommended is 80%. The recommendation position of the video to be recommended is then adjusted so that it is placed in a position with better exposure, thereby quickly obtaining user behavior data, improving the effectiveness of information recommendation, and responding to the information recommendation requests for the target users.

[0097] In some embodiments, an information recommendation plugin can be embedded in the client running on the terminal to implement an AI-based information recommendation method locally on the client. For example, after the terminal obtains an information recommendation request for a target user, it calls the information recommendation plugin to implement an AI-based information recommendation method. This method involves projecting features based on the projection space corresponding to each recommendation feature of the task to be recommended, performing feature interaction based on each recommendation feature and the projected features of the corresponding projection space, making predictions based on the feature domains corresponding to multiple recommendation features, and executing the task to be recommended based on recommendation metrics to respond to the information recommendation request for the target user.

[0098] As an application example, for a news application, when a target user opens the news client running on their terminal, the terminal automatically obtains a news recommendation request for the target user, calls the information recommendation plugin, and executes an AI-based information recommendation method based on multiple recommendation features of the task to be recommended. This determines the recommendation index (e.g., whether the target user will click on the recommended news) for the target user. When the recommendation index indicates that the target user will click on the recommended news, the recommendation task is executed, thereby quickly obtaining user behavior data, improving the effectiveness of information recommendation, and responding to the information recommendation request for the target user.

[0099] The structure of the electronic device for information recommendation provided in the embodiments of this application is described below. See also... Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device 500 for information recommendation provided in an embodiment of this application. The example given is that the electronic device 500 is a server. Figure 2 The illustrated electronic device 500 for information recommendation 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 via a bus system 540. It is understood that the bus system 540 is used to implement communication between these components. In addition to a data bus, the bus system 540 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 2 The general labeled all buses as Bus System 540.

[0100] 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. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0101] Memory 550 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 550 described in this application embodiment is intended to include any suitable type of memory. Memory 550 may optionally include one or more storage devices physically located away from processor 510.

[0102] In some embodiments, memory 550 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0103] Operating system 551 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0104] The network communication module 552 is used to reach other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0105] In some embodiments, the AI-based information recommendation device provided in this application can be implemented in software. For example, it can be an information recommendation service in a server as described above, or an information recommendation plugin in a terminal as described above. Of course, it is not limited to this; the AI-based information recommendation device provided in this application can be provided in various software embodiments, including various forms such as applications, software, software modules, scripts, or code.

[0106] Figure 2 An AI-based information recommendation device 555, stored in a memory 550, is shown. It can be software in the form of programs and plugins, such as information recommendation plugins, and includes a series of modules, including an acquisition module 5551, a feature interaction module 5552, a prediction module 5553, and a recommendation module 5554. The acquisition module 5551, the feature interaction module 5552, the prediction module 5553, and the recommendation module 5554 are used to implement the information recommendation function provided in the embodiments of this application.

[0107] As mentioned above, the AI-based information recommendation method provided in this application can be implemented by various types of electronic devices. See also Figure 3 , Figure 3This is a flowchart illustrating the information recommendation method based on artificial intelligence provided in the embodiments of this application, combined with... Figure 3 The steps shown are explained.

[0108] In the following steps, the information to be recommended can be text, images, graphic content, video, or other data. Recommendation metrics are indicators used to guide the recommendation system in making recommendations. Examples include whether a target user will click on the recommended information, whether the target user is interested in the recommended information, whether the target user will convert based on the recommended information, and whether the target user will rate the recommended information.

[0109] In step 101, multiple recommendation features of the task to be recommended are obtained. The multiple recommendation features include at least one item feature of the information to be recommended and at least one user feature of the target user.

[0110] For example, feature extraction processing can be performed on the information to be recommended using offline or online methods to obtain the item 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 the category and price of the information to be recommended.

[0111] For example, feature extraction processing can be performed on the user profile of the target user through offline or online methods to obtain the user characteristics of the target user. These user characteristics are used to represent the characteristics of the target user, such as the target user's age, preferences, gender, etc.

[0112] For example, after the recommendation features of the task to be recommended (item features of the information to be recommended and user features of the target user) are extracted in advance through offline means, when the target user opens the client running on the terminal, the terminal automatically obtains the information recommendation request for the target user (including the target user's identifier), and obtains the pre-extracted recommendation features of the task to be recommended based on the information recommendation request for the target user, so as to perform feature interaction operations based on the recommendation features of the task to be recommended in the future.

[0113] In step 102, based on the projection space corresponding to each recommended feature, feature projection processing is performed on multiple recommended features to obtain the projected features of the projection space.

[0114] For example, after obtaining the recommendation features of the task to be recommended, for the projection space corresponding to each recommendation feature, the feature interaction network is used to perform feature projection processing on multiple recommendation features to obtain the projection features (i.e., projection representation) of each projection space. For example, for the projection space i corresponding to recommendation feature i, feature projection is performed on all recommendation features to obtain the projection features of all recommendation features in projection space i, i.e., the projection features of projection space i.

[0115] In some embodiments, based on the projection space corresponding to each recommended feature, feature projection processing is performed on multiple recommended features to obtain the projected features of the projection space, including: performing nonlinear feature extraction processing on multiple recommended features based on the weights of the projection space corresponding to each recommended feature to obtain the extracted features of the projection space; and multiplying the extracted features of the projection space with the projection matrix of the projection space to obtain the projected features of the projection space.

[0116] like Figure 6 As shown, nonlinear feature extraction can be achieved in the following way: based on the weights of the projection space corresponding to each recommended feature, multiple recommended features are scaled to obtain scaled recommended features, and the scaled recommended features are concatenated to obtain the extracted features of the projection space.

[0117] like Figure 6 As shown, the formula for feature projection is: ,in, Let j represent the recommended feature. Let represent the weight of the corresponding recommendation feature j in the projection space i. Let S represent the projection matrix of projection space i, S represent the set of recommended features, and K represent the number of recommended features. The projection features representing projection space i are not limited to those in the embodiments of this application. It can also be a transformation formula for other projections.

[0118] In step 103, feature interaction processing is performed based on each recommended feature and the corresponding projection features of the projection space to obtain the feature domain corresponding to each recommended feature.

[0119] For example, in obtaining the projection features of projection space i Then, the projection features and recommendation features Perform feature interaction to obtain the feature domain corresponding to the recommended feature i. (i.e., feature domain-aware feature representation), thereby obtaining the feature domain corresponding to each recommended feature. This allows for the projection processing of multiple recommendation features in different projection spaces, and feature interaction is performed on the projected features in each projection space, thereby improving the diversity of the feature domain and thus improving the accuracy of the recommendation metrics in the future.

[0120] In some embodiments, the feature domain corresponding to each recommended feature is obtained by performing feature interaction processing on each recommended feature and the projected features of the corresponding projection space, including: multiplying each recommended feature and the projected features of the corresponding projection space to obtain the feature vector of each recommended feature; and mapping the feature vector of each recommended feature to obtain the feature domain corresponding to each recommended feature.

[0121] like Figure 6 As shown, the formula for feature interaction processing is: ,in, Let i represent the recommended feature. Represents the projection characteristics of projection space i. This represents a mapping process. This represents the feature domain corresponding to the recommended feature i. The embodiments of this application are not limited to this. Other feature interaction transformation formulas also apply.

[0122] See Figure 4 , Figure 4 This is an optional flowchart illustrating an artificial intelligence-based information recommendation method provided in an embodiment of this application. Figure 4 Show Figure 3 Step 102 can also be implemented through step 1021A: the projection space corresponding to each recommended feature includes multiple channels; in step 1021A, based on any channel of the projection space corresponding to each recommended feature, feature projection processing is performed on multiple recommended features to obtain the projection feature of any channel; Figure 4 Show Figure 3 Step 103 can also be implemented through steps 1031A-1032A: In step 1031A, feature interaction processing is performed on each recommended feature and the projection feature of any corresponding channel to obtain the feature domain corresponding to any channel; In step 1032A, the feature domains corresponding to multiple channels in the projection space are fused to obtain the feature domain corresponding to each recommended feature.

[0123] See Figure 7 For the projection space i, the formula for the feature projection is: ,in, Let j represent the recommended feature. Let represent the weight of the recommended feature j in channel h of projection space i. Let S represent the projection matrix of channel h in projection space i, S represent the set of recommended features, and K represent the number of recommended features. The projection characteristics of channel h in projection space i are not limited to those in the embodiments of this application. It can also be a transformation formula for other projections.

[0124] See Figure 7 The formula for feature interaction processing is: ,in, Let i represent the recommended feature. The projection characteristics of channel h in projection space i are represented. This represents a mapping process. This application's embodiments represent the feature domain corresponding to channel h of recommended feature i, but are not limited to this. Other feature interaction transformation formulas also apply.

[0125] For example, the formula for fusion is: Where H represents the number of channels. This represents any combination function, such as element-wise summation or element-wise averaging. That is, summing the feature domains corresponding to multiple channels in the projection space and using the summation result as the feature domain corresponding to each recommended feature; or averaging the feature domains corresponding to multiple channels in the projection space and using the average result as the feature domain corresponding to each recommended feature.

[0126] See Figure 5 , Figure 5 This is an optional flowchart illustrating an artificial intelligence-based information recommendation method provided in an embodiment of this application. Figure 5 Show Figure 3 Step 102 can also be implemented through step 1021B: the feature interaction network for feature interaction includes multiple cascaded feature interaction layers; in step 1021B, the following processing is performed through any feature interaction layer in the feature interaction network: based on the projection space corresponding to each recommended feature, feature projection is performed on the multiple recommended features input to any feature interaction layer to obtain the projected features of any feature interaction layer. Figure 5 Show Figure 3 Step 103 can also be implemented through steps 1031B-1032B: Perform the following processing through any feature interaction layer in the feature interaction network: perform feature interaction processing based on multiple recommended features input to any feature interaction layer and the projected features of the corresponding feature interaction layer to obtain the feature domain of any feature interaction layer, and output the multiple recommended features included in the feature domain; in step 1032B, the feature domain of the last feature interaction layer is determined to be the feature domain corresponding to each recommended feature.

[0127] The multiple recommended features input to the first feature interaction layer are the multiple recommended features included in the task to be recommended. The multiple recommended features input to the subsequent feature interaction layers are the multiple recommended features output by the previous feature interaction layer of the subsequent feature interaction layers. The subsequent feature interaction layers are the feature interaction layers other than the first feature interaction layer among the multiple cascaded feature interaction layers.

[0128] See Figure 8 The feature interaction network consists of L cascaded feature interaction layers, targeting the Lth feature interaction layer. l The nth feature interaction layer, based on the projection space corresponding to each recommended feature, processes the input to the nth feature interaction layer. l Projecting multiple recommendation features from the feature interaction layer yields the first feature. l The projected features of the feature interaction layer are based on the input to the first feature layer. l Multiple recommendation features of the feature interaction layer and the first feature layer l The projected features of the first feature interaction layer are processed through feature interaction to obtain the second feature interaction layer. l The feature domain of the first feature interaction layer is output, along with multiple recommended features included in the feature domain (i.e., the first feature interaction layer). l The feature domain of the feature interaction layer is used as the first feature domain. l The feature interaction layer input is given to the first... l+ The recommended features of the first feature interaction layer are used to determine the feature domain of the last feature interaction layer as the feature domain corresponding to each recommended feature.

[0129] For example, the processing procedure of a feature interaction network is as follows: ,in, Indicates the first l The processing of each feature interaction layer Indicates input to the first l Recommended features of the feature interaction layer Indicates the first l The feature domain of each feature interaction layer.

[0130] In some embodiments, feature interaction processing is performed based on multiple recommended features input to any feature interaction layer and the corresponding projected features of any feature interaction layer to obtain the feature domain of any feature interaction layer, including: performing feature interaction processing based on multiple recommended features input to any feature interaction layer and the corresponding projected features of any feature interaction layer to obtain the latent features of any feature interaction layer; and summing the latent features of any feature interaction layer with the multiple recommended features input to any feature interaction layer to obtain the feature domain of any feature interaction layer.

[0131] For example, to include feature interaction information at any level in the final output, residual connections are used in each layer of feature interaction; that is, the processing procedure of the feature interaction network is as follows: ,in, Indicates the first l The processing of each feature interaction layer Indicates input to the first l Recommended features of the feature interaction layer Indicates the first l The feature domain of each feature interaction layer.

[0132] In step 104, index prediction processing is performed based on the feature domains corresponding to multiple recommendation features to obtain the recommendation index for the target user corresponding to the information to be recommended.

[0133] For example, after obtaining the feature domains corresponding to multiple recommendation features, index prediction processing is performed based on the feature domains corresponding to multiple recommendation features to obtain the recommendation indexes for the target users corresponding to the information to be recommended. For example, whether the target user will click on the information to be recommended, whether the target user is interested in the information to be recommended, whether the target user will convert based on the information to be recommended, whether the target user will rate the information to be recommended, etc., and then the recommendation task is executed based on the recommendation indexes for the target users corresponding to the information to be recommended, thereby improving the accuracy of information recommendation.

[0134] In some embodiments, the indicator prediction processing is implemented through a first prediction model and a second prediction model; the indicator prediction processing based on the feature domains corresponding to multiple recommendation features to obtain the recommendation indicator for the target user corresponding to the information to be recommended includes: performing indicator prediction processing on the feature domains corresponding to multiple recommendation features through the first prediction model to obtain the recommendation indicator of the first prediction model; performing indicator prediction processing on multiple recommendation features of the recommendation task to be recommended through the second prediction model to obtain the recommendation indicator of the second prediction model; and obtaining the recommendation indicator for the target user corresponding to the information to be recommended based on the recommendation indicator of the first prediction model and the recommendation indicator of the second prediction model.

[0135] For example, the process of using the first prediction model to predict the index of the feature domains corresponding to multiple recommendation features is as follows: The first prediction model is used to concatenate the feature domains corresponding to multiple recommendation features to obtain the concatenated feature domain; the concatenated feature domain is weighted based on the weights of the first prediction model to obtain the weighted feature domain; the bias of the first prediction model is summed with the weighted feature domain to obtain the recommendation index of the first prediction model.

[0136] For example, the first prediction model includes a single forward layer, through which the metric is predicted. Where W represents the weights of the forward layer, and b represents the bias of the forward layer. The first prediction model in this embodiment is not limited to the forward layer; other prediction models are also applicable.

[0137] In some embodiments, the second prediction model includes multiple cascaded hidden layers; the second prediction model is used to perform index prediction processing on multiple recommendation features of the recommendation task to obtain the recommendation index of the second prediction model, including: mapping multiple recommendation features of the recommendation task through the first hidden layer of the multiple cascaded hidden layers; outputting the mapping result of the first hidden layer to the subsequent cascaded hidden layers, so as to continue the mapping processing and output the mapping result in the subsequent cascaded hidden layers, until the last hidden layer is output; and using the mapping result output by the last hidden layer as the recommendation index of the second prediction model.

[0138] For example, the second prediction model is a multilayer perceptron (MLP), which is used to predict indicators. In this context, MLP represents the processing procedure of a multilayer perceptron (which includes multiple cascaded hidden layers). This represents multiple recommendation features for the task to be recommended. The recommended metric for prediction by the multilayer perceptron is indicated. The second prediction model in this embodiment is not limited to the multilayer perceptron, and other prediction models are also applicable.

[0139] For example, the process of mapping and outputting the mapping results in subsequent cascaded hidden layers is as follows: The following processing is performed in the j-th hidden layer of multiple cascaded hidden layers: the mapping result of the (j-1)-th hidden layer is weighted based on the weight of the j-th hidden layer to obtain the weighted mapping result; the bias of the j-th hidden layer is summed with the weighted mapping result to obtain the mapping result of the j-th hidden layer, and the mapping result of the j-th hidden layer is output to the (j+1)-th hidden layer; where j is an incrementing natural number and its value ranges from 2≤j≤N-1, and N is the number of multiple cascaded hidden layers.

[0140] In some embodiments, the recommendation index for the target user corresponding to the information to be recommended is obtained based on the recommendation index of the first prediction model and the recommendation index of the second prediction model, including: summing the recommendation index of the first prediction model and the recommendation index of the second prediction model to obtain the summed recommendation index; and normalizing the summed recommendation index to obtain the recommendation index for the target user corresponding to the information to be recommended.

[0141] For example, the recommendation metric for the target user corresponding to the information to be recommended is ,in, This represents the recommended index predicted by the first prediction model. This indicates the recommended index predicted by the second prediction model. This represents a normalization function, but the embodiments of this application are not limited to a normalization function. Other normalization functions are also applicable.

[0142] In step 105, the recommendation task is executed based on the recommendation metrics of the target user corresponding to the information to be recommended.

[0143] For example, after obtaining the recommendation metrics of the target user corresponding to the information to be recommended, the recommendation task is executed based on the recommendation metrics of the target user corresponding to the information to be recommended. For example, when the recommendation metrics of the target user corresponding to the information to be recommended indicate that the target user will click on the information to be recommended, the recommendation task is executed, thereby quickly obtaining user behavior data and improving the effect of information recommendation.

[0144] The following will describe an exemplary application of the embodiments of this application in a real-world application scenario.

[0145] This application can be applied to various product applications, including e-commerce shopping, video (or music) recommendations, news and information feed recommendations, and lifestyle service scenarios. It can infer the optimal set of items based on the user's known interests and specific environmental information, thereby achieving a win-win situation for the platform, content providers, and users.

[0146] For example, in movie recommendations, embodiments of this application can use scene-related features and user-related features as model inputs to train a CTR prediction model, which can then be deployed online to recommend items that users may be interested in, significantly improving the accuracy of the recommendation system and thereby increasing user satisfaction with the product. Similarly, in online advertising scenarios, the trained CTR prediction model can be used to accurately recommend commercial advertisements that users are likely to click on, directly increasing the company's revenue.

[0147] Because the embodiments of this application fully consider the limitations of extraction for different feature domains, they have superior extraction capabilities in feature interaction. They do not require redundant model training, and while ensuring model training efficiency, they can also improve the accuracy of CTR prediction, provide a better user experience, and generate more significant commercial revenue.

[0148] Click-through rate (CTR) prediction has broad application prospects in personalized recommendation systems. It can estimate the probability that a user will click on a list of recommended items in a specific context, enabling users to efficiently obtain information of interest even in situations of information overload. Building an accurate CTR prediction model is a crucial step for personalized recommendation systems to deliver content efficiently; the accuracy of its scoring directly impacts the user experience and brings significant commercial returns to the company's products.

[0149] In specific recommendation scenarios, since the feature spaces of different user interests are often sparse and high-dimensional, existing click-through rate (CTR) prediction has significant room for improvement in two aspects: 1) Manual feature combination can improve the accuracy of the model, and CTR prediction models have significant room for improvement in extracting high-order feature representations; 2) Existing deep models use vector products to constrain the interaction strength between feature domains, or simply project a single feature space onto multiple shared spaces, but this cannot obtain deep interaction information between different feature domains. These problems lead to time-consuming and cumbersome training processes and poor prediction performance, which to some extent limits the application effect of CTR prediction technology.

[0150] In related technologies, click-through rate (CTR) prediction methods based on machine learning include the following two: 1. CTR prediction models based on field-aware factorization machines (FFM); 2. CTR prediction models based on deep learning models. The applicant has found that both of these approaches have their own shortcomings:

[0151] 1) The CTR prediction model based on FFM has a high computational complexity because it needs to learn k-dimensional latent vectors of n features in f domains during training. On the other hand, FFM models and their variants are limited by their inherent structure, and can only extract low-order interactions between features, which limits the model's predictive ability.

[0152] 2) The CTR prediction model based on the deep model uses scalar product to constrain the interaction strength between feature domains, but it simply extends a single shared space to multiple shared projection spaces, which cannot capture the complete information of different projection spaces, resulting in a decrease in prediction performance.

[0153] To address the aforementioned issues, this application proposes an AI-based information recommendation method. This method enhances the nonlinear extraction of multi-domain features by strengthening the model, thereby meeting the increased demands for CTR prediction accuracy in recommendation systems. The method employs a feature interaction network to learn the semantic diversity of each feature domain (one feature corresponds to one feature domain). This network possesses field-aware properties and can be integrated into most deep models. Furthermore, it incorporates a hierarchical structure to improve the high-level ambiguity of feature interactions. Combining the feature interaction network and the hierarchical structure effectively enhances the accuracy of click-through rate prediction.

[0154] like Figure 9 As shown, this application addresses the limitations of CTR models in recommendation systems when processing interaction representations across different feature domains, and the need to improve nonlinear extraction capabilities. It proposes a feature interaction network based on feature domain awareness, comprising the following steps:

[0155] Step 1: Build the dataset.

[0156] First, each movie rating sample is converted into the format required for training, and configured into sparse features, continuous features, and labels. The labels are the user's historical ratings of movies, which can be further transformed according to the optimization objective. Then, the data is input into a feature domain-aware feature interaction network, which uses the feature domain-aware interaction module (i.e., the feature interaction layer) to learn key feature interaction representations, and combines the hierarchical structure to further enhance the prediction accuracy. Finally, the model predicts the click probability corresponding to each input sample.

[0157] The information recommendation method based on artificial intelligence in this application includes the following steps:

[0158] For example, a dataset is a collection of users' historical rating records for movies, with different data sizes, such as 1M (containing 10,000 ratings), 10M (containing 100,000 ratings), and 20M (containing 200,000 ratings).

[0159] This application uses a 1M dataset, which includes approximately 1 million rating records from 6040 unique users for 3900 movies. Each sample contains movie item features, user information features, and the user's historical ratings for the movie. The user ratings are based on a 5-star scale, increasing in increments of half a star (0.5 stars to 5 stars). There are a total of 7 categories of features related to movie items and user information.

[0160] In this embodiment of the application, all collected samples are randomly divided into training set, validation set and test set in a ratio of 8:1:1.

[0161] Step 2: Data preprocessing.

[0162] This application embodiment performs preprocessing and transformation on numerical features and user ratings:

[0163] 1) In movie rating, the original values ​​are divided into positive and negative samples with a threshold of 3. Samples with a rating greater than or equal to 3 are positive samples, and samples with a rating less than 3 are negative samples.

[0164] 2) Numerical features were standardized according to the rule: if feature x>2, standardized feature y = log2(x). Other standardization operations, such as mean-variance standardization, can also be used in this embodiment.

[0165] Step 3: Construct a feature interaction network based on feature domain awareness.

[0166] like Figure 10As shown, considering that each feature domain is located in a different projection space, this application proposes a feature domain perception interaction module and constructs a corresponding feature interaction network.

[0167] like Figure 10 As shown, each dashed block represents a projection space associated with the feature domain. For features... (i.e., the original features of the sample) The corresponding projection space i, all input features ( to ) will first be based on weight The scaled features are then stitched together and then projected through a matrix. Projection, the vector obtained after projection Again with Element-wise multiplication yields New representation of features The projection matrix in the embodiments of this application and weight It is specific to the feature Therefore, for features The corresponding feature domain can project all features onto In the corresponding projection space, and then with By performing feature interactions, information in the feature domain (i.e., new representations) can be captured. ).and The process of feature interaction can be obtained from the following formulas (1)-(2):

[0168] (1)

[0169] Where S represents the set of features and K represents the number of features.

[0170] (2)

[0171] in, It is a feature in its corresponding projection matrix New combined features are formed by feature interactions with other features. This results in a new set of feature vectors. This will be used for further feature interactions. Specifically, in training the dataset, the total number of features for movie items and user information is 7, therefore the feature domain K is 7. This feature set is then interacted with other features in the corresponding projection space to obtain a new set of feature domain vectors, i.e., a new feature set.

[0172] In this embodiment, multiple channels are extended for each feature domain, meaning each feature domain contains multiple sets of projection matrices and weights to improve the expressiveness of the model. For example, each feature domain has H channels, and the multi-channel feature interaction representation is shown in formulas (3)-(4):

[0173] (3)

[0174] (4)

[0175] Each channel h contains a separate projection matrix. and weight Ultimately, the characteristics The update is calculated from all channels, as shown in formula (5):

[0176] (5)

[0177] in, This represents any combination function, such as element-wise summation or element-wise averaging. The number of channels H can be determined based on the number of training features, typically an integer within the interval [2, 6]. In this embodiment, H is configured as 2.

[0178] On the other hand, a simple single-layer domain-aware feature interaction layer can only capture second-order feature interactions, while higher-order feature interactions are crucial for click prediction tasks. Therefore, this embodiment stacks multiple domain-aware feature interaction layers to capture higher-order feature interaction information, as shown in formula (6):

[0179] (6)

[0180] in, This represents a domain-aware feature interaction layer. Indicates the first l The input of the layer, L represents the total number of feature interaction layers, and L is configured as 3 in this embodiment. This represents the initial eigenvector, i.e. .

[0181] In order to include feature interaction information of arbitrary levels in the final output, this embodiment uses residual connections in each layer of feature interaction, as shown in formula (7):

[0182] (7)

[0183] After obtaining the L-order feature interaction information, this embodiment of the application uses a single feedforward layer to calculate the output probability, as shown in formula (8):

[0184] (8)

[0185] Where W represents the weight of the forward layer and b represents the bias of the forward layer.

[0186] In order to capture implicit high-order feature interaction information, this embodiment of the application also uses an independent multilayer perceptron (MLP) to calculate the probability based on the initial feature vector, as shown in formula (9):

[0187] (9)

[0188] The above and The values ​​are added together and normalized to obtain the final output value of the model, which is in the range [0, 1], representing the probability that the model predicts whether the sample will be clicked, as shown in formula (10):

[0189] (10)

[0190] Step 4: Construct the loss function for the model.

[0191] In this embodiment, the cross-entropy loss function (softmax loss) is used as the model's loss function, where y represents the actual tag that was clicked online. This represents the probability that the model predicts for this sample during training. The training objective is to optimize the model's prediction of whether a sample is clicked, as shown in formula (11):

[0192] (11)

[0193] in, This indicates whether the i-th sample is a true click tag. This represents the probability that the model predicts for the i-th sample during training.

[0194] Step 5: Training a feature interaction network based on feature domain awareness.

[0195] This application employs a first-order optimization algorithm (Adam optimizer) to solve for the model parameters, and uses the Xavier pattern (an efficient neural network initialization method that avoids the decay of activation value variance by maintaining a Gaussian distribution for the output values ​​of each layer) to initialize the model parameters. The neural network parameters are iteratively updated based on the training data. During the solution process, training samples from the dataset (including movie item features, user features, and corresponding labels) are fed into the model for training, and model optimization is completed through error backpropagation.

[0196] In summary, the embodiments of this application have the following beneficial effects:

[0197] 1) This application proposes a feature domain-aware interaction module that can realize feature interaction in different projection spaces, thereby capturing the high-order semantic diversity of each feature domain, obtaining the polysemy of each feature domain at the high-order semantic level, and realizing efficient feature interaction in a specific projection space without needing too much prior knowledge of manually combining interactive features.

[0198] 2) The embodiments of this application combine the proposed feature domain perception interaction module with the hierarchical structure to jointly extract high-order feature representations, which can accurately learn the high-order interaction of user behavior and item information in movie recommendation and other recommendation scenarios, thereby improving the accuracy of CTR model prediction;

[0199] 3) The feature domain-aware interaction module proposed in this application can be directly applied to common deep models in click-through rate prediction tasks in a modular manner, thereby training an efficient and accurate recommendation model. This allows the recommendation model to be applied to movie recommendation scenarios, resulting in a better user experience and higher commercial benefits.

[0200] This concludes the description of the artificial intelligence-based information recommendation method provided in this application, using the exemplary application and implementation of the server provided in the embodiments of this application. This application also provides an information recommendation device. In practical applications, the functional modules in the information recommendation device can be collaboratively implemented using the hardware resources of electronic devices (such as terminal devices, servers, or server clusters), such as computing resources like processors, communication resources (such as those used to support various communication methods like optical fiber and cellular), and memory. Figure 2 An AI-based information recommendation device 555 stored in memory 550 is shown. It can be software in the form of programs and plug-ins, such as software modules designed in programming languages ​​such as C / C++ and Java, application software designed in programming languages ​​such as C / C++ and Java, or dedicated software modules, application programming interfaces, plug-ins, cloud services, etc. in large software systems. Examples of different implementation methods are given below.

[0201] Example 1: Information recommendation devices are mobile applications and modules.

[0202] The information recommendation device 555 in this embodiment can provide a software module designed using programming languages ​​such as C / C++ and Java, which is embedded in various mobile applications based on systems such as Android or iOS (stored as executable instructions in the storage medium of the mobile device and executed by the processor of the mobile device), thereby directly using the computing resources of the mobile device itself to complete the relevant information recommendation tasks, and periodically or irregularly transmitting the processing results to a remote server through various network communication methods, or saving them locally on the mobile device.

[0203] Example 2: The information recommendation device is a server application and platform.

[0204] The information recommendation device 555 in this embodiment can be a dedicated software module in an application software or large software system designed using programming languages ​​such as C / C++ and Java. It runs on the server side (stored in the server-side storage medium as executable instructions and executed by the server-side processor). The server uses its own computing resources to complete the relevant information recommendation tasks.

[0205] This application embodiment can also provide an information recommendation platform (for information recommendation) for use by individuals, groups or organizations, by mounting a customized, easy-to-interact web interface or other user interfaces (UI) on a distributed, parallel computing platform composed of multiple servers.

[0206] Example 3: The information recommendation device consists of server-side application programming interfaces (APIs) and plugins.

[0207] The information recommendation device 555 in this embodiment can be provided as a server-side API or plugin for users to call, to execute the AI-based information recommendation method of this embodiment, and to be embedded in various applications.

[0208] Example 4: Information recommendation devices are mobile device client APIs and plugins.

[0209] The information recommendation device 555 in this embodiment can be provided as an API or plugin on a mobile device for users to call in order to execute the artificial intelligence-based information recommendation method of this embodiment.

[0210] Example 5: The information recommendation device is an open service in the cloud.

[0211] The information recommendation device 555 in this embodiment can provide a cloud service for information recommendation developed for users, allowing individuals, groups or organizations to make information recommendations.

[0212] The information recommendation device 555 includes a series of modules, including an acquisition module 5551, a feature interaction module 5552, a prediction module 5553, and a recommendation module 5554. The following describes the scheme for the cooperation of the various modules in the information recommendation device 555 provided in this embodiment to achieve information recommendation.

[0213] The acquisition module 5551 is used to acquire multiple recommendation features of the task to be recommended, the multiple recommendation features including at least one item feature of the information to be recommended and at least one user feature of the target user; the feature interaction module 5552 is used to perform feature projection processing on the multiple recommendation features based on the projection space corresponding to each recommendation feature to obtain the projection features of the projection space; and to perform feature interaction processing based on each recommendation feature and the projection features corresponding to the projection space to obtain the feature domain corresponding to each recommendation feature; the prediction module 5553 is used to perform index prediction processing based on the feature domain corresponding to the multiple recommendation features to obtain the recommendation index of the information to be recommended corresponding to the target user; and the recommendation module 5554 is used to execute the task to be recommended based on the recommendation index of the information to be recommended corresponding to the target user.

[0214] In some embodiments, the feature interaction module 5552 is further configured to perform nonlinear feature extraction processing on the plurality of recommended features based on the weight of the projection space corresponding to each recommended feature, to obtain the extracted features of the projection space; and to multiply the extracted features of the projection space with the projection matrix of the projection space to obtain the projected features of the projection space.

[0215] In some embodiments, the feature interaction module 5552 is further configured to scale the plurality of recommended features based on the weight of the projection space corresponding to each of the recommended features to obtain scaled plurality of recommended features; and to stitch the scaled plurality of recommended features together to obtain the extracted features of the projection space.

[0216] In some embodiments, the feature interaction module 5552 is further configured to multiply each of the recommended features and the projection features corresponding to the projection space to obtain a feature vector of each of the recommended features; and to map the feature vector of each of the recommended features to obtain a feature domain corresponding to each of the recommended features.

[0217] In some embodiments, the projection space corresponding to each recommended feature includes multiple channels; the feature interaction module 5552 is further configured to perform feature projection processing on the multiple recommended features based on any one of the channels in the projection space corresponding to each recommended feature, to obtain the projection feature of any one of the channels; perform feature interaction processing on each recommended feature and the projection feature corresponding to any one of the channels, to obtain the feature domain corresponding to any one of the channels; and perform fusion processing on the feature domains corresponding to the multiple channels of the projection space, respectively, to obtain the feature domain corresponding to each recommended feature.

[0218] In some embodiments, the feature interaction module 5552 is further configured to sum the feature domains corresponding to the multiple channels of the projection space, and use the summation result as the feature domain corresponding to each of the recommended features; or, to average the feature domains corresponding to the multiple channels of the projection space, and use the average result as the feature domain corresponding to each of the recommended features.

[0219] In some embodiments, the feature interaction network for feature interaction includes multiple cascaded feature interaction layers; the feature interaction module 5552 is further configured to perform the following processing through any of the feature interaction layers in the feature interaction network: based on the projection space corresponding to each of the recommended features, perform feature projection processing on the multiple recommended features input to any of the feature interaction layers to obtain the projected features of any of the feature interaction layers; wherein, the multiple recommended features input to the first feature interaction layer are multiple recommended features included in the task to be recommended, the multiple recommended features input to the subsequent feature interaction layers are multiple recommended features output by the previous feature interaction layer of the subsequent feature interaction layer, and the subsequent feature interaction layer is the feature interaction layer other than the first feature interaction layer among the multiple cascaded feature interaction layers; perform the following processing through any of the feature interaction layers in the feature interaction network: perform feature interaction processing based on the multiple recommended features input to any of the feature interaction layers and the projected features corresponding to any of the feature interaction layers to obtain the feature domain of any of the feature interaction layers, and output the multiple recommended features included in the feature domain; determine the feature domain of the last feature interaction layer as the feature domain corresponding to each of the recommended features.

[0220] In some embodiments, the feature interaction module 5552 is further configured to perform feature interaction processing based on the plurality of recommended features input to any feature interaction layer and the projected features corresponding to any feature interaction layer, to obtain the latent features of any feature interaction layer; and to sum the latent features of any feature interaction layer with the plurality of recommended features input to any feature interaction layer to obtain the feature domain of any feature interaction layer.

[0221] In some embodiments, the metric prediction processing is implemented through a first prediction model and a second prediction model; the prediction module 5553 is further configured to perform metric prediction processing on the feature domains corresponding to the plurality of recommendation features through the first prediction model to obtain the recommendation metric of the first prediction model; perform metric prediction processing on the plurality of recommendation features of the task to be recommended through the second prediction model to obtain the recommendation metric of the second prediction model; and obtain the recommendation metric of the target user corresponding to the information to be recommended based on the recommendation metric of the first prediction model and the recommendation metric of the second prediction model.

[0222] In some embodiments, the prediction module 5553 is further configured to concatenate the feature domains corresponding to the plurality of recommendation features using the first prediction model to obtain a concatenated feature domain; weight the concatenated feature domain based on the weights of the first prediction model to obtain a weighted feature domain; and sum the bias of the first prediction model with the weighted feature domain to obtain the recommendation index of the first prediction model.

[0223] In some embodiments, the prediction module 5553 is further configured to sum the recommendation metrics of the first prediction model and the recommendation metrics of the second prediction model to obtain a summed recommendation metric; and to normalize the summed recommendation metric to obtain a recommendation metric for the target user corresponding to the information to be recommended.

[0224] In some embodiments, the second prediction model includes multiple cascaded hidden layers; the prediction module 5553 is further configured to map multiple recommendation features of the task to be recommended through the first hidden layer of the multiple cascaded hidden layers; output the mapping result of the first hidden layer to the subsequent cascaded hidden layers, so as to continue the mapping processing and output the mapping result in the subsequent cascaded hidden layers, until the last hidden layer is output; and use the mapping result output by the last hidden layer as the recommendation index of the second prediction model.

[0225] In some embodiments, the prediction module 5553 is further configured to perform the following processing through the j-th hidden layer of the plurality of cascaded hidden layers: weighting the mapping result of the (j-1)-th hidden layer based on the weight of the j-th hidden layer to obtain a weighted mapping result; summing the bias of the j-th hidden layer with the weighted mapping result to obtain the mapping result of the j-th hidden layer; and outputting the mapping result of the j-th hidden layer to the (j+1)-th hidden layer; wherein j is an incrementing natural number and its value range is 2≤j≤N-1, and N is the number of the plurality of cascaded hidden layers.

[0226] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. The processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the artificial intelligence-based information recommendation method described above in this application.

[0227] This application provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to execute the artificial intelligence-based information recommendation method provided in this application. For example... Figure 3-5 The example shown is an information recommendation method based on artificial intelligence.

[0228] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0229] In some embodiments, executable instructions may take 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 standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0230] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0231] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0232] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. An information recommendation method based on artificial intelligence, characterized in that, The method comprises: obtaining a plurality of recommendation features of a to-be-recommended task, the plurality of recommendation features comprising at least one item feature of to-be-recommended information and at least one user feature of a target user; performing feature projection processing on the plurality of recommendation features based on a projection space corresponding to each of the recommendation features to obtain a projection feature of the projection space, wherein the feature projection processing comprises: performing scaling processing on the plurality of recommendation features based on a weight of the projection space corresponding to each of the recommendation features to obtain scaled plurality of recommendation features; performing splicing processing on the scaled plurality of recommendation features to obtain an extracted feature of the projection space; and multiplying the extracted feature of the projection space by a projection matrix of the projection space to obtain the projection feature of the projection space; multiplying each of the recommendation features and the projection feature corresponding to the projection space to obtain a feature vector of each of the recommendation features; and performing mapping processing on the feature vector of each of the recommendation features to obtain a feature domain corresponding to each of the recommendation features; performing index prediction processing based on the feature domain corresponding to the plurality of recommendation features to obtain a recommendation index of the to-be-recommended information corresponding to the target user; and executing the to-be-recommended task based on the recommendation index of the to-be-recommended information corresponding to the target user.

2. The method of claim 1, wherein: the projection space corresponding to each of the recommendation features comprises a plurality of channels; the feature projection processing on the plurality of recommendation features based on the projection space corresponding to each of the recommendation features to obtain the projection feature of the projection space comprises: performing feature projection processing on the plurality of recommendation features based on any of the channels of the projection space corresponding to each of the recommendation features to obtain a projection feature of any of the channels.

3. The method of claim 2, wherein, The method further comprises: performing feature interaction processing on each of the recommendation features and the projection feature corresponding to any of the channels to obtain a feature domain corresponding to any of the channels; and performing fusion processing on the feature domains corresponding to the plurality of channels of the projection space respectively to obtain the feature domain corresponding to each of the recommendation features.

4. The method of claim 3, wherein, The fusion processing on the feature domains corresponding to the plurality of channels of the projection space respectively to obtain the feature domain corresponding to each of the recommendation features comprises: performing summation processing on the feature domains corresponding to the plurality of channels of the projection space respectively, and taking a result of the summation processing as the feature domain corresponding to each of the recommendation features; or performing average processing on the feature domains corresponding to the plurality of channels of the projection space respectively, and taking a result of the average processing as the feature domain corresponding to each of the recommendation features.

5. The method of claim 1, wherein: a feature interaction network used for feature interaction comprises a plurality of cascaded feature interaction layers; the feature projection processing on the plurality of recommendation features based on the projection space corresponding to each of the recommendation features to obtain the projection feature of the projection space comprises: performing the following processing through any of the feature interaction layers in the feature interaction network: perform feature projection processing on the multiple recommendation features input to any of the feature interaction layers based on the projection space corresponding to each of the recommendation features, to obtain projection features of any of the feature interaction layers; wherein the multiple recommendation features input to the first feature interaction layer are multiple recommendation features included in the to-be-recommended task, the multiple recommendation features input to a subsequent feature interaction layer are multiple recommendation features output by a previous feature interaction layer of the subsequent feature interaction layer, and the subsequent feature interaction layer is a feature interaction layer other than the first feature interaction layer in the multiple cascaded feature interaction layers.

6. The method of claim 5, wherein, The method further includes: performing the following processing by any of the feature interaction layers in the feature interaction network: perform feature interaction processing on the multiple recommendation features input to any of the feature interaction layers and the projection features corresponding to any of the feature interaction layers, to obtain a feature field of any of the feature interaction layers, and output multiple recommendation features included in the feature field; determining the feature field of the last feature interaction layer as the feature field corresponding to each of the recommendation features.

7. The method of claim 6, wherein, The feature interaction processing on the multiple recommendation features input to any of the feature interaction layers and the projection features corresponding to any of the feature interaction layers to obtain a feature field of any of the feature interaction layers includes: performing feature interaction processing on the multiple recommendation features input to any of the feature interaction layers and the projection features corresponding to any of the feature interaction layers, to obtain hidden features of any of the feature interaction layers; performing summation processing on the hidden features of any of the feature interaction layers and the multiple recommendation features input to any of the feature interaction layers, to obtain a feature field of any of the feature interaction layers.

8. The method of claim 1, wherein the index prediction processing is implemented by a first prediction model and a second prediction model; the index prediction processing based on the feature fields corresponding to the multiple recommendation features to obtain a recommendation index of the to-be-recommended information corresponding to the target user includes: performing index prediction processing on the feature fields corresponding to the multiple recommendation features by the first prediction model, to obtain a recommendation index of the first prediction model; performing index prediction processing on the multiple recommendation features of the to-be-recommended task by the second prediction model, to obtain a recommendation index of the second prediction model; obtaining a recommendation index of the to-be-recommended information corresponding to the target user based on the recommendation index of the first prediction model and the recommendation index of the second prediction model.

9. The method of claim 8, wherein, The index prediction processing on the feature fields corresponding to the multiple recommendation features by the first prediction model to obtain a recommendation index of the first prediction model includes: performing concatenation processing on the feature fields corresponding to the multiple recommendation features by the first prediction model, to obtain concatenated feature fields; performing weighting processing on the concatenated feature fields based on weights of the first prediction model, to obtain weighted feature fields; performing summation processing on the bias of the first prediction model and the weighted feature fields, to obtain a recommendation index of the first prediction model.

10. The method of claim 8, wherein, the second prediction model comprises a plurality of cascaded hidden layers; the index prediction processing on the plurality of recommendation features of the to-be-recommended task by the second prediction model to obtain a recommendation index of the second prediction model comprises: mapping processing on the plurality of recommendation features of the to-be-recommended task by a first hidden layer of the plurality of cascaded hidden layers; outputting a mapping result of the first hidden layer to a subsequent cascaded hidden layer for continuing mapping processing and mapping result output in the subsequent cascaded hidden layer until outputting to a last hidden layer; taking the mapping result output by the last hidden layer as the recommendation index of the second prediction model.

11. The method of claim 10, wherein, the continuing mapping processing and mapping result output in the subsequent cascaded hidden layer comprises: performing the following processing by a jth hidden layer of the plurality of cascaded hidden layers: weighting processing on a mapping result of a (j-1) th hidden layer based on a weight of the jth hidden layer to obtain a weighted mapping result; summing processing on the weighted mapping result and a bias of the jth hidden layer to obtain a mapping result of the jth hidden layer and output the mapping result of the jth hidden layer to a (j+1) th hidden layer; wherein j is a natural number increasing and takes a value range of 2≤j≤N-1, and N is a quantity of the plurality of cascaded hidden layers.

12. An information recommendation apparatus based on artificial intelligence, characterized by comprising: the apparatus comprises: an acquisition module configured to acquire a plurality of recommendation features of a to-be-recommended task, the plurality of recommendation features comprising at least one item feature of to-be-recommended information and at least one user feature of a target user; a feature interaction module configured to perform feature projection processing on the plurality of recommendation features based on a projection space corresponding to each recommendation feature to obtain a projection feature of the projection space, wherein the feature projection processing comprises: performing scaling processing on the plurality of recommendation features based on a weight of the projection space corresponding to each recommendation feature to obtain scaled plurality of recommendation features; performing splicing processing on the scaled plurality of recommendation features to obtain an extracted feature of the projection space; and performing multiplication processing on the extracted feature of the projection space and a projection matrix of the projection space to obtain the projection feature of the projection space; performing multiplication processing on each recommendation feature and the projection feature corresponding to the projection space to obtain a feature vector of each recommendation feature; and performing mapping processing on the feature vector of each recommendation feature to obtain a feature domain corresponding to each recommendation feature; a prediction module configured to perform index prediction processing based on the feature domain corresponding to the plurality of recommendation features to obtain a recommendation index of the to-be-recommended information corresponding to the target user; a recommendation module configured to perform a to-be-recommended task based on the recommendation index of the to-be-recommended information corresponding to the target user.

13. The apparatus of claim 12, wherein, the projection space corresponding to each recommendation feature comprises a plurality of channels; and the feature interaction module is further configured to: The multiple recommendation features are subjected to feature projection processing based on any of the channels of the projection space corresponding to each of the recommendation features, to obtain the projection features of any of the channels.

14. The apparatus of claim 13, wherein, The feature interaction module is further configured to: The feature interaction module is further configured to: The feature interaction module is further configured to:

15. An electronic device, comprising: The feature interaction module is further configured to: The electronic device comprises: a memory configured to store executable instructions; 16. A computer readable storage medium characterized by: a processor configured to execute the executable instructions stored in the memory, and implement the information recommendation method based on artificial intelligence according to any one of claims 1 to 11.

17. A computer program product comprising computer instructions, characterized in that, The executable instructions are stored in the memory and configured to be executed by the processor, and implement the information recommendation method based on artificial intelligence according to any one of claims 1 to 11. The computer instructions are executed by the processor, and implement the information recommendation method based on artificial intelligence according to any one of claims 1 to 11.

Images