Explanation Method, Device, Equipment and Readable Storage Medium for Recommendation System

Abstract

An embodiment of the present application provides an explanation method, device, equipment and computer-readable storage medium for a recommendation system, including: in response to a first request, obtaining first recommended content for the first request, where the first recommended content is the recommended content expected to be obtained from the recommendation system, and performing a recommended content comparison operation: in response to sending the first request to the recommendation system, obtaining second recommended content recommended by the recommendation system according to the first request; when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, then perform an update process on the first request; repeat the steps of the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtain a first explanation for the recommendation system.

Description

Technical Field

[0001] The present application relates to the field of computer technology. Specifically, the present application relates to an explanation method, device, equipment, and computer-readable storage medium for a recommendation system. Background Art

[0002] In the prior art, the goal of a recommendation system is to recommend recommended content that a user may be interested in. However, to make users trust the recommendation system, it is necessary to explain the recommendation system, that is, to provide users with reasons for the recommendation. For example, when the recommendation system recommends a certain restaurant to a user, the reasons for the recommendation are that the environment of the restaurant is good, the dishes are delicious, and the price is cheap, etc. The explanations made for the recommendation system in the existing solutions, that is, the explanations made for the models or features in the recommendation system, cannot explain the causal relationship between user behavior (user requests) and the recommendation results (recommended content) of the recommendation system. At the same time, for a black-box recommendation system, in the case where the models and features in the recommendation system are unknown, or the models and features in the recommendation system are very complex, it is difficult for the existing solutions to explain the recommendation results of the recommendation system, that is, the causal relationship between user behavior and the recommendation results of the recommendation system cannot be explained. Summary of the Invention

[0003] In view of the shortcomings of the existing methods, the present application provides an explanation method, device, equipment, and computer-readable storage medium for a recommendation system to solve the problem of how to achieve the ability to explain the causal relationship between user requests and the recommended content of the recommendation system.

[0004] In a first aspect, the present application provides an explanation method for a recommendation system, including:

[0005] In response to a first request, obtain a first recommended content for the first request, where the first recommended content is the recommended content expected to be obtained from the recommendation system, and perform a recommended content comparison operation:

[0006] In response to sending the first request to the recommendation system, obtain a second recommended content recommended by the recommendation system according to the first request;

[0007] When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, then perform an update process on the first request;

[0008] Repeat the steps of the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtain a first explanation for the recommendation system;

[0009] The first explanation includes that within the preset number of requests, the recommendation system has the response ability to recommend the first recommended content for the first request.

[0010] In one embodiment, when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of request times is not less than a preset number of request times, a second explanation for the recommendation system is obtained; the second explanation includes that within the preset number of request times, the recommendation system does not have the response ability to recommend the first recommended content for the first request.

[0011] In one embodiment, the updating process of the first request includes:

[0012] Construct a first historical request set, where the first historical request set includes the first request;

[0013] Input each historical request in the first historical request set into a preset prediction model respectively to obtain content labels corresponding to the respective historical requests. The content labels are used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request, and the content labels include relevant information of the historical recommended content;

[0014] Determine the similarity between the historical feature vectors corresponding to the respective historical recommended contents and the first feature vector corresponding to the first recommended content, and use the historical request corresponding to the minimum similarity in the obtained similarities as the updated first request.

[0015] In one embodiment, the prediction model is trained in the following manner:

[0016] Based on a preset second historical request set, construct a training sample set; based on the training sample set, perform iterative training on an initial prediction model to be trained to obtain a trained prediction model;

[0017] During the iterative training process of the initial prediction model, perform the following processing:

[0018] Use the sample requests included in the training sample set as the input samples of the initial prediction model, and use the content labels corresponding to the sample requests as the output results of the initial prediction model. Substitute the input samples and output results into the loss function corresponding to the initial prediction model to obtain a loss function value, and update the model parameters of the initial prediction model based on the loss function value until the model parameters of the initial prediction model are updated with the prediction model parameters obtained when the loss function value meets a predetermined condition to obtain a trained prediction model.

[0019] In one embodiment, sending the first request to the recommendation system includes:

[0020] Generate an operation instruction based on a preset decision algorithm, and transmit the operation instruction to any application program;

[0021] In response to a first request generated by any application based on an operation instruction, send the first request to a recommendation system.

[0022] In one embodiment, when the similarity between a second feature vector corresponding to a second recommended content and a first feature vector corresponding to a first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, then perform an update process on the first request, including:

[0023] When the distance between a second feature vector corresponding to a second recommended content and a first feature vector corresponding to a first recommended content is greater than a preset distance, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, then perform an update process on the first request;

[0024] The distance between the second feature vector and the first feature vector is negatively correlated with the similarity.

[0025] In one embodiment, when the similarity between a second feature vector corresponding to a second recommended content and a first feature vector corresponding to a first recommended content is less than a preset similarity, and the total number of requests is not less than a preset number of requests, then obtain a second explanation for the recommendation system, including:

[0026] When the similarity between a second feature vector corresponding to a second recommended content and a first feature vector corresponding to a first recommended content is less than a preset similarity, the total number of requests is not less than a preset number of requests, and the second recommended content obtained each time in the total number of requests is the same, then obtain a second explanation. The second explanation includes determining that the first request is not a factor for the recommendation system to recommend the second recommended content, and the second recommended content includes a hot event.

[0027] In a second aspect, the present application provides an explanation device for a recommendation system, including:

[0028] A first processing module, configured to obtain a first recommended content for the first request in response to the first request, where the first recommended content is the recommended content expected to be obtained from the recommendation system;

[0029] A second processing module, configured to perform a comparison operation on the recommended content:

[0030] In response to sending the first request to the recommendation system, obtain a second recommended content recommended by the recommendation system according to the first request;

[0031] When the similarity between a second feature vector corresponding to the second recommended content and a first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, then perform an update process on the first request;

[0032] Repeat the steps of the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtain the first explanation for the recommendation system.

[0033] The first explanation includes that within the preset number of requests, the recommendation system has the ability to recommend the first recommended content for the first request.

[0034] In a third aspect, the present application provides an electronic device, including: a processor, a memory, and a bus;

[0035] The bus is used to connect the processor and the memory;

[0036] The memory is used to store operation instructions;

[0037] The processor is used to execute the explanation method for the recommendation system in the first aspect of the present application by calling the operation instructions.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, and the computer program is used to execute the explanation method for the recommendation system in the first aspect of the present application.

[0039] The technical solutions provided by the embodiments of the present application at least have the following beneficial effects:

[0040] Compare the second recommended content recommended by the recommendation system according to the first request (user request, that is, user behavior) with the first recommended content. When within the preset number of requests, the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is not less than the preset similarity, then the recommendation system has the ability to recommend the first recommended content for the first request; in this way, the following explanation can be made for the recommendation system: in a limited number of interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the user request, the user can obtain the recommended content that the user desires, and a causal relationship is formed between the user request and the recommended content of the recommendation system, that is, the user request corresponds to the generation of the recommended content. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments of the present application.

[0042] Figure 1 It is a schematic diagram of the system architecture provided by the embodiments of the present application;

[0043] Figure 2 It is a schematic flow chart of an explanation method for a recommendation system provided by the embodiments of the present application;

[0044] Figure 3A schematic flowchart of another method for explaining a recommendation system provided by an embodiment of the present application;

[0045] Figure 4 A schematic structural diagram of an apparatus for explaining a recommendation system provided by an embodiment of the present application;

[0046] Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0047] To make the objectives, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0048] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals indicate the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be construed as a limitation to the present application.

[0049] Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the", and "said" used herein may also include the plural forms. It should be further understood that the term "including" used in the specification of the present application means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more related listed items.

[0050] The embodiment of the present application provides a method for explaining a recommendation system for a recommendation system in the field of instant messaging. The method for explaining the recommendation system involves the technical field of machine learning in the field of artificial intelligence and various fields of cloud technology, such as cloud computing and cloud services in cloud technology.

[0051] Cloud computing is a computing model that distributes computing tasks across a resource pool composed of a large number of computing devices, enabling various application systems to obtain computing power, storage space, and information services as needed. The network that provides resources is called the "cloud". The resources in the "cloud" seem to users to be infinitely scalable, and can be obtained at any time, used on demand, expanded at any time, and paid according to usage.

[0052] As a basic capabilities provider of cloud computing, a cloud computing resource pool (referred to as a cloud platform, generally called an IaaS (Infrastructure as a Service) platform) will be established, and various types of virtual resources will be deployed in the resource pool for external customers to choose and use. The cloud computing resource pool mainly includes: computing devices (virtual machines, including operating systems), storage devices, and network devices.

[0053] Logically divided, the PaaS (Platform as a Service) layer can be deployed on the IaaS (Infrastructure as a Service) layer, and the SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. Or the SaaS can be directly deployed on the IaaS. PaaS is a platform for software operation, such as databases, web containers, etc. SaaS is various business software, such as web portals, mass text message senders, etc. Generally speaking, SaaS and PaaS are upper layers relative to IaaS.

[0054] The so-called artificial intelligence cloud service is generally also called AIaaS (AI as a Service). This is a current mainstream service method for artificial intelligence platforms. Specifically, the AIaaS platform will split several common AI services and provide independent or packaged services in the cloud. This service model is similar to opening an AI-themed mall: all developers can access and use one or more artificial intelligence services provided by the platform through the API interface. Some senior developers can also use the AI frameworks and AI infrastructure provided by the platform to deploy and operate their own exclusive cloud artificial intelligence services.

[0055] Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable machines to have the functions of perception, reasoning, and decision-making.

[0056] Artificial intelligence technology is an interdisciplinary subject that involves a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, autonomous driving, and intelligent transportation.

[0057] Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0058] To better understand and illustrate the solutions of the embodiments of this application, some technical terms involved in the embodiments of this application are briefly described below.

[0059] Recommendation system: The emergence and popularization of the Internet have brought a large amount of information to users, meeting the users' information needs in the information age. However, with the rapid growth of the online information volume brought about by the rapid development of the network, users are unable to obtain the truly useful part of the information from the large amount of information, and the efficiency of using information has instead decreased. This is the so-called information overload problem. A very promising way to solve the information overload problem is the recommendation system. A recommendation system is a personalized information recommendation system that recommends information, products, etc. that users are interested in to users based on the users' information needs, interests, etc. The recommendation system discovers users' interest points by studying users' interest preferences and performing personalized calculations, thereby guiding users to discover their own information needs.

[0060] Feed stream: Feed stream is information flow. In the Internet field, feed stream products such as Moments, Weibo, etc., and photo sharing websites are also a form of feed stream products. App can have a module named Dynamics, Message Square, etc., which is also a feed stream product.

[0061] Euclidean distance: Euclidean distance is a commonly used distance definition, which refers to the real distance between two points in m-dimensional space, or the natural length of a vector (that is, the distance from the point to the origin). The Euclidean distance in two-dimensional and three-dimensional space is the actual distance between two points.

[0062] Cosine similarity: Cosine similarity uses the cosine value of the angle between two vectors in the vector space as a measure of the difference between two individuals. The closer the cosine value is to 1, the closer the angle is to 0 degrees, that is, the more similar the two vectors are.

[0063] The solution provided by the embodiments of the present application involves cloud technology. The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0064] The solution provided in the embodiment of the present application can be applicable to any application scenario in the field of cloud technology that requires an explanation of the recommendation system. Through this solution, the recommendation system can be explained as follows: in a limited number of interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the user's request, and the user can obtain the predetermined recommended content. There is a causal relationship between the user request and the recommended content of the recommendation system, that is, the recommended content is generated corresponding to the user request.

[0065] In order to better understand the solution provided by the embodiment of the present application, the solution is described below in conjunction with a specific application scenario.

[0066] In one embodiment, Figure 1 A schematic diagram of a system architecture for explaining a recommendation system applicable to an embodiment of the present application is shown in FIG. It can be understood that the explanation method for the recommendation system provided in the embodiment of the present application can be applicable to but not limited to the following applications: Figure 1 In the application scenario shown.

[0067] In this example, Figure 1As shown, the system architecture for the explanation of the recommendation system in this example may include, but is not limited to, a recommendation adversarial explanation system 101 and a recommendation system 102, and the recommendation adversarial explanation system 101 and the recommendation system 102 may interact via a network. The recommendation adversarial explanation system 101 includes an adversarial explanation decision engine 1011 and terminals 1012. Figure 1 The agent-1, agent-2, … agent-N shown in Figure 1 are all terminals 1012. The adversarial explanation decision engine 1011 generates user instructions (operation instructions) according to a preset decision algorithm and sends the user instructions to the terminals 1012; the user instructions are, for example, refresh, like, comment, etc. The terminals 1012 receive the user instructions sent by the adversarial explanation decision engine 1011, and the terminals 1012 send user requests to the recommendation system 102 according to the user instructions. The recommendation system 102 sends recommendation results (recommended content) to the terminals 1012 according to the received user requests. The terminals 1012 send the received recommendation results to the adversarial explanation decision engine 1011. The adversarial explanation decision engine 1011 receives the recommendation results sent by each terminal 1012 and stores the corresponding user instructions and environmental data.

[0068] It can be understood that the above is only an example, and this embodiment is not limited here.

[0069] Among them, the terminals 1012 can be smartphones (such as Android phones, iOS phones, etc.), mobile phone emulators, tablets, laptops, digital broadcast receivers, MIDs (Mobile Internet Devices), PDAs (Personal Digital Assistants), desktop computers, in-vehicle terminals (such as in-vehicle navigation terminals), smart speakers, smart watches, etc. The adversarial explanation decision engine 1011 or the recommendation system 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server or server cluster that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDNs (Content Delivery Networks), and big data and artificial intelligence platforms. The above network may include, but is not limited to: wired networks, wireless networks, where the wired network includes: local area networks, metropolitan area networks, and wide area networks, and the wireless network includes: Bluetooth, Wi-Fi, and other networks that implement wireless communication. Specifically, it can also be determined based on the actual application scenario requirements and is not limited here.

[0070] See Figure 2 , Figure 2The figure shows a schematic flowchart of an explanation method for a recommendation system provided by an embodiment of the present application. Among them, this method can be executed by any electronic device, such as a recommendation adversarial explanation system. As an alternative embodiment, this method can be executed by a recommendation adversarial explanation system. For the sake of convenient description, in the description of some alternative embodiments below, the recommendation adversarial explanation system will be taken as an example of the execution subject of this method. As Figure 2 shown, the explanation method for the recommendation system provided by the embodiment of the present application includes the following steps:

[0071] S101, in response to a first request, obtain first recommended content for the first request, where the first recommended content is the recommended content expected to be obtained from the recommendation system.

[0072] In one embodiment, the first recommended content can be text, picture, video, audio, etc. For example, the first recommended content is a short video related to a game, and the feature vector corresponding to the short video of the game is [1.0, 2.1, 2.3, 1.2, 5.4].

[0073] In one embodiment, the first request is a user request used to represent user behavior. For example, the user likes a short video of a target object in a game. The first request is used to request the first recommended content from the recommendation system. The first recommended content can be preset and used to evaluate whether the recommendation system has the response ability to recommend the first recommended content for the first request.

[0074] S102, perform a recommended content comparison operation:

[0075] In response to sending the first request to the recommendation system, obtain second recommended content recommended by the recommendation system according to the first request;

[0076] When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, update the first request;

[0077] Repeat the steps of the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtain a first explanation for the recommendation system;

[0078] The first explanation includes that within the preset number of requests, the recommendation system has the response ability to recommend the first recommended content for the first request.

[0079] In one embodiment, the initial value of the total number of requests n for the recommendation adversarial explanation system to request recommended content from the recommendation system is m, and m is a positive integer.

[0080] For example, it is recommended that the adversarial interpretation system sends a first request to the recommendation system on a certain occasion. If the similarity between the second feature vector Y corresponding to the second recommended content and the first feature vector X corresponding to the first recommended content is less than a preset similarity, the total number of requests n is incremented by 1, that is, n is updated to n + 1. For example, when the adversarial interpretation system sends the first request to the recommendation system for the first time and m is 0, n is updated from the initial value of 0 to 1.

[0081] In one embodiment, the difference between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content, that is, the distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content, such as the Euclidean distance, can be used to measure the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content. The smaller the distance between the second feature vector and the first feature vector, the greater the similarity between the second feature vector and the first feature vector.

[0082] In one embodiment, the cosine similarity is used to measure the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content, that is, the cosine value of the angle between the second feature vector and the first feature vector, which can be used to measure the similarity between the second feature vector and the first feature vector. The closer the cosine value is to 1, the closer the angle is to 0 degrees, that is, the more similar the two vectors are, and the greater the similarity between the second feature vector and the first feature vector.

[0083] For example, the first request is a user request used to characterize user behavior. The user behavior is that the user comments on a video of a giant panda. When the user comments on the video of the giant panda for the first time, that is, the recommendation adversarial explanation system sends the first request to the recommendation system. The recommended content returned by the recommendation system is a video of a bear. Among them, the first recommended content is a video of a giant panda, and the second recommended content is a video of a bear. The similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, and the total number of requests 1 for requesting recommended content from the recommendation system is less than the preset number of requests 5. Then, the recommendation adversarial explanation system updates the first request to obtain the updated first request. When the user comments on the video of the giant panda for the second time, that is, the recommendation adversarial explanation system sends the updated first request to the recommendation system for the second time. The recommended content returned by the recommendation system is a video of a giant panda. Among them, the first recommended content is a video of a giant panda, and the second recommended content is a video of a giant panda. The similarity between the second feature vector and the first feature vector is not less than the preset similarity, and the first explanation for the recommendation system is obtained. The first explanation includes that within the preset number of requests 5, the recommendation system has the ability to recommend a video of a giant panda in response to the user's comment on the video of the giant panda. Thus, it can be shown that: in the limited interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the comment on the video of the giant panda. The user can obtain the video of the giant panda that the user desires. A causal relationship is formed between the user request (the user's comment on the video of the giant panda) and the recommended content of the recommendation system (the video of the giant panda), that is, the user's comment on the video of the giant panda causes the recommendation system to recommend the video of the giant panda.

[0084] In the embodiments of the present application, the following explanation can be made for the recommendation system: in the limited interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the user request. The user can obtain the recommended content that the user desires. A causal relationship is formed between the user request and the recommended content of the recommendation system, that is, the user request corresponding to generates the recommended content.

[0085] In one embodiment, when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, and the total number of requests is not less than the preset number of requests, the second explanation for the recommendation system is obtained. The second explanation includes that within the preset number of requests, the recommendation system does not have the ability to recommend the first recommended content in response to the first request.

[0086] For example, the first request is a user request used to represent a user behavior, where the user behavior is that the user listens to the songs of singer B in album A. When the user listens to the songs of singer B for the first time, that is, the recommendation adversarial explanation system sends the first request to the recommendation system, and the recommended content returned by the recommendation system is the songs of other singers in album A except singer B. Among them, the first recommended content is the songs of singer B, and the second recommended content is the songs of other singers in album A. The similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, and the total number of requests 1 for requesting recommended content from the recommendation system is less than the preset number of requests 10. Then, the recommendation adversarial explanation system performs an update process on the first request to obtain the updated first request. When the user listens to the songs of singer B multiple times, that is, the recommendation adversarial explanation system sends the updated first request to the recommendation system multiple times, and the recommended content returned by the recommendation system is always the songs of other singers in album A. Among them, the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, and the total number of requests 10 is not less than the preset number of requests 10. Then, the second explanation for the recommendation system is obtained; the second explanation includes that within the preset number of requests 10, the recommendation system does not have the ability to recommend the songs of singer B for the user's listening to the songs of singer B in album A. Thus, it can be shown that: in a limited number of interactions between the user and the recommendation system, the recommendation system cannot respond quickly to the listening of the songs of singer B, the user cannot get the songs of singer B that they want to listen to, and there is no causal relationship between the user's request (listening to the songs of singer B) and the recommended content of the recommendation system (the songs of other singers in album A), that is, the user's listening to the songs of singer B will not cause the recommendation system to recommend the songs of singer B.

[0087] In one embodiment, the prediction model is trained in the following manner:

[0088] Based on a preset second historical request set, a training sample set is constructed; based on the training sample set, the initial prediction model to be trained is iteratively trained to obtain the trained prediction model;

[0089] During the iterative training of the initial prediction model, the following processing is performed:

[0090] Taking the sample requests included in the training sample set as the input samples of the initial prediction model, and taking the content labels corresponding to the sample requests as the output results of the initial prediction model, substituting the input samples and the output results into the loss function corresponding to the initial prediction model to obtain the loss function value, and updating the model parameters of the initial prediction model based on the loss function value until the model parameters of the initial prediction model are updated with the prediction model parameters obtained when the loss function value satisfies the predetermined condition, so as to obtain the trained prediction model.

[0091] In one embodiment, the user opening request is constructed by the adversarial interpretation decision engine 1011 in Figure 1 . The user opening request can be a recommendation function that needs to be explained, such as the QQ Miniworld square page. The adversarial interpretation decision engine 1011 also saves key information such as device information and behavior operation time. Among them, the device information includes that the operating system of the terminal 1012 is the Android system, etc., and the operation time includes the time point of the user operation (generating an operation instruction), etc. The adversarial interpretation decision engine 1011 generates an operation instruction (user instruction) based on a preset decision algorithm, and transmits the operation instruction to any application app on the terminal 1012, such as the app of the QQ Miniworld square page. In response to the first request generated by any application app based on the operation instruction, the terminal 1012 sends the first request to the recommendation system 102. The recommendation system 102 returns recommended content to the terminal 1012. The recommended content includes, for example, short videos, Feed streams, etc. Among them, the short video can be a corresponding short video displayed in the form of a video stream of a spatial video, such as a football video. The terminal 1012 transmits the recommended content to the adversarial interpretation decision engine 1011, and the adversarial interpretation decision engine 1011 records the recommended content and models the relationship between user behavior and recommended content, that is, content = F(behavior); where content represents a content label, and the content label is used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request. The content label includes relevant information of the historical recommended content; behavior represents the historical user request, that is, the historical request, such as liking a beautiful woman 3 times, etc.; the prediction model F can be used to predict: under the specific behavior, which content the recommendation system will most likely recommend. The prediction model F can be obtained through a multi-class machine learning method.

[0092] In one embodiment, the update process of the first request includes steps A1 - A3:

[0093] Step A1, construct a first historical request set, and the first historical request set includes the first request.

[0094] Step A2, input each historical request in the first historical request set into a preset prediction model respectively to obtain the content label corresponding to each historical request. The content label is used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request, and the content label includes relevant information of the historical recommended content.

[0095] Step A3, determine the similarity between the historical feature vector corresponding to each historical recommended content and the first feature vector corresponding to the first recommended content, and use the historical request corresponding to the minimum similarity among the obtained similarities as the updated first request.

[0096] In one embodiment, a user request (the first request) is generated according to a user behavior generation algorithm, that is, the first request is updated according to the user behavior generation algorithm. The goal of the user behavior generation algorithm is: under the generated user request (such as the historical requests in the first historical request set), the prediction model estimates the user behavior that minimizes the difference between the vector corresponding to the next recommended content and the given target content vector (the first feature vector corresponding to the first recommended content), that is, use content = F(behavior) to inversely solve for behavior. Specifically, all combinations of behaviors are traversed to obtain the corresponding content and the second feature vector X corresponding to the content, and the combination of behaviors with the smallest Euclidean distance between the second feature vector X and the first feature vector Y is determined. The user behavior generation algorithm is based on Figure 1 the user behaviors of all terminals 1012 and the corresponding recommendation results in it, and can correct the prediction model most quickly and give user behaviors that can be explored efficiently.

[0097] In one embodiment, sending the first request to the recommendation system includes:

[0098] Generating an operation instruction based on a preset decision algorithm and transmitting the operation instruction to any application;

[0099] In response to the first request generated by any application based on the operation instruction, sending the first request to the recommendation system.

[0100] For example, the operation instruction can be refresh, like, comment, etc., and the application can be an APP application installed on the terminal.

[0101] In one embodiment, when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, and the total number of requests for recommended content to the recommendation system is less than the preset number of requests, then the first request is updated, including:

[0102] When the distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is greater than the preset distance, and the total number of requests for recommended content to the recommendation system is less than the preset number of requests, then the first request is updated;

[0103] The distance between the second feature vector and the first feature vector is negatively correlated with the similarity.

[0104] In one embodiment, the distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is the Euclidean distance, which can be used to measure the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content. The smaller the distance between the second feature vector and the first feature vector, the greater the similarity between the second feature vector and the first feature vector, that is, the distance between the second feature vector and the first feature vector is negatively correlated with the similarity. The distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is greater than a preset distance, that is, the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity.

[0105] In one embodiment, when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests is not less than a preset number of requests, then a second explanation for the recommendation system is obtained, including:

[0106] When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, the total number of requests is not less than a preset number of requests, and the second recommended content obtained each time in the total number of requests is the same, then a second explanation is obtained. The second explanation includes determining that the first request is not a factor constituting the recommendation system's recommendation of the second recommended content, and the second recommended content includes a hot event.

[0107] For example, the first request is a user request, which is used to characterize user behavior. The user behavior is that the user likes a picture of a certain scenery. Every time the user likes a picture of a certain scenery, the recommendation adversarial interpretation system sends a first request to the recommendation system, and the recommended content returned by the recommendation system is a picture of a scene in the National Games being held. Among them, the first recommended content is a picture of a certain scenery, and the second recommended content is a picture of a scene in the National Games being held. The similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, the total number of requests is not less than the preset number of requests, and the second recommended content obtained each time in the total number of requests is the same, then the second interpretation is obtained, and the second interpretation includes determining that liking a picture of a certain scenery is not a factor in the recommendation system recommending a picture of a scene in the National Games being held, and the National Games being held is a network hot spot, that is, a hot event. In the limited interactions between users and the recommendation system, the recommendation system cannot respond quickly to likes on pictures of certain scenery, and users cannot get the pictures of certain scenery they want. There is no causal relationship between the user's request (liking a picture of a certain scenery) and the recommendation system's recommendation content (pictures of scenes from the National Games currently being held), that is, liking a picture of a certain scenery will not cause the recommendation system to recommend a corresponding picture of the certain scenery.

[0108] In the embodiment of the present application, the recommendation system can be explained as follows: in a limited number of interactions between the user and the recommendation system, the recommendation system cannot respond quickly to the user request, the user cannot obtain the predetermined recommended content, and there is no causal relationship between the user request and the recommended content of the recommendation system, that is, the user request cannot generate the corresponding recommended content.

[0109] In order to better understand the method provided in the embodiment of the present application, the scheme of the embodiment of the present application is further explained below with reference to examples of specific application scenarios.

[0110] See also Figure 3 , Figure 3 The flowchart of an explanation method for a recommendation system provided by an embodiment of the present application is shown, wherein the method can be executed by any electronic device, such as a recommendation confrontation explanation system. As an optional implementation, the method can be executed by a recommendation confrontation explanation system. For the convenience of description, in the description of some optional embodiments below, the recommendation confrontation explanation system will be used as an example to illustrate the execution subject of the method. Figure 3 As shown, the interpretation method for the recommendation system provided in the embodiment of the present application includes the following steps:

[0111] S201, setting an expected content expression vector Y, and assigning an initial value of the number of loops n to 0.

[0112] In one embodiment, the expected content expression vector Y can be the first feature vector corresponding to the first recommended content, where the first recommended content is the recommended content expected to be obtained from the recommendation system. The first recommended content can be text, picture, video, audio, etc. For example, the first recommended content is a short video about playing Chicken, and the content expression vector Y corresponding to the short video about playing Chicken is [1.0, 2.1, 2.3, 1.2, 5.4]. The number of loop times n can be the total number of requests n for requesting recommended content from the recommendation system.

[0113] S202: Send a user request to the recommendation system and receive the content expression vector X returned by the recommendation system.

[0114] In one embodiment, through Figure 1 the adversarial interpretation decision engine 1011 constructs a user opening request, which can be a recommended function to be explained, such as the QQ Miniworld square page. The adversarial interpretation decision engine 1011 also saves key information such as device information and behavior operation time. Among them, the device information includes that the operating system of the terminal 1012 is the Android system, etc., and the operation time includes the time point of the user operation (generating an operation instruction), etc. The adversarial interpretation decision engine 1011 generates an operation instruction (user instruction) based on a preset decision algorithm, and transmits the operation instruction to any application app on the terminal 1012, such as the app of the QQ Miniworld square page. In response to the first request (user request) generated by any application app based on the operation instruction, the terminal 1012 sends the first request to the recommendation system 102. The recommendation system 102 returns recommended content to the terminal 1012. The recommended content can be, for example, short videos, Feed streams, etc. Among them, the short video can be the corresponding short video displayed in the form of a video stream of a spatial video, such as a football video. The terminal 1012 transmits the recommended content to the adversarial interpretation decision engine 1011, and the adversarial interpretation decision engine 1011 records the recommended content and models the relationship between user behavior and recommended content, that is, content = F(behavior); where content represents a content label, and the content label is used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request. The content label includes relevant information of the historical recommended content; behavior represents the historical user request, that is, the historical request, such as liking a handsome guy 5 times, commenting on a football event 11 times, etc.; the prediction model F can be used to predict: under the specific behavior, which content the recommendation system will most probably recommend. The content label is input into an external system, such as a content understanding system, to obtain the corresponding recommended content, and according to the corresponding recommended content, the content expression vector X is obtained, where the content expression vector X can be the second feature vector corresponding to the second recommended content.

[0115] S203, determine whether the distance between the content expression vector X and the content expression vector Y is less than a preset first threshold. When it is determined that the distance between the content expression vector X and the content expression vector Y is less than the preset first threshold, proceed to S204 for processing; when it is determined that the distance between the content expression vector X and the content expression vector Y is not less than the preset first threshold, proceed to S205 for processing.

[0116] In one embodiment, the difference between the content expression vector X and the content expression vector Y, that is, the distance between the content expression vector X and the content expression vector Y, such as the Euclidean distance, the smaller the distance between the content expression vector X and the content expression vector Y, the greater the similarity between the content expression vector X and the content expression vector Y.

[0117] S204, output the sequence length n of the user operation, and output the corresponding user operation sequence and the recommended content sequence.

[0118] In one embodiment, the sequence length n of the user operation can be the n that records the number of loop iterations, that is, the total number of requests for recommended content to the recommendation system. The user operation sequence includes each user request sent, and the recommended content sequence includes the recommended content returned by the recommendation system for each user request.

[0119] S205, determine whether the current sequence length n is less than a second threshold. When it is determined that the current sequence length n is less than the second threshold, proceed to S206 for processing; when it is determined that the current sequence length n is not less than the second threshold, proceed to S207 for processing.

[0120] In one embodiment, the current sequence length n can be the n that records the number of loop iterations, that is, the total number of requests for recommended content to the recommendation system, and the second threshold can be a preset number of requests.

[0121] S206, generate a user request according to the user behavior generation algorithm, and at the same time assign the loop count n to n + 1.

[0122] In one embodiment, a user request (the first request) is generated according to a user behavior generation algorithm, that is, the first request is updated according to the user behavior generation algorithm. The goal of the user behavior generation algorithm is: under the generated user request (such as the historical requests in the first historical request set), the prediction model estimates the user behavior that minimizes the difference between the vector corresponding to the next recommended content and the given target content vector (the first feature vector corresponding to the first recommended content), that is, use content = F(behavior) to inversely find behavior. Specifically, all combinations of behavior are traversed to obtain the corresponding content and the content expression vector X corresponding to the content, and the combination of behavior with the smallest Euclidean distance between the content expression vector X and the content expression vector Y is determined. The user behavior generation algorithm is based on Figure 1 the user behaviors and corresponding recommendation results of all terminals 1012 in it, and can correct the prediction model most quickly and give user behaviors that can be explored efficiently.

[0123] S207, output the sequence length n of the user operations, and output the corresponding user operation sequence, recommended content sequence, and the minimum vector difference of the historical recommendation results.

[0124] In one embodiment, the sequence length n of the user operations can be n that records the number of loop times, that is, the total number of requests for recommended content to the recommendation system. The user operation sequence includes each user request sent, and the recommended content sequence includes the recommended content returned by the recommendation system for each user request. The vector differences between the historical feature vectors corresponding to each historical recommended content and the content expression vector Y are determined, and the smallest vector difference among the obtained vector differences is determined as the minimum vector difference of the historical recommendation result (historical recommended content).

[0125] S208, obtain an explanation for the recommendation system.

[0126] In one embodiment, through S201 - S207, the user requests and corresponding recommendation results (recommended content) are obtained. When n is greater than or equal to the second threshold, it indicates that the user cannot obtain the predetermined desired content in a limited number of behaviors. The predetermined desired content corresponds to the expected content expression vector Y, indicating that the recommendation system does not have an obvious response ability to user behaviors. When n is less than the second threshold, it indicates that the recommendation system can obtain the predetermined desired content in a limited number of user interactions. The predetermined desired content corresponds to the expected content expression vector Y, indicating that the recommendation system has a fast response ability to user behaviors. At the same time, output content = F(behavior), and F will be used as the prediction model of this recommendation system.

[0127] In the embodiments of the present application, the explanation method for the recommendation system is used in the existing recommendation system and the investigated external recommendation system, and the quantitative evaluation and analysis results of the existing recommendation system and the investigated external recommendation system are given, which is of great help to the evaluation, optimization and investigation of the existing recommendation system and the investigated external recommendation system. By means of adversarial learning, it is obtained whether the recommendation system has a causal relationship with the user behavior, the causal relationship is quantitatively modeled, and based on the causal relationship obtained from the modeling, it is determined what kind of behavior sequence will produce what kind of recommendation results. At the same time, the recommendation reasons not constituted by the user behavior are also given. Thus, it is possible to understand how the user behavior and the context environment affect the recommendation results for the recommendation system, and the expected recommendation results can be obtained by controlling the user behavior.

[0128] Based on the same inventive concept, the embodiments of the present application further provide an explanation device for a recommendation system. The structural schematic diagram of the device is as Figure 4 shown. The explanation device 40 for the recommendation system includes a first processing module 401 and a second processing module 402.

[0129] The first processing module 401 is configured to, in response to a first request, obtain a first recommended content for the first request, where the first recommended content is the recommended content expected to be obtained from the recommendation system.

[0130] The second processing module 402 is configured to perform a recommended content comparison operation:

[0131] In response to sending the first request to the recommendation system, obtain a second recommended content recommended by the recommendation system according to the first request;

[0132] When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, the first request is updated;

[0133] Repeat the steps of performing the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtain a first explanation for the recommendation system;

[0134] The first explanation includes that within the preset number of requests, the recommendation system has the response ability to recommend the first recommended content for the first request.

[0135] In one embodiment, the second processing module 402 is further configured to:

[0136] When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than the preset similarity, and the total number of requests is not less than the preset number of requests, a second explanation for the recommendation system is obtained; the second explanation includes that within the preset number of requests, the recommendation system does not have the ability to recommend the first recommended content in response to the first request.

[0137] In one embodiment, the second processing module 402 is specifically configured to:

[0138] Construct a first historical request set, where the first historical request set includes the first request;

[0139] Input each historical request in the first historical request set into a preset prediction model respectively to obtain content labels corresponding to the respective historical requests. The content labels are used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request, and the content labels include the relevant information of the historical recommended content;

[0140] Determine the similarity between the historical feature vectors corresponding to the respective historical recommended contents and the first feature vector corresponding to the first recommended content, and use the historical request corresponding to the minimum similarity among the obtained similarities as the updated first request.

[0141] In one embodiment, the second processing module 402 is specifically configured to:

[0142] Based on a preset second historical request set, construct a training sample set; based on the training sample set, perform iterative training on the initial prediction model to be trained to obtain a trained prediction model;

[0143] Perform the following processing during the iterative training of the initial prediction model:

[0144] Use the sample requests included in the training sample set as the input samples of the initial prediction model, and use the content labels corresponding to the sample requests as the output results of the initial prediction model. Substitute the input samples and output results into the loss function corresponding to the initial prediction model to obtain a loss function value, and perform an update process on the model parameters of the initial prediction model based on the loss function value until the model parameters of the initial prediction model are updated with the prediction model parameters obtained when the loss function value meets the predetermined condition, to obtain a trained prediction model.

[0145] In one embodiment, the second processing module 402 is specifically configured to:

[0146] Generate an operation instruction based on a preset decision algorithm, and transmit the operation instruction to any application program;

[0147] In response to a first request generated by any application program based on the operation instruction, send the first request to the recommendation system.

[0148] In one embodiment, when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests to the recommendation system for recommended content is less than a preset number of requests, the second processing module 402 is specifically configured to:

[0149] When the distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is greater than a preset distance, and the total number of requests to the recommendation system for recommended content is less than a preset number of requests, updating the first request;

[0150] The distance between the second eigenvector and the first eigenvector is negatively correlated with the similarity.

[0151] In one embodiment, when the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests is not less than a preset number of requests, the second processing module 402 is specifically configured to:

[0152] When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, the total number of requests is not less than a preset number of requests, and the second recommended content obtained each time in the total number of requests is the same, a second explanation is obtained, and the second explanation includes determining that the first request is not a factor in the recommendation system recommending the second recommended content, and the second recommended content includes a hot event.

[0153] The application of the embodiments of the present application has at least the following beneficial effects:

[0154] The second recommended content recommended by the recommendation system according to the first request (user request, i.e., user behavior) is compared with the first recommended content. When, within a preset number of requests, the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is not less than the preset similarity, the recommendation system has the response capability to recommend the first recommended content for the first request. In this way, the recommendation system can be explained as follows: in a limited number of interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the user request, the user can get the predetermined recommended content, and there is a causal relationship between the user request and the recommended content of the recommendation system, i.e., the recommended content is generated corresponding to the user request.

[0155] Based on the same inventive concept, the present application also provides an electronic device, the structural diagram of which is as follows: Figure 5As shown, the electronic device 9000 includes at least one processor 9001, a memory 9002, and a bus 9003. At least one processor 9001 is electrically connected to the memory 9002. The memory 9002 is configured to store at least one computer-executable instruction, and the processor 9001 is configured to execute the at least one computer-executable instruction, so as to perform the steps of any one of the explanation methods for a recommendation system provided in any embodiment or any alternative implementation manner of the present application.

[0156] Furthermore, the processor 9001 may be an FPGA (Field-Programmable Gate Array) or other devices with logical processing capabilities, such as an MCU (Microcontroller Unit) or a CPU (Central Processing Unit).

[0157] Applying the embodiments of the present application has at least the following beneficial effects:

[0158] Compare the second recommended content recommended by the recommendation system according to the first request (user request, i.e., user behavior) with the first recommended content. When within a preset number of request times, the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is not less than the preset similarity, then the recommendation system has the response ability to recommend the first recommended content for the first request. In this way, the following explanation can be made for the recommendation system: In a limited number of interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the user request, and the user can obtain the recommended content that the user desires. A causal relationship is formed between the user request and the recommended content of the recommendation system, that is, the user request corresponds to the generation of the recommended content.

[0159] Based on the same inventive concept, the embodiments of the present application also provide a computer-readable storage medium storing a computer program, which is used to implement the steps of any one of the explanation methods for a recommendation system provided in any embodiment or any alternative implementation manner of the present application when executed by a processor.

[0160] The computer-readable storage medium provided in the embodiments of the present application includes, but is not limited to, any type of disk (including floppy disk, hard disk, optical disk, CD-ROM, and magneto-optical disk), ROM (Read-Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, magnetic card or optical card. That is, the readable storage medium includes any medium that can store or transmit information in a readable form by a device (e.g., a computer).

[0161] The application of the embodiments of the present application has at least the following beneficial effects:

[0162] The second recommended content recommended by the recommendation system according to the first request (user request, i.e., user behavior) is compared with the first recommended content. When, within a preset number of requests, the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is not less than the preset similarity, the recommendation system has the response capability to recommend the first recommended content for the first request. In this way, the recommendation system can be explained as follows: in a limited number of interactions between the user and the recommendation system, the recommendation system can respond relatively quickly to the user request, the user can get the predetermined recommended content, and there is a causal relationship between the user request and the recommended content of the recommendation system, i.e., the recommended content is generated corresponding to the user request.

[0163] The embodiment of the present application also provides a computer program product including instructions, which, when executed on a computer device, enables the computer device to execute the interpretation method for the recommendation system provided by the above-mentioned various method embodiments.

[0164] Those skilled in the art will appreciate that each block in these structure diagrams and / or block diagrams and / or flow charts and combinations of blocks in these structure diagrams and / or block diagrams and / or flow charts can be implemented using computer programs. Those skilled in the art will appreciate that these computer program products can be provided to a general-purpose computer, a professional computer, or a processor of other programmable data processing methods to implement, thereby executing the schemes specified in the blocks or multiple blocks of the structure diagrams and / or block diagrams and / or flow charts disclosed in this application through the processor of the computer or other programmable data processing method.

[0165] Those skilled in the art can understand that the various operations, methods, steps, measures, and solutions in the processes discussed in this application can be alternated, changed, combined, or deleted. Further, other steps, measures, and solutions in the various operations, methods, and processes discussed in this application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, those in the prior art that have steps, measures, and solutions in the various operations, methods, and processes disclosed in this application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0166] The above are only some embodiments of this application. It should be noted that for those of ordinary skill in the art, without departing from the principle of this application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. An explanation method for a recommendation system, characterized in that, Including: In response to a first request, obtain first recommended content for the first request, where the first recommended content is the recommended content expected to be obtained from a recommendation system, and perform a recommended content comparison operation: In response to sending the first request to the recommendation system, obtain second recommended content recommended by the recommendation system according to the first request; When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, update the first request; Repeat the steps of the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtain a first explanation for the recommendation system; The first explanation includes that within the preset number of requests, the recommendation system has the ability to recommend the first recommended content for the first request; The updating process of the first request includes: Construct a first historical request set, where the first historical request set includes the first request; Input each historical request in the first historical request set into a preset prediction model respectively, and obtain content labels corresponding to the respective historical requests. The content labels are used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request, and the content labels include relevant information of the historical recommended content; Determine the similarity between the historical feature vectors corresponding to the respective historical recommended contents and the first feature vector corresponding to the first recommended content, and use the historical request corresponding to the minimum similarity among the obtained similarities as the updated first request.

2. The method according to claim 1, wherein Also including: When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests is not less than the preset number of requests, obtain a second explanation for the recommendation system; The second explanation includes that within the preset number of requests, the recommendation system does not have the ability to recommend the first recommended content for the first request.

3. The method according to claim 1, wherein The prediction model is trained through the following method: Based on a preset second historical request set, construct a training sample set; based on the training sample set, perform iterative training on an initial prediction model to be trained to obtain a trained prediction model; During the iterative training process of the initial prediction model, perform the following processing: Use the sample requests included in the training sample set as the input samples of the initial prediction model, and use the content labels corresponding to the sample requests as the output results of the initial prediction model. Substitute the input samples and the output results into the loss function corresponding to the initial prediction model to obtain a loss function value, and update the model parameters of the initial prediction model based on the loss function value until the model parameters of the initial prediction model are updated with the prediction model parameters obtained when the loss function value meets a predetermined condition, and a trained prediction model is obtained.

4. The method according to claim 1, characterized in that Sending the first request to the recommendation system includes: Generating an operation instruction based on a preset decision algorithm and transmitting the operation instruction to any application; In response to a first request generated by any application based on the operation instruction, sending the first request to the recommendation system.

5. The method according to claim 1, characterized in that When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for recommended content sent to the recommendation system is less than a preset number of requests, updating the first request includes: When the distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is greater than a preset distance, and the total number of requests for recommended content sent to the recommendation system is less than a preset number of requests, updating the first request; The distance between the second feature vector and the first feature vector is negatively correlated with the similarity.

6. The method according to claim 2, wherein When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests is not less than a preset number of requests, obtaining a second explanation for the recommendation system includes: When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, the total number of requests is not less than a preset number of requests, and the second recommended content obtained each time in the total number of requests is the same, obtaining the second explanation, where the second explanation includes determining that the first request is not a factor for the recommendation system to recommend the second recommended content, and the second recommended content includes a hot event.

7. An explanation device for a recommendation system, characterized in that Including: A first processing module, configured to obtain a first recommended content for the first request in response to the first request, where the first recommended content is the recommended content expected to be obtained from the recommendation system; A second processing module, configured to perform a recommended content comparison operation: In response to sending the first request to the recommendation system, obtaining a second recommended content recommended by the recommendation system according to the first request; When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests for recommended content sent to the recommendation system is less than a preset number of requests, updating the first request; Repeatedly executing the steps of the recommended content comparison operation until the similarity between the second feature vector and the first feature vector is not less than a preset similarity, obtaining a first explanation for the recommendation system; The first explanation includes that within the preset number of requests, the recommendation system has the ability to recommend the first recommended content for the first request; When the second processing module updates the first request, it is specifically configured to: Construct a first historical request set, where the first historical request set includes the first request; Input each historical request in the first historical request set into a preset prediction model respectively to obtain a content label corresponding to each historical request, where the content label is used to identify the historical recommended content recommended by the recommendation system in response to the corresponding historical request, and the content label includes relevant information of the historical recommended content; Determine the similarity between the historical feature vector corresponding to each historical recommended content and the first feature vector corresponding to the first recommended content, and use the historical request corresponding to the minimum similarity among the obtained similarities as the updated first request.

8. The device according to claim 7, characterized in that, The second processing module is further configured to: When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, and the total number of requests is not less than a preset number of requests, obtain a second explanation for the recommendation system; The second explanation includes that within the preset number of requests, the recommendation system does not have the ability to recommend the first recommended content in response to the first request.

9. The device according to claim 7, characterized in that, The prediction model is trained through the following method: construct a training sample set based on a preset second historical request set; perform iterative training on an initial prediction model to be trained based on the training sample set to obtain a trained prediction model; During the iterative training process of the initial prediction model, perform the following processing: Use the sample requests included in the training sample set as the input samples of the initial prediction model, and use the content labels corresponding to the sample requests as the output results of the initial prediction model. Substitute the input samples and the output results into the loss function corresponding to the initial prediction model to obtain a loss function value, and update the model parameters of the initial prediction model based on the loss function value until the model parameters of the initial prediction model are updated with the prediction model parameters obtained when the loss function value meets a predetermined condition to obtain a trained prediction model.

10. The device according to claim 7, characterized in that, The second processing module is specifically configured to: Generate an operation instruction based on a preset decision algorithm, and transmit the operation instruction to any application program; In response to the first request generated by any application program based on the operation instruction, send the first request to the recommendation system.

11. The device according to claim 7, characterized in that, The second processing module is specifically configured to: When the distance between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is greater than a preset distance, and the total number of requests for requesting recommended content from the recommendation system is less than a preset number of requests, perform an update process on the first request; The distance between the second feature vector and the first feature vector is negatively correlated with the similarity.

12. The device according to claim 8, characterized in that The second processing module is specifically configured to: When the similarity between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content is less than a preset similarity, the total number of requests is not less than a preset number of requests, and each obtained second recommended content in the total number of requests is the same, then the second explanation is obtained. The second explanation includes determining that the first request is not a factor for the recommendation system to recommend the second recommended content, and the second recommended content includes a hot event.

13. An electronic device, characterized in that, Comprising: a processor and a memory; the memory is used for storing a computer program; the processor is configured to execute the explanation method for a recommendation system according to any one of claims 1-6 by calling the computer program.

14. A computer-readable storage medium, characterized in that, A computer program is stored, and the computer program is used to implement the explanation method for a recommendation system according to any one of claims 1-6 when being executed by a processor.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the explanation method for a recommendation system according to any one of claims 1-6.

Images