CN113822429A

CN113822429A - Interpretation method, device and equipment for recommendation system and readable storage medium

Info

Publication number: CN113822429A
Application number: CN202110874423.6A
Authority: CN
Inventors: 张逾
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-07-30
Filing date: 2021-07-30
Publication date: 2021-12-21

Abstract

The embodiment of the application provides an interpretation method, an interpretation device, interpretation equipment and a computer-readable storage medium for a recommendation system, wherein the interpretation method comprises the following steps: responding to the first request, obtaining first recommended content aiming at the first request, wherein the first recommended content is recommended content expected to be obtained from a recommending system, and performing recommended content comparison operation: responding to the first request sent to the recommendation system, and obtaining second recommendation content recommended by the recommendation system according to the first request; 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 smaller than a preset similarity, and the total request times of requesting the recommended content from the recommendation system are smaller than the preset request times, updating the first request; and repeating the step of comparing the recommended content until the similarity between the second characteristic vector and the first characteristic vector is not less than the preset similarity, and obtaining a first explanation aiming at the recommendation system.

Description

Interpretation method, device and equipment for recommendation system and readable storage medium

Technical Field

The present application relates to the field of computer technologies, and in particular, to an interpretation method, an interpretation apparatus, an interpretation device, and a computer-readable storage medium for a recommendation system.

Background

The goal of prior art recommendation systems is to recommend recommended content to a user that may be of interest. However, to let the user trust the recommendation system, it needs to be interpreted with respect to the recommendation system, that is, to provide the recommendation reason for the user, for example, the recommendation system recommends a restaurant to the user, where the restaurant is good in environment, delicious in dishes, and low in price. The existing scheme aims at the explanation of the recommendation system, namely the explanation of the model or the characteristics in the recommendation system, and the causal relationship between the user behavior (user request) and the recommendation result (recommendation content) of the recommendation system cannot be explained. Meanwhile, for the recommendation system of the black box, under the condition that the model and the characteristics in the recommendation system are not known or are very complex, the recommendation result of the recommendation system is difficult to interpret by the existing scheme, namely the causal relationship between the user behavior and the recommendation result of the recommendation system cannot be interpreted.

Disclosure of Invention

Aiming at the defects of the existing mode, the application provides an interpretation method, a device, equipment and a computer readable storage medium aiming at a recommendation system, which are used for solving the problem of how to realize the effect-cause relationship between a user request and the recommended content of the recommendation system.

In a first aspect, the present application provides an interpretation method for a recommendation system, including:

responding to the first request, obtaining first recommended content aiming at the first request, wherein the first recommended content is recommended content expected to be obtained from a recommending system, and performing recommended content comparison operation:

responding to the first request sent to the recommendation system, and obtaining second recommendation content recommended by the recommendation system according to the first request;

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 smaller than a preset similarity, and the total request times of requesting the recommended content from the recommendation system are smaller than the preset request times, updating the first request;

repeating the step of comparing the recommended contents until the similarity between the second feature vector and the first feature vector is not less than the preset similarity, and obtaining a first explanation aiming at the recommendation system;

the first interpretation comprises that the recommendation system has the response capability of recommending the first recommended content for the first request within the preset request times.

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 request times is not less than a preset request times, a second explanation for the recommendation system is obtained; the second interpretation comprises that the recommending system does not have the response capability of recommending the first recommended content for the first request within the preset request times.

In one embodiment, the update processing of the first request includes:

constructing a first history request set, wherein the first history request set comprises first requests;

respectively inputting each history request in the first history request set to a preset prediction model to obtain a content tag corresponding to each history request, wherein the content tag is used for identifying history recommended content recommended by a recommendation system in response to the corresponding history request, and comprises related information of the history recommended content;

and determining the acquaintance degree between the historical characteristic vector corresponding to each historical recommended content and the first characteristic vector corresponding to the first recommended content, and taking the historical request corresponding to the minimum similarity in the obtained acquaintance degrees as the updated first request.

In one embodiment, the predictive model is trained by:

constructing a training sample set based on a preset second history request set; performing iterative training on the initial prediction model to be trained based on the training sample set to obtain a trained prediction model;

the following processing is executed in the initial prediction model iterative training process:

taking a sample request included in a training sample set as an input sample of an initial prediction model, taking a content label corresponding to the sample request as an output result of the initial prediction model, substituting the input sample and the output result into a loss function corresponding to the initial prediction model to obtain a loss function value, updating model parameters of the initial prediction model based on the loss function value, and updating the model parameters of the initial prediction model according to the prediction model parameters obtained when the loss function value meets a preset condition to obtain a trained prediction model.

In one embodiment, sending the first request to a recommendation system includes:

generating an operation instruction based on a preset decision algorithm, and transmitting the operation instruction to any application program;

and responding to a first request generated by any application program based on the operation instruction, and sending the first request to the recommendation system.

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 smaller than the preset similarity, and the total request times of requesting the recommendation system for recommending content is smaller than the preset request times, performing update processing on the first request, including:

when the distance between a second feature vector corresponding to the second recommended content and a first feature vector corresponding to the first recommended content is larger than a preset distance and the total request times of requesting the recommended content from the recommendation system are smaller than the preset request times, updating the first request;

the distance between the second feature vector and the first feature vector is inversely related to the similarity.

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, a second explanation for the recommendation system is obtained, including:

and 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 smaller than a preset similarity, the total request times are not smaller than the preset request times, and the second recommended content obtained every time in the total request times is the same, obtaining a second explanation, wherein the second explanation comprises a factor that the first request does not form the recommendation of the second recommended content by the recommendation system, and the second recommended content comprises a hot event.

In a second aspect, the present application provides an interpretation apparatus for a recommendation system, comprising:

the first processing module is used for responding to the first request and obtaining first recommended content aiming at the first request, wherein the first recommended content is recommended content expected to be obtained from a recommending system;

the second processing module is used for executing the comparison operation of the recommended content:

In a third aspect, the present application provides an electronic device, comprising: a processor, a memory, and a bus;

a bus for connecting the processor and the memory;

a memory for storing operating instructions;

and the processor is used for executing the interpretation method aiming at the recommendation system in the first aspect of the application by calling the operation instruction.

In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program for executing the interpretation method for the recommendation system of the first aspect of the present application.

The technical scheme provided by the embodiment of the application at least has the following beneficial effects:

comparing second recommended content recommended by a recommendation system according to a first request (user request, namely user behavior) with first recommended content, and if 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 not less than a preset similarity within a preset request number, the recommendation system has the response capability of recommending the first recommended content for the first request; as such, the following explanation can be made for the recommendation system: in limited interactions between the user and the recommendation system, the recommendation system can respond to the user request more quickly, the user can obtain the preset desired recommended content, and a causal relationship is formed between the user request and the recommended content of the recommendation system, namely the recommended content is generated corresponding to the user request.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings used in the description of the embodiments of the present application will be briefly described below.

Fig. 1 is a schematic diagram of a system architecture provided in an embodiment of the present application;

fig. 2 is a schematic flowchart of an explanation method for a recommendation system according to an embodiment of the present application;

FIG. 3 is a schematic flowchart of another explanation method for a recommendation system according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of an explanation device for a recommendation system according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

In order to make the objects, features and advantages of the present invention more apparent 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, and it is apparent that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Reference will now be made in detail to embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are exemplary only for the purpose of explaining the present application and are not to be construed as limiting the present application.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.

The embodiment of the application provides an explanation method for a recommendation system in the field of instant messaging, and the explanation method for the recommendation system relates to the technical field of machine learning in the field of artificial intelligence and various fields of cloud technology, such as cloud computing, cloud service and the like in the cloud technology.

Cloud computing (cloud computing) is a computing model that distributes computing tasks over a pool of resources formed by a large number of computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The network that provides the resources is referred to as the "cloud". Resources in the "cloud" appear to the user as being infinitely expandable and available at any time, available on demand, expandable at any time, and paid for on-demand.

As a basic capability provider of cloud computing, a cloud computing resource pool (called as an ifas (Infrastructure as a Service) platform for short is established, and multiple types of virtual resources are deployed in the resource pool and are selectively used by external clients.

According to the logic function division, a PaaS (Platform as a Service) layer can be deployed on an IaaS (Infrastructure as a Service) layer, a SaaS (Software as a Service) layer is deployed on the PaaS layer, and the SaaS can be directly deployed on the IaaS. PaaS is a platform on which software runs, such as a database, a web container, etc. SaaS is a variety of business software, such as web portal, sms, and mass texting. Generally speaking, SaaS and PaaS are upper layers relative to IaaS.

So-called artificial intelligence cloud services are also commonly referred to as AIaaS (AIas a Service, chinese "AI as a Service"). The method is a service mode of an artificial intelligence platform, and particularly, the AIaaS platform splits several types of common AI services and provides independent or packaged services at a cloud. This service model is similar to the one opened in an AI theme mall: all developers can access one or more artificial intelligence services provided by the platform through an API (application programming interface), and part of the qualified developers can also use an AI framework and an AI infrastructure provided by the platform to deploy and operate and maintain the self-dedicated cloud artificial intelligence services.

Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and the like.

Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

For better understanding and description of the embodiments of the present application, some technical terms used in the embodiments of the present application will be briefly described below.

The recommendation system comprises: the appearance and popularization of the internet bring a great deal of information to users, and the requirement of the users on the information in the information age is met, but the quantity of the information on the internet is greatly increased along with the rapid development of the network, so that the users cannot obtain the part of information which is really useful for the users when facing a great amount of information, and the use efficiency of the information is reduced on the contrary, which is the so-called information overload problem. A very potential approach to solving the information overload problem is a recommendation system, which is a personalized information recommendation system that recommends information, products, and the like that a user is interested in to the user according to the information needs, interests, and the like of the user. The recommendation system carries out personalized calculation by researching the interest preference of the user, and finds the interest points of the user through the recommendation system, thereby guiding the user to find the own information requirement.

Feed stream: a Feed stream is an information stream; in the internet field, Feed stream products such as friend circles, microblogs and the like, and a picture sharing website is also a form of Feed stream product; the App may have a module named dynamic, message plaza, etc. which is also a Feed stream product.

Euclidean distance: euclidean distance is a commonly used definition of distance, referring to the true distance between two points in m-dimensional space, or the natural length of a vector (i.e., the distance of the point from the origin). The euclidean distance in two and three dimensions is the actual distance between two points.

Cosine similarity: cosine similarity measures the difference between two individuals by using the cosine value of the included angle between two vectors in the vector space. The closer the cosine value is to 1, the closer the angle is to 0 degrees, i.e. the more similar the two vectors are.

The technical scheme provided by the embodiment of the present application relates to a cloud technology, and the following detailed description is provided on the technical scheme of the present application and how to solve the technical problem in the technical scheme of the present application with specific embodiments. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments. Embodiments of the present application will be described below with reference to the accompanying drawings.

The scheme provided by the embodiment of the application can be suitable for any application scene needing explanation for the recommendation system in the cloud technical field, and the following explanation can be carried out for the recommendation system through the scheme: in limited interactions between the user and the recommendation system, the recommendation system can respond to the user request more quickly, the user can obtain the preset desired recommended content, and a causal relationship is formed between the user request and the recommended content of the recommendation system, namely the recommended content is generated corresponding to the user request.

In order to better understand the scheme provided by the embodiment of the present application, the scheme is described below with reference to a specific application scenario.

In an embodiment, fig. 1 shows a system architecture diagram of an explanation for a recommendation system, to which the embodiment of the present application is applied, and it is understood that the explanation method for a recommendation system provided in the embodiment of the present application may be applied to, but is not limited to, the application scenario shown in fig. 1.

In this example, as shown in fig. 1, the system architecture of the explanation for the recommendation system in this example may include, but is not limited to, a recommendation countermeasure interpretation system 101 and a recommendation system 102, and interaction between the recommendation countermeasure interpretation system 101 and the recommendation system 102 may be performed through a network. It is recommended that countermeasure interpretation system 101 includes countermeasure interpretation decision engine 1011 and terminal 1012, agent-1, agent-2, … agent-N shown in FIG. 1 are all terminal 1012. The countermeasure interpretation decision engine 1011 generates a user instruction (operation instruction) according to a preset decision algorithm, and sends the user instruction to the terminal 1012; user commands such as refresh, like, comment, etc. The terminal 1012 receives the user instruction sent by the anti-interpretation decision engine 1011 and the terminal 1012 sends the user request to the recommender system 102 according to the user instruction. The recommendation system 102 transmits a recommendation result (recommended content) to the terminal 1012 according to the received user request. The terminal 1012 sends the received recommendation to the confrontation interpretation decision engine 1011. The countermeasure interpretation decision engine 1011 receives the recommendation sent by each terminal 1012 and stores the corresponding user instructions and environmental data.

It is understood that the above is only an example, and the present embodiment is not limited thereto.

The terminal 1012 may be a smart phone (e.g., an Android phone, an iOS phone, etc.), a phone simulator, a tablet computer, a notebook computer, a digital broadcast receiver, an MID (Mobile Internet Devices), a PDA (personal digital assistant), a desktop computer, a vehicle terminal (e.g., a vehicle navigation terminal), a smart speaker, a smart watch, etc. The countermeasure interpretation decision engine 1011 or the recommendation system 102 may be an independent physical server, may also be a server cluster or a distributed system formed by a plurality of physical servers, and may also be a cloud server or a server cluster that provides basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a CDN (Content Delivery Network), a big data and artificial intelligence platform, and the like. Such networks may include, but are not limited to: a wired network, a wireless network, wherein the wired network comprises: a local area network, a metropolitan area network, and a wide area network, the wireless network comprising: bluetooth, Wi-Fi, and other networks that enable wireless communication. The determination may also be based on the requirements of the actual application scenario, and is not limited herein.

Referring to fig. 2, fig. 2 shows a flowchart of an explanation method for a recommendation system provided in an embodiment of the present application, where the method may be executed by any electronic device, such as a recommendation countermeasure interpretation system, as an alternative implementation, the method may be executed by the recommendation countermeasure interpretation system, and for convenience of description, in the following description of some alternative embodiments, the recommendation countermeasure interpretation system will be described as an example of an execution subject of the method. As shown in fig. 2, the explanation method for the recommendation system provided in the embodiment of the present application includes the following steps:

s101, responding to the first request, and obtaining first recommended content aiming at the first request, wherein the first recommended content is recommended content expected to be obtained from a recommending system.

In one embodiment, the first recommended content may be text, pictures, video, audio, and the like. 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 ].

In one embodiment, the first request is a user request for characterizing user behavior, such as a user's approval for a short video of a bid-winning object in a game. The first request is for requesting a first recommended content from a recommendation system. The first recommended content may be preset to evaluate whether the recommendation system has a response capability of recommending the first recommended content for the first request.

S102, performing comparison operation of recommended contents:

In one embodiment, the initial value of the total number of requests n for the recommended content from the recommendation countering interpretation system to the recommendation system is m, where m is a positive integer.

For example, 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 smaller than the preset similarity, the total number of times n of requests is accumulated to 1, that is, n is updated to n +1, for example, if the recommended countermeasure interpretation system sends the first request to the recommendation system for the first time, and if m is 0, n is updated from the initial value 0 to 1.

In one embodiment, a 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, a distance, for example, a euclidean distance, between the second feature vector corresponding to the second recommended content and the first feature vector corresponding to the first recommended content may be used to measure a 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.

In one embodiment, the similarity between the second eigenvector corresponding to the second recommended content and the first eigenvector corresponding to the first recommended content is measured through cosine similarity, that is, the cosine value of the included angle between the second eigenvector and the first eigenvector can be used for measuring the similarity between the second eigenvector and the first eigenvector. The closer the cosine value is to 1, the closer the included angle is to 0 degrees, i.e. the more similar the two vectors are, the greater the similarity between the second feature vector and the first feature vector is.

For example, the first request is a user request for characterizing user behavior, and the user behavior is that the user comments on a panda video. The user commends the panda video for the first time, namely the recommendation countermeasure interpretation system sends a first request to the recommendation system, the recommendation content returned by the recommendation system is a koala video, wherein the first recommendation content is the panda video, the second recommendation content is the koala video, the similarity between a second feature vector corresponding to the second recommendation content and a first feature vector corresponding to the first recommendation content is smaller than a preset similarity, and the total request times 1 of requesting the recommendation content from the recommendation system is smaller than the preset request times 5, so that the recommendation countermeasure interpretation system updates the first request to obtain the updated first request. The user commends the panda video for the second time, namely the recommendation confrontation interpretation system sends the updated first request to the recommendation system for the second time, the recommendation content returned by the recommendation system is the panda video, the first recommendation content is the panda video, the second recommendation content is the panda video, the similarity between the second feature vector and the first feature vector is not smaller than the preset similarity, and the first interpretation aiming at the recommendation system is obtained; the first explanation includes that within the preset request times 5, the recommendation system has the capability of recommending the panda video according to the comment of the user on the panda video, so that the following can be explained: in limited interactions between a user and the recommendation system, the recommendation system can respond to comments of the panda videos relatively quickly, the user can obtain a preset wanted panda video, and a causal relationship is formed between a user request (the user comments on the panda video) and recommendation contents (the panda video) of the recommendation system, namely the user comments on the panda video enable the recommendation system to recommend the panda videos corresponding to the recommendation system.

In the embodiment of the present application, the following explanation may be performed for the recommendation system: in limited interactions between the user and the recommendation system, the recommendation system can respond to the user request more quickly, the user can obtain the preset desired recommended content, and a causal relationship is formed between the user request and the recommended content of the recommendation system, namely the recommended content is generated corresponding to the user request.

Illustratively, the first request is a user request for characterizing user behavior, which is the user listening to the songs of singer B from song album A. The user listens to the song of the singer B for the first time, namely the recommendation countermeasure interpretation system sends a first request to the recommendation system, the recommendation content returned by the recommendation system is the songs of other singers except the singer B in the song album A, wherein the first recommendation content is the song of the singer B, the second recommendation content is the songs of other singers in the song album A, the similarity between the second characteristic vector corresponding to the second recommendation content and the first characteristic vector corresponding to the first recommendation content is smaller than the preset similarity, and the total request times 1 of requesting the recommendation content from the recommendation system is smaller than the preset request times 10, then the recommendation countermeasure interpretation system updates the first request to obtain the updated first request. The user listens to the song of the singer B for multiple times, namely the recommendation countermeasure interpretation system sends the updated first request to the recommendation system for multiple times, the recommended contents returned by the recommendation system are all songs of other singers in the song album A, wherein 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 smaller than the preset similarity, and the total request times 10 are not smaller than the preset request times 10, so that a second interpretation aiming at the recommendation system is obtained; the second interpretation includes that the recommendation system does not have the ability to recommend the singer's B songs for the user to listen to the singer's B songs on the song album a within the preset number of requests 10. It can thus be stated that: in limited interactions between a user and a recommendation system, the recommendation system cannot respond to songs of a singer B quickly, the user cannot obtain songs of the singer B which is preset to listen, and a causal relationship is not formed between a user request (for listening to the songs of the singer B) and recommended contents (for listening to the songs of other singers in a song album A) of the recommendation system, namely, the user cannot listen to the songs of the singer B to cause the recommendation system to recommend the songs of the singer B.

In one embodiment, the predictive model is trained by:

In one embodiment, the user open request is constructed by the antagonism interpretation decision engine 1011 in FIG. 1, and may be a recommendation function that requires interpretation, such as a qq Small world Square Page. The countermeasure interpretation decision engine 1011 stores key information such as device information including that the operating system of the terminal 1012 is an android system and behavior operation time including a time point of user operation (generation of an operation instruction) and the like at the same time. The countermeasure interpretation decision engine 1011 generates an operation instruction (user instruction) based on a preset decision algorithm, transmits the operation instruction to any application app on the terminal 1012, for example, an app of a qq small world square page, and in response to a first request generated by any application app based on the operation instruction, the terminal 1012 transmits the first request to the recommendation system 102. The recommender system 102 returns a recommendation to the terminal 1012, for example a short video, Feed stream, etc., wherein the short video may be a corresponding short video presented as a video stream of a spatial video, for example a football-like video. The terminal 1012 transmits the recommended content to the countermeasure interpretation decision engine 1011, and the countermeasure interpretation decision engine 1011 records the recommended content and performs the modeling of the relationship between the user behavior and the recommended content, that is, content ═ f (behavior); the content represents a content tag, the content tag is used for identifying historical recommended content recommended by a recommendation system in response to a corresponding historical request, and the content tag comprises related information of the historical recommended content; behavior represents historical user requests, i.e., historical requests, such as liking beauty 3 times, commenting military content 10 times, etc.; the prediction model F can be used to predict: which content is recommended by a recommendation system with the maximum probability under a specific behavior, and the prediction model F can be obtained by a multi-classification machine learning method.

In one embodiment, the update process is performed on the first request, including steps A1-A3:

step A1, a first set of history requests is constructed, the first set of history requests including a first request.

Step A2, respectively inputting each history request in the first history request set to a preset prediction model, and obtaining a content tag corresponding to each history request, where the content tag is used to identify history recommended content recommended by a recommendation system in response to the corresponding history request, and the content tag includes related information of the history recommended content.

Step A3, determining the acquaintance between the history feature vector corresponding to each history recommended content and the first feature vector corresponding to the first recommended content, and taking the history request corresponding to the minimum similarity in the obtained acquaintance as the updated first request.

In one embodiment, the user request (first request) is generated according to a user behavior generation algorithm, i.e., the first request is updated according to the user behavior generation algorithm. The goals of the user behavior generation algorithm are: in the generated user request (for example, a history request in the first history request set), the prediction model estimates a user behavior that minimizes a difference between a vector corresponding to a next recommended content and a given target content vector (a first feature vector corresponding to a first recommended content), that is, a content ═ f (behavior) is used to reversely solve the behavior, specifically, all combinations of behaviors are traversed to obtain a corresponding content and a second feature vector X corresponding to the content, and a combination of the behavior with the smallest euclidean distance between the second feature vector X and the first feature vector Y is determined. The user behavior generation algorithm can modify the prediction model most quickly and provide user behaviors that can be explored efficiently according to the user behaviors of all the terminals 1012 in fig. 1 and corresponding recommendation results.

For example, the operation instruction may be refresh, approval, comment, and the like, and the application may be an APP application installed on the terminal.

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 a euclidean distance, and the euclidean distance may 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, i.e. the distance between the second feature vector and the first feature vector is inversely related to 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 larger than a preset distance, namely 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 smaller than a preset similarity.

For example, the first request is a user request for characterizing user behavior, and the user behavior is that the user likes pictures of a landscape. The user approves the picture of a certain landscape each time, namely, the recommendation countermeasure interpretation system sends a first request to the recommendation system, the recommendation contents returned by the recommendation system are all pictures of scenes in the national sports meeting which is being held, wherein the first recommendation contents are the pictures of the certain landscape, the second recommendation contents are the pictures of the scenes in the national sports meeting which is being held, the similarity between a second feature vector corresponding to the second recommendation contents and a first feature vector corresponding to the first recommendation contents is smaller than the preset similarity, the total request times are not smaller than the preset request times, and the second recommendation contents obtained in each time in the total request times are the same, so that a second interpretation is obtained, the second interpretation comprises a factor of confirming that the approval of the pictures of the certain landscape is not a factor of the pictures which constitute the scene which is being held by the recommendation system in the national sports meeting which is being held, and the national sports meeting which is being held is a network hotspot, i.e. a hotspot event. In limited interactions between a user and a recommendation system, the recommendation system cannot respond to the praise of the picture of the certain landscape quickly, the user cannot obtain the picture of the certain landscape, and no causal relationship is formed between the user request (praise the picture of the certain landscape) and the recommendation content (picture of the scene in the national sporting meeting) of the recommendation system, namely, praise of the picture of the certain landscape does not enable the recommendation system to recommend the picture of the certain landscape.

In the embodiment of the present application, the following explanation may be performed for the recommendation system: in limited interactions between a user and a recommendation system, the recommendation system cannot respond to a user request more quickly, the user cannot obtain preset desired recommended content, and a causal relationship is not formed between the user request and the recommended content of the recommendation system, namely the user request cannot generate the recommended content correspondingly.

In order to better understand the method provided by the embodiment of the present application, the following further describes the scheme of the embodiment of the present application with reference to an example of a specific application scenario.

Referring to fig. 3, fig. 3 shows a flowchart of an explanation method for a recommendation system provided in an embodiment of the present application, where the method may be executed by any electronic device, such as a recommendation countermeasure interpretation system, as an alternative implementation, the method may be executed by the recommendation countermeasure interpretation system, and for convenience of description, in the following description of some alternative embodiments, the recommendation countermeasure interpretation system will be described as an example of an execution subject of the method. As shown in fig. 3, the interpretation method for the recommendation system provided in the embodiment of the present application includes the following steps:

s201, setting the expected content expression vector Y, and assigning the initial value of the cycle number n to 0.

In one embodiment, the expected content expression vector Y may be a first feature vector corresponding to a first recommended content, the first recommended content being a recommended content expected to be obtained from a recommendation system. The first recommended content may be text, pictures, video, audio, and the like. For example, the first recommended content is a short video for eating chicken, and the content expression vector Y corresponding to the short video for eating chicken is [1.0,2.1,2.3,1.2,5.4 ]. The number of loops n may be a total number of requests n to request recommended content from the recommendation system.

S202, sending a user request to the recommendation system, and receiving a content expression vector X returned by the recommendation system.

In one embodiment, the user open request is constructed by the antagonism interpretation decision engine 1011 in FIG. 1, and may be a recommendation function that requires interpretation, such as a qq Small world Square Page. The countermeasure interpretation decision engine 1011 stores key information such as device information including that the operating system of the terminal 1012 is an android system and behavior operation time including a time point of user operation (generation of an operation instruction) and the like at the same time. The countermeasure interpretation decision engine 1011 generates an operation instruction (user instruction) based on a preset decision algorithm, transmits the operation instruction to any application app on the terminal 1012, for example, an app of a qq small world square page, and in response to a 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 recommender system 102 returns a recommendation to the terminal 1012, for example a short video, Feed stream, etc., wherein the short video may be a corresponding short video presented as a video stream of a spatial video, for example a football-like video. The terminal 1012 transmits the recommended content to the countermeasure interpretation decision engine 1011, and the countermeasure interpretation decision engine 1011 records the recommended content and performs the modeling of the relationship between the user behavior and the recommended content, that is, content ═ f (behavior); the content represents a content tag, the content tag is used for identifying historical recommended content recommended by a recommendation system in response to a corresponding historical request, and the content tag comprises related information of the historical recommended content; behavior represents historical user requests, i.e., historical requests, such as 5 prasuals in captain, 11 comments on a football event, etc.; the prediction model F can be used to predict: under a specific behavior, the recommendation system will recommend which content with the highest probability. The content tag is input to an external system, for example, a content understanding system, to obtain corresponding recommended content, and a content expression vector X is obtained according to the corresponding recommended content, where the content expression vector X may be a second feature vector corresponding to a second recommended content.

S203, judging whether the distance between the content expression vector X and the content expression vector Y is smaller than a preset first threshold value, and turning to the S204 when the distance between the content expression vector X and the content expression vector Y is smaller than the preset first threshold value; 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, processing goes to S205.

In one embodiment, the difference between the content expression vector X and the content expression vector Y, i.e. the distance between the content expression vector X and the content expression vector Y, e.g. 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.

And S204, outputting the sequence length n of the user operation, and outputting a corresponding user operation sequence and a recommended content sequence.

In one embodiment, the length n of the sequence of user operations may be n of the number of recording cycles, that is, the total number of requests for requesting recommended content from the recommendation system, the sequence of user operations includes a user request sent each time, and the sequence of recommended content includes recommended content returned by the recommendation system for each user request.

S205, judging whether the length n of the current sequence is smaller than a second threshold value, and turning to S206 for processing when the length n of the current sequence is smaller than the second threshold value; when it is determined that the current sequence length n is not less than the second threshold, processing proceeds to S207.

In one embodiment, the length n of the current sequence may be n of the number of recording cycles, that is, the total number of requests for requesting recommendation content from the recommendation system, and the second threshold may be a preset number of requests.

S206, generating a user request according to a user behavior generation algorithm, and assigning the cycle number n as n + 1.

In one embodiment, the user request (first request) is generated according to a user behavior generation algorithm, i.e., the first request is updated according to the user behavior generation algorithm. The goals of the user behavior generation algorithm are: in the generated user request (for example, a history request in the first history request set), the prediction model estimates a user behavior that minimizes a difference between a vector corresponding to a next recommended content and a given target content vector (a first feature vector corresponding to a first recommended content), that is, a content ═ f (behavior) is used to reversely solve the behavior, specifically, all combinations of behaviors are traversed to obtain a corresponding content and a content expression vector X corresponding to the content, and the combination of the 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 can modify the prediction model most quickly and provide user behaviors that can be explored efficiently according to the user behaviors of all the terminals 1012 in fig. 1 and corresponding recommendation results.

And S207, outputting the sequence length n of the user operation, and outputting the corresponding user operation sequence, the recommended content sequence and the minimum vector difference of the historical recommendation result.

In one embodiment, the length n of the sequence of user operations may be n of the number of recording cycles, that is, the total number of requests for requesting recommended content from the recommendation system, the sequence of user operations includes a user request sent each time, and the sequence of recommended content includes recommended content returned by the recommendation system for each user request. Determining vector differences between the history feature vectors and the content expression vectors Y corresponding to the history recommended contents respectively, and determining the minimum vector difference in the obtained vector differences as the minimum vector difference of the history recommended results (history recommended contents).

S208, obtaining an explanation for the recommendation system.

In one embodiment, through S201-S207, the user request and the corresponding recommendation result (push content) are obtained, and 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, and the predetermined desired content corresponds to the expected content expression vector Y, which indicates that the recommendation system does not have obvious responsiveness to the user behavior. When n is smaller than the second threshold, the fact that the recommendation system can obtain the preset desired content in a limited number of user interactions is shown, the preset desired content corresponds to the expected content expression vector Y, and the fact that the recommendation system has a quick response capability to user behaviors is shown. And simultaneously outputting content which is F (behavior), wherein F is used as a prediction model of the recommendation system.

In the embodiment of the application, the interpretation method for the recommendation system is used for the existing recommendation system and the investigated external recommendation system, quantitative evaluation and analysis results of the existing recommendation system and the investigated external recommendation system are provided, and great help is provided for evaluation, optimization and investigation of the existing recommendation system and the investigated external recommendation system. Whether the recommendation system and the user behavior form a causal relationship is obtained in a counterstudy mode, the causal relationship is quantitatively modeled, the causal relationship obtained through modeling determines what behavior sequence can generate what recommendation result, and meanwhile, the recommendation reason which is not formed by the user behavior is given, so that the recommendation system can know how the user behavior and the context influence the recommendation result, and the expected recommendation result can be obtained by controlling the user behavior.

Based on the same inventive concept, the embodiment of the present application further provides an interpretation apparatus for a recommendation system, a schematic structural diagram of the apparatus is shown in fig. 4, and the interpretation apparatus 40 for a recommendation system includes a first processing module 401 and a second processing module 402.

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

A second processing module 402, configured to perform a recommended content comparison operation:

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

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 smaller than a preset similarity, and the total request times are not smaller than the preset request times, a second explanation aiming at the recommendation system is obtained; the second interpretation comprises that the recommending system does not have the response capability of recommending the first recommended content for the first request within the preset request times.

In an embodiment, the second processing module 402 is specifically configured to:

In an 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 smaller than the preset similarity, and the total request frequency of requesting the recommendation system for recommending content is smaller than the preset request frequency, the second processing module 402 is specifically configured to:

In an 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:

The application of the embodiment of the application has at least the following beneficial effects:

Based on the same inventive concept, the embodiment of the present application further provides an electronic device, a schematic structural diagram of which is shown in fig. 5, where the electronic device 9000 includes at least one processor 9001, a memory 9002, and a bus 9003, and 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 to perform the steps of any of the interpretation methods for the recommendation system as provided by any of the embodiments or any alternative embodiments in the present application.

Further, the processor 9001 may be an FPGA (Field-Programmable Gate Array) or other devices with logic processing capability, such as an MCU (micro controller Unit) and a CPU (Central processing Unit).

Based on the same inventive concept, the embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and the computer program is used for implementing the steps of any one of the interpretation methods for the recommendation system provided in any one of the embodiments or any one of the alternative embodiments of the present application when the computer program is executed by a processor.

The computer-readable storage medium provided by the embodiments of the present application includes, but is not limited to, any type of disk including floppy disks, hard disks, optical disks, CD-ROMs, and magneto-optical disks, ROMs (Read-Only memories), RAMs (Random Access memories), EPROMs (Erasable Programmable Read-Only memories), EEPROMs (Electrically Erasable Programmable Read-Only memories), flash memories, magnetic cards, or optical cards. That is, a readable storage medium includes any medium that stores or transmits information in a form readable by a device (e.g., a computer).

Embodiments of the present application further provide a computer program product containing instructions, which when run on a computer device, cause the computer device to execute the interpretation method for the recommendation system provided in the foregoing method embodiments.

It will be understood by those within the art that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by computer programs. Those skilled in the art will appreciate that the computer program product may be implemented by a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, implement the aspects specified in the block or blocks of the block diagrams and/or flowchart illustrations disclosed herein.

Those of skill in the art will appreciate that the various operations, methods, steps in the processes, acts, or solutions discussed in this application can be interchanged, modified, combined, or eliminated. Further, other steps, measures, or schemes in various operations, methods, or flows that have been discussed in this application can be alternated, altered, rearranged, broken down, combined, or deleted. Further, steps, measures, schemes in the prior art having various operations, methods, procedures disclosed in the present application may also be alternated, modified, rearranged, decomposed, combined, or deleted.

The foregoing is only a partial embodiment of the present application, and it should be noted that, for those skilled in the art, several modifications and decorations can be made without departing from the principle of the present application, and these modifications and decorations should also be regarded as the protection scope of the present application.

Claims

1. An interpretation method for a recommendation system, comprising:

responding to a first request, acquiring first recommended content aiming at the first request, wherein the first recommended content is recommended content expected to be obtained from a recommendation system, and performing recommended content comparison operation:

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 smaller than a preset similarity, and the total request times of requesting the recommendation content from the recommendation system are smaller than the preset request times, updating the first request;

repeatedly executing the step of comparing the recommended content until the similarity between the second feature vector and the first feature vector is not less than a preset similarity, and obtaining a first explanation aiming at the recommendation system;

the first explanation comprises that the recommendation system has the response capability of recommending the first recommended content for the first request within the preset request times.

2. The method of claim 1, further comprising:

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 smaller than a preset similarity, and the total request times are not smaller than preset request times, obtaining a second explanation for the recommendation system;

the second interpretation comprises that the recommending system does not have the response capability of recommending the first recommended content for the first request within the preset request times.

3. The method of claim 1, wherein the updating the first request comprises:

building a first history request set, wherein the first history request set comprises the first request;

respectively inputting each history request in the first history request set to a preset prediction model to obtain a content tag corresponding to each history request, wherein the content tag is used for identifying history recommended content recommended by the recommendation system in response to the corresponding history request, and the content tag comprises relevant information of the history recommended content;

4. The method of claim 3, wherein the predictive model is trained by:

constructing a training sample set based on a preset second history request set; performing iterative training on the initial prediction model to be trained on the basis of the training sample set to obtain a trained prediction model;

performing the following processing in the initial prediction model iterative training process:

and taking a sample request included in the training sample set as an input sample of the initial prediction model, taking a content label corresponding to the sample request as an output result of the initial prediction model, substituting the input sample and the output result into a loss function corresponding to the initial prediction model to obtain a loss function value, updating model parameters of the initial prediction model based on the loss function value, and updating the model parameters of the initial prediction model according to the prediction model parameters obtained when the loss function value meets a preset condition to obtain the trained prediction model.

5. The method of claim 1, wherein sending the first request to the recommendation system comprises:

6. The method according to claim 1, 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 smaller than a preset similarity, and the total number of requests for requesting recommended content from the recommendation system is smaller than a preset number of requests, performing update processing on the first request includes:

7. 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 smaller than a preset similarity, and the total number of requests is not smaller than a preset number of requests, obtaining a second interpretation for the recommendation system includes:

and 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 smaller than a preset similarity, the total request frequency is not smaller than a preset request frequency, and the second recommended content obtained every time in the total request frequency is the same, obtaining a second interpretation, wherein the second interpretation comprises a factor that the first request does not form the recommendation of the second recommended content by the recommendation system, and the second recommended content comprises a hotspot event.

8. An interpretation apparatus for a recommendation system, comprising:

the first processing module is used for responding to a first request, and obtaining first recommended content aiming at the first request, wherein the first recommended content is recommended content expected to be obtained from a recommending system;

9. The apparatus of claim 8, wherein the second processing module is further configured to:

10. The apparatus of claim 8, wherein the second processing module is specifically configured to:

11. The apparatus of claim 8, wherein the second processing module is specifically configured to:

12. The apparatus of claim 8, wherein the second processing module is specifically configured to:

13. The apparatus according to claim 9, wherein the second processing module is specifically configured to:

14. An electronic device, comprising: a processor, a memory;

the memory for storing a computer program;

the processor is configured to execute the interpretation method for the recommendation system according to any one of claims 1 to 7 by calling the computer program.

15. A computer-readable storage medium, characterized in that a computer program is stored which, when being executed by a processor, is adapted to carry out the interpretation method for a recommendation system according to any one of claims 1-7.