Resource recommendation method, device, electronic device, product and medium

Abstract

The embodiments of the present application disclose resource recommendation methods, devices, electronic devices, products and media, which can be applied to the field of artificial intelligence technology. The method includes: obtaining an initial resource network; determining an object among N objects that has object access behavior to both the first recommended resource and the second recommended resource as a target object, and determining the edge weight of the target edge according to the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval of the target object to the first recommended resource and the second recommended resource, and the object activity of the target object; determining the initial resource network with the edge weight of the target edge as the target resource network, and recommending recommended resources to N objects according to the target resource network. By adopting the embodiments of the present application, the accuracy and diversity of resource recommendations for objects can be improved. The present application can be applied to various scenarios such as cloud technology, big data, artificial intelligence, smart transportation, and assisted driving.

Description

Technical Field

[0001] The present application is applied to the field of artificial intelligence technology, and in particular relates to a resource recommendation method, device, electronic device, product and medium. Background Art

[0002] Resource recommendations have been widely used in various software, websites or systems. At present, the training data of the recommendation model of the resource recommendation system mainly depends on the resource exposure data. For popular resources, since there are enough positive samples in the training data, the final resource recommendation system tends to recommend popular resources. Popular resources eventually get more exposure opportunities, while unpopular resources cannot get exposure opportunities, which leads to the problem of popularity bias, and the accuracy and diversity of resource recommendations are low. Summary of the invention

[0003] The embodiments of the present application provide a resource recommendation method, device, electronic device, product and medium, which can improve the accuracy and diversity of resource recommendation for objects.

[0004] On the one hand, an embodiment of the present application discloses a resource recommendation method, the method comprising:

[0005] An initial resource network is obtained; the initial resource network is constructed according to the object access behaviors of N objects for M recommended resources, where N and M are both positive integers, and the initial resource network includes a plurality of resource nodes and edges between the plurality of resource nodes, where one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors for both recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is indicated as a target edge, and the target edge is used to connect a resource node corresponding to the first recommended resource and a resource node corresponding to the second recommended resource;

[0006] Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval between the target object and the first recommended resource and the second recommended resource, and the object activity of the target object;

[0007] Determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0008] An initial resource network having the edge weight of the target edge is determined as the target resource network, and recommended resources are recommended for the N objects according to the target resource network.

[0009] On the one hand, an embodiment of the present application discloses a resource recommendation device, the device comprising:

[0010] An acquisition unit is used to acquire an initial resource network; the initial resource network is constructed according to the object access behaviors of N objects to M recommended resources, N and M are both positive integers, the initial resource network includes multiple resource nodes and edges between the multiple resource nodes, one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors to the recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is represented as a target edge, and the target edge is used to connect the resource node corresponding to the first recommended resource and the resource node corresponding to the second recommended resource;

[0011] a processing unit, configured to determine an object among the N objects that has object access behaviors to both the first recommended resource and the second recommended resource as a target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval of the target object to the first recommended resource and the second recommended resource, and the object activity of the target object;

[0012] The processing unit is further used to determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval and the object activity;

[0013] The processing unit is further configured to determine an initial resource network having a connection weight of a target connection as a target resource network, and recommend recommended resources to the N objects according to the target resource network.

[0014] In one aspect, an embodiment of the present application provides a computer program product or a computer program, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in various optional modes such as the above aspect.

[0015] On the one hand, an embodiment of the present application provides an electronic device, the electronic device includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to perform the following steps:

[0016] An initial resource network is obtained; the initial resource network is constructed according to the object access behaviors of N objects for M recommended resources, where N and M are both positive integers, and the initial resource network includes a plurality of resource nodes and edges between the plurality of resource nodes, where one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors for both recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is indicated as a target edge, and the target edge is used to connect a resource node corresponding to the first recommended resource and a resource node corresponding to the second recommended resource;

[0017] Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval between the target object and the first recommended resource and the second recommended resource, and the object activity of the target object;

[0018] Determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0019] An initial resource network having the edge weight of the target edge is determined as the target resource network, and recommended resources are recommended for the N objects according to the target resource network.

[0020] On the one hand, an embodiment of the present application provides a computer-readable storage medium, in which computer program instructions are stored. When the computer program instructions are executed by a processor, the computer program instructions are used to perform the following steps:

[0021] An initial resource network is obtained; the initial resource network is constructed according to the object access behaviors of N objects for M recommended resources, where N and M are both positive integers, and the initial resource network includes a plurality of resource nodes and edges between the plurality of resource nodes, where one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors for both recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is indicated as a target edge, and the target edge is used to connect a resource node corresponding to the first recommended resource and a resource node corresponding to the second recommended resource;

[0022] Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval between the target object and the first recommended resource and the second recommended resource, and the object activity of the target object;

[0023] Determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0024] An initial resource network having the edge weight of the target edge is determined as the target resource network, and recommended resources are recommended for the N objects according to the target resource network.

[0025] The embodiment of the present application can construct an initial resource network based on the object access behavior of the object to the recommended resources, and for any two recommended resources accessed by the same object, the edge weights of the edges corresponding to the two recommended resources in the initial resource network can be determined based on the resource access heat of the two recommended resources, the object activity of the objects that access both the two recommended resources, and the time interval for the same object to access the two recommended resources, so as to obtain the target resource network, and then recommend the recommended resources based on the target resource network. Therefore, when constructing the target resource network, the weight values ​​of the edges between the resource nodes corresponding to the two recommended resources can be determined by factors such as the resource access heat of the two recommended resources and the object activity of the objects that access both the recommended resources, which can improve the accuracy of the determined edge weights, and then recommend resources to the object through the target resource network with more accurate edge weights, thereby improving the accuracy and diversity of resource recommendations for the object. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 It is a structural diagram of a resource recommendation system provided in an embodiment of the present application;

[0028] Figure 2 It is a flowchart of a resource recommendation method provided in an embodiment of the present application;

[0029] Figure 3 This is a schematic diagram of the effect of an initial resource network provided in an embodiment of the present application;

[0030] Figure 4 This is a schematic diagram of the effect of a target resource network provided by an embodiment of the present application;

[0031] Figure 5 It is a flowchart of a resource recommendation method provided in an embodiment of the present application;

[0032] Figure 6 It is a flowchart of a resource recommendation method provided in an embodiment of the present application;

[0033] Figure 7 This is a schematic diagram of the effect of generating a candidate recommended resource set provided by an embodiment of the present application;

[0034] Figure 8 It is a flowchart of a resource recommendation method provided in an embodiment of the present application;

[0035] Fig. 9 is a structural diagram of a resource recommendation device provided in an embodiment of the present application;

[0036] Fig.10 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0038] The embodiment of the present application proposes a resource recommendation scheme, which can construct an initial resource network based on the object access behavior of the object to the recommended resource (which can be referred to as the resource), and for any two recommended resources accessed by the same object, the edge weights of the edges corresponding to the two recommended resources in the initial resource network can be determined based on the resource access heat of the two recommended resources, the object activity of the object that accesses both the two recommended resources, and the time interval between the same object accessing the two recommended resources, so as to obtain the target resource network, and then recommend the recommended resource based on the target resource network. Therefore, when constructing the target resource network, the weight value of the edge between the resource nodes corresponding to the two recommended resources can be determined by factors such as the resource access heat of the two recommended resources and the object activity of the object that accesses both the recommended resources, which can improve the accuracy of the determined edge weights, and then recommend resources to the object through the target resource network with more accurate edge weights, thereby improving the accuracy and diversity of resource recommendations for the object.

[0039] In a possible implementation, the present application embodiment can be applied to a resource recommendation system, see Figure 1 , Figure 1It is a structural diagram of a resource recommendation system provided by an embodiment of the present application, and the resource recommendation system may include a client and a server. Among them, the server can generate a target resource network according to the above-mentioned resource recommendation scheme. Since in the generated target resource network, the weights of the popular recommended resources are penalized with respect to the recommended resources visited by other recommended resources and objects with high activity, it is helpful to reduce the popularity deviation problem of the resources finally recommended. After the server determines the resource to be recommended in the end according to the target resource network, the resource information of the recommended resource (such as a picture of the resource, a title of the resource, etc.) can be sent to the client for display, that is, the recommendation of the recommended resource is realized. The client can be associated with a corresponding object. After the client receives the resource information returned by the server, the resource information can be displayed. If the object is interested in the displayed resource information, the resource corresponding to the displayed resource information can be accessed through the client, that is, an object access behavior for the resource is generated, so that the subsequent server obtains the object access behavior and recommends new resources according to the resources of interest to the object. Among them, the object access behavior can be a click on a resource; it can also be a behavior of clicking on a resource to display detailed content and browsing for a certain period of time, thereby reducing the situation where the object accidentally clicks on the recommended resource and causes inaccurate recommendation results; or it can also include other positive access behaviors to resources (such as likes, collections, searches for similar resources, etc.), which are not limited here.

[0040] It should be noted that, unless otherwise specified, in the specific implementation of this application, data related to object information, resource information, etc., when the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0041] It should be noted that this application can display a prompt interface or pop-up window before collecting relevant user data (such as the above-mentioned object access behavior and object activity) and during the process of collecting relevant user data. The prompt interface or pop-up window is used to prompt the user that its relevant data is currently being collected, so that this application only starts to execute the relevant steps of obtaining user-related data after obtaining the user's confirmation operation on the prompt interface or pop-up window, otherwise (that is, when the user's confirmation operation on the prompt interface or pop-up window is not obtained), the relevant steps of obtaining user-related data are terminated, that is, the user's relevant data is not obtained. In other words, all user data collected by this application are collected with the user's consent and authorization, and the collection, use and processing of relevant user data need to comply with the relevant laws, regulations and standards of relevant countries and regions.

[0042] In a possible implementation, the embodiments of the present application can also be applied to the field of big data technology, such as processing the above-mentioned large amount of object access behavior data through big data technology. Big data refers to a collection of data that cannot be captured, managed and processed by conventional software tools within a certain time frame. It is a massive, high-growth and diversified information asset that requires a new processing model to have stronger decision-making power, insight discovery and process optimization capabilities. With the advent of the cloud era, big data has also attracted more and more attention. Big data requires special technologies to effectively process large amounts of data within a tolerable time. Technologies suitable for big data include large-scale parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet and scalable storage systems.

[0043] The technical solution of the present application can be used in electronic devices, such as the server of the resource recommendation system mentioned above. The electronic device can be a terminal, a server, or other devices for resource recommendation, which is not limited in the present application. Optional. The server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Terminals include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, etc. The embodiments of the present invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc.

[0044] It is understood that the above scenarios are only examples and do not constitute a limitation on the application scenarios of the technical solutions provided in the embodiments of the present application. The technical solutions of the present application can also be applied to other scenarios. For example, it is known to those skilled in the art that with the evolution of the system architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0045] Based on the above description, the present application embodiment proposes a resource recommendation method. Figure 2 , Figure 2 1 is a flow chart of a resource recommendation method provided in an embodiment of the present application. The method can be executed by the electronic device mentioned above. The resource recommendation method can include the following steps.

[0046] S201. Acquire an initial resource network.

[0047] Among them, the initial resource network is constructed according to the object access behavior of N objects to M recommended resources, N and M are both positive integers, and the initial resource network includes multiple resource nodes and edges between multiple resource nodes, and one resource node corresponds to one recommended resource. The object access behavior is collected with the consent and authorization of the user. Recommended resources can be products that can be recommended, accessed or ordered in the client, such as clothing products, insurance products, etc.; the recommended resources can also be product advertisements recommended and accessed in the client, such as software application advertisements, brand advertisements, etc.; the recommended resources can also be information data, such as news, popular science pictures and texts, popular science videos, etc., which are not limited here.

[0048] In the initial resource network, an edge is used to connect two resource nodes, and an edge is used to indicate that there is at least one object that has object access behavior to both recommended resources corresponding to the two resource nodes connected by the edge. For example, if object 1 has object access behavior to both recommended resource 1 and recommended resource 2, then there can be an edge between recommended resource 1 and recommended resource 2. As mentioned above, the object access behavior can be a click behavior on a resource, a collection behavior on a resource, a behavior of clicking on a resource to display detailed content and browsing for a certain period of time, etc.

[0049] Among them, there is at least one object that has object access behavior on both recommended resources, which means that there is at least one object that accesses both recommended resources, and the two recommended resources have an access co-occurrence relationship. In other words, an edge can be used to connect resource nodes corresponding to recommended resources with an access co-occurrence relationship. For example, if object 1 accesses both recommended resource 1 and recommended resource 2, there is an access co-occurrence relationship between recommended resource 1 and recommended resource 2, and then there can be an edge between recommended resource 1 and recommended resource 2.

[0050] Among them, the N objects can be objects that generate object access behaviors within a first time range, and the first time range can be a time period between any time point before the current time and the current time. The current time can be the time for obtaining the object access behaviors of the N objects. For example, the first time range can be 1 year or 5 years before the current time, etc. The N objects can also be all objects registered in the client, and there is no limitation here. The M recommended resources can be resources that have been accessed within a second time range, and the second time range can be a time period between any time point before the current time and the current time. The second time range can be the same as or different from the first time range, and there is no limitation here. For example, the second time range can be 1 year or 5 years before the current time, etc. The M recommended resources can also be all resources that can be recommended, and there is no limitation here.

[0051] For example, the N objects may include object 1, object 2, object 3, and object 4, and the M recommended resources may include resource A, resource B, resource C, resource D, resource E, resource F, and resource G. The initial resource network constructed according to the object access behaviors of the N objects to the M recommended resources can be found in Figure 3 , Figure 3 Schematic diagram of the effect of an initial resource network provided by an embodiment of the present application. Figure 3 As shown, there is an edge between the resource node corresponding to resource A and the resource node corresponding to resource B. This is because both resource A and resource B are accessed by the same object (at least one of the above N objects). Similarly, resource A and resource F are accessed by the same object, resource B and resource C are accessed by the same object, and so on. It will not be elaborated here. There is no edge between the resource node corresponding to resource A and the resource node corresponding to resource C. This is because resource A and resource C are not accessed by any of the above N objects, that is, any of the above N objects may only access resource A, or only access resource C, or neither resource A nor resource C is accessed.

[0052] Any edge of the initial resource network may be represented as a target edge, and the target edge is used to connect the first recommended resource and the second recommended resource.

[0053] S202. Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the time interval between the target object's access to the first recommended resource and the second recommended resource, and the object activity of the target object.

[0054] The number of the target objects may be one or more, which is not limited here.

[0055] The resource access heat of the first recommended resource refers to the number of times the first recommended resource is accessed by N objects in the historical period, and the resource access heat of the second recommended resource refers to the number of times the second recommended resource is accessed by N objects in the historical period. The historical period can be the time period between the time point before the current time and the current time, and the current time can be the time for obtaining the resource access heat of the above-mentioned first recommended resource. The length of the historical period can be defined according to actual needs, such as being defined as a historical period of 1 year before the current time or 5 years before the current time, etc. The number of times accessed by N objects is the sum of the number of times the recommended resource is accessed by the above-mentioned N objects. It can be understood that if the number of times accessed by N objects in the historical period is greater, the resource access heat of the recommended resource is greater, and if the number of times accessed by N objects in the historical period is less, the resource access heat of the recommended resource is smaller. In one possible implementation, the resource access heat may be a value calculated based on the number of times the recommended resource is accessed by N objects in a historical period. For example, the number of times the recommended resource is accessed by N objects in a historical period may be divided by the length of the historical period to obtain the corresponding resource access heat. For another example, the number of times the recommended resource is accessed by N objects in a historical period may be divided by the time difference between the first access and the last access in the historical period to obtain the corresponding resource access heat. No limitation is imposed here.

[0056] The object activity of the target object refers to the number of resource accesses of the target object to the M recommended resources in the historical period. The number of resource accesses to the M recommended resources in the historical period is the sum of the number of resource accesses of the target object to the above-mentioned M recommended resources. It can be understood that if the number of resource accesses to the M recommended resources in the historical period is more, the greater the object activity of the target object is, and if the number of resource accesses to the M recommended resources in the historical period is less, the smaller the object activity of the target object is. In a possible implementation, the object activity of the target object can be a value obtained by calculating the number of resource accesses of the target object to the M recommended resources in the historical period. For example, the number of resource accesses to the M recommended resources in the historical period can be divided by the length of the historical period to obtain the corresponding object activity. For example, the number of resource accesses to the M recommended resources in the historical period can be divided by the time difference between the first access to the recommended resource and the last access to the recommended resource in the historical period to obtain the corresponding object activity. There is no limitation here.

[0057] The time interval between the target object's access to the first recommended resource and the second recommended resource refers to the time interval between the time the target object accesses the first recommended resource and the time the target object accesses the second recommended resource. For example, the time the target object accesses the first recommended resource is t1, and the time the target object accesses the second recommended resource is t2, then the access time interval is t2-t1. In one possible implementation, the target object may access the first recommended resource and the second recommended resource multiple times, then the minimum time interval between the target object's access to the first recommended resource and the second recommended resource may be used as the above-mentioned access time interval. For example, the time the target object accesses the first recommended resource is t1, t2, and t3, and the time the target object accesses the second recommended resource is t4, t5, and t6, among which the time interval between t4 and t3 is the minimum, then t4-t3 is the time interval between the target object's access to the first recommended resource and the second recommended resource.

[0058] S203: Determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity.

[0059] Among them, the edge weight of the target edge can be used to indicate the degree of association between the two recommended resources connected by the target edge. The larger the edge weight, the higher the degree of association between the two recommended resources connected. If the edge weight is smaller, the lower the degree of association between the two recommended resources connected. It can be understood that the degree of association between two recommended resources can affect the subsequent recommendation of resources. Generally speaking, recommending resources for objects is to recall resources based on the degree of association between the resources visited by the object and each recommended resource. Therefore, when calculating the edge weight in this step, a penalty is imposed on the resource access heat and object activity of the resource to reduce the edge weights between the recommended resources with high heat and the recommended resources visited by the objects with high activity, so as to reduce the probability of recalling the recommended resources with high heat and the recommended resources visited by the objects with high activity, thereby reducing the popularity deviation of resource recommendations.

[0060] In a possible implementation manner, the edge weight of the target edge determined in the embodiment of the present application can be calculated by the following formula (1):

[0061]

[0062] Wherein, i in formula (1) represents recommended resource i, j represents recommended resource j, and recommended resource i and recommended resource j can be any one of the M recommended resources, and recommended resource i and recommended resource j are not the same resource. ijThat is, it represents the edge weight of the edge between the resource node corresponding to the recommended resource i and the resource node corresponding to the recommended resource j. u represents any object that accesses both the recommended resource i and the recommended resource j. ui -t uj ) means the time t that any object accesses the recommended resource i ui and the time t at which the recommended resource j is accessed uj The time interval between visits. u That is, it indicates the object activity of any of the above objects. i represents the resource access popularity of the recommended resource i, p j represents the resource access popularity of the recommended resource j. Then, 1 / δ(t ui -t uj )*log(1+q u )*p i p j ) (recorded as the target value), and then the target values ​​corresponding to each object that accesses both recommended resource i and recommended resource j are added together to obtain the edge weight of the edge between the resource nodes corresponding to recommended resource i and recommended resource j. It can be understood that formula (1) takes into account the access interval between recommended resource i and recommended resource j. The shorter the access interval, the greater the similarity between the two recommended resources, and the higher the edge weight should be; formula (1) takes into account the object activity of the object that has object access behavior to both recommended resource i and recommended resource j. The higher the object activity, the smaller the corresponding edge weight, thereby achieving the penalty for the correlation between the resources accessed by the active objects; formula (1) takes into account the resource access heat of recommended resource i and recommended resource j. The higher the resource access heat, the smaller the corresponding edge weight. Therefore, by imposing penalties on the resource access popularity of the two recommended resources and the object activity of the objects that access both recommended resources, the weight value of the edge between the resource nodes corresponding to the two recommended resources is reduced, which is equivalent to reducing the correlation between popular resources and between the resources accessed by active objects. When recommending resources subsequently, the chances of popular resources being recommended can be reduced, and the chances of unpopular resources being recommended can be increased, which helps to reduce the popularity bias problem of resource recommendations and improve the accuracy and diversity of resource recommendations for objects.

[0063] S204: Determine the initial resource network having the edge weight of the target edge as the target resource network, and recommend recommended resources to the N objects according to the target resource network.

[0064] The target resource network may be a resource network in which edge weights are added to each edge based on the initial resource network. It is understandable that since the target edge is any edge in the initial resource network, the corresponding edge weights of each edge in the generated target resource network can be calculated through steps S202-S203.

[0065] For example, see Figure 4 , Figure 4 : is a schematic diagram of the effect of a target resource network provided by an embodiment of the present application. The target resource network can be Figure 3 Each edge in the initial resource network shown determines the resource network with the corresponding edge weight, such as Figure 4 The edge weight of the edge between the resource node corresponding to resource A and the resource node corresponding to resource B is 0.8, the edge weight of the edge between the resource node corresponding to resource B and the resource node corresponding to resource C is 0.7, and so on, which will not be elaborated here.

[0066] In a possible implementation, the recommendation of recommended resources for N objects according to the target resource network can be determined according to the recommended resources corresponding to the resource nodes with edges between the resource nodes of the historically visited recommended resources of the object in the target resource network, that is, determined according to the recommended resources corresponding to the candidate resource nodes associated with the historically visited recommended resources. Among them, the historically visited recommended resources can be the recommended resources that the object has visited in the above-mentioned historical period. For example, the recommendation of recommended resources for N objects according to the target resource network can include the following steps: for each object, the resource nodes with edges between the resource nodes of the historically visited recommended resources of the object in the target resource network are determined as candidate resource nodes associated with the historically visited recommended resources; the candidate resource nodes are sorted in descending order according to the edge weights between each candidate resource node and the resource node of the historically visited recommended resources, and the recommended resources in the top R of the sorting are used as the recommended resources for recommending the object. Among them, if there are multiple recommended resources for the object's historical access, then according to the recommended resources ranked in the top R as the recommended resources for the object, the union or intersection of the recommended resources ranked in the top R among the candidate resource nodes associated with each historically accessed recommended resource can be determined as the recommended resource for the object. In this way, the resources to be recommended can be quickly determined directly based on the edge weights between the resource nodes of the recommended resources and the resource nodes of the recommended resources for the object's historical access. Since the target resource network is penalized by the resource access heat of the resources and the object activity of the objects, the recommended resources determined here for the object are less affected by the resource access heat and the object activity. Some resources with lower resource access heat and lower object activity can also be recommended, which reduces the popularity bias of resource recommendations and increases the diversity of recommended resources.

[0067] In a possible implementation, the edge weights between each candidate resource node and the resource node of the historically visited recommended resource are sorted in descending order, and after the recommended resources ranked in the top R are determined, the recommended resources ranked in the top R among the candidate resource nodes associated with each historically visited recommended resource can be scored to determine the recommended resource for the object based on the score. For a specific description, please refer to the subsequent Figure 6 The relevant description of the embodiments is not repeated here.

[0068] The embodiment of the present application can construct an initial resource network based on the object access behavior of the object to the recommended resources, and for any two recommended resources accessed by the same object, the edge weights of the edges corresponding to the two recommended resources in the initial resource network can be determined based on the resource access heat of the two recommended resources, the object activity of the objects that access both the two recommended resources, and the time interval for the same object to access the two recommended resources, so as to obtain the target resource network, and then recommend the recommended resources based on the target resource network. Therefore, when constructing the target resource network, the weight values ​​of the edges between the resource nodes corresponding to the two recommended resources can be determined by factors such as the resource access heat of the two recommended resources and the object activity of the objects that access both the recommended resources, which can improve the accuracy of the determined edge weights, and then recommend resources to the object through the target resource network with more accurate edge weights, thereby improving the accuracy and diversity of resource recommendations for the object.

[0069] See also Figure 5 , Figure 5 1 is a flow chart of a resource recommendation method provided in an embodiment of the present application. The method can be executed by the electronic device mentioned above. The resource recommendation method can include the following steps.

[0070] S501: Acquire an initial resource network.

[0071] S502. Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the time interval between the target object's access to the first recommended resource and the second recommended resource, and the object activity of the target object.

[0072] Among them, steps S501-S502 can refer to steps S201-S202 and are not repeated here.

[0073] S503: Calculate the access co-occurrence weight between the first recommended resource and the second recommended resource according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity.

[0074] The access co-occurrence weight may be a weight value of the edge between the recommended resources having the access co-occurrence relationship. The access co-occurrence weight between the first recommended resource and the second recommended resource is the access co-occurrence weight of the edge between the resource node corresponding to the first recommended resource and the resource node corresponding to the second recommended resource. Referring to the description of step S203, the access co-occurrence weight between the first recommended resource and the second recommended resource may be calculated by formula (1), that is, W in formula (1) is ijIndicates the access co-occurrence weight between the first recommended resource and the second recommended resource, which will not be described here.

[0075] S504: Obtain resource similarity between the first recommended resource and the second recommended resource, and determine a similarity weight between the first recommended resource and the second recommended resource according to the resource similarity.

[0076] Among them, the resource similarity can be the similarity of the associated resource data of the recommended resource, and the associated resource data can be at least one of the associated text data and the associated image data. For example, the associated text data can be the title, introduction, etc. of the recommended resource, and the associated image data can be the illustrations in the recommended resource, the cover image of the recommended resource, etc., which are not limited here. The similarity weight can be a weight value used to represent the similarity between the recommended resources. The similarity weight between the first recommended resource and the second recommended resource, that is, the similarity weight of the edge between the resource node corresponding to the first recommended resource and the resource node corresponding to the second recommended resource. The greater the resource similarity, the greater the similarity weight between the recommended resources, and the smaller the resource similarity, the smaller the similarity weight between the recommended resources. It can be understood that by introducing the similarity weight, the correlation between the recommended resources can be reflected from the recommended resources themselves, so that the edge weights of each edge in the resource network finally determined can more accurately reflect the correlation between the recommended resources, which in turn helps to determine the recommended resources with a higher correlation with the historical access objects of the object for recommendation, thereby improving the accuracy of resource recommendation.

[0077] In a possible implementation, obtaining the resource similarity between the first recommended resource and the second recommended resource may specifically include the following steps: ① obtaining the first associated resource data of the first recommended resource. The first associated resource data includes at least one of the associated text data and associated image data of the first recommended resource. ② obtaining the second associated resource data of the second recommended resource. The second associated resource data includes at least one of the associated text data and associated image data of the second recommended resource. ③ generating a first embedding vector of the first associated resource data, and generating a second embedding vector of the second associated resource data. The first embedding vector may be an embedding vector obtained by processing the associated text data of the first recommended resource according to a natural language processing model (NLP), the first embedding vector may also be an embedding vector obtained by processing the associated image data of the first recommended resource according to an image processing model (NLP), or the first embedding vector may also be an embedding vector obtained by combining an embedding vector obtained by processing the associated text data of the first recommended resource and an embedding vector obtained by processing the associated image data of the first recommended resource, which is not limited here. Similarly, the second embedding vector may be an embedding vector obtained by processing the associated text data of the second recommended resource according to a natural language processing model (NLP), or an embedding vector obtained by processing the associated image data of the second recommended resource according to an image processing model (NLP), or an embedding vector obtained by combining an embedding vector obtained by processing the associated text data of the second recommended resource and an embedding vector obtained by processing the associated image data of the second recommended resource, which is not limited here. ④ Determine the resource similarity between the first recommended resource and the second recommended resource according to the first embedding vector and the second embedding vector. In the step ④, the resource similarity is determined, and the cosine similarity of the first embedding vector and the second embedding vector can be calculated as the resource similarity between the first recommended resource and the second recommended resource. Thus, the similarity of the recommended resource itself can be reflected by the similarity of the embedding vector, and the embedding vector corresponding to the recommended resource can be quickly determined by various pre-trained image processing models or natural language processing models, which greatly improves the calculation rate of resource similarity.

[0078] In a possible implementation, the similarity weight between the first recommended resource and the second recommended resource is determined according to the resource similarity. The similarity weight between the first recommended resource and the second recommended resource can be directly determined according to the resource similarity. For details, see the following formula (2):

[0079] W ij = cosine(w i *w j ) (2)

[0080] Where i in formula (2) represents recommended resource i, j represents recommended resource j, and W ij That is, it represents the similarity weight between recommended resource i and recommended resource j. i represents the embedding vector corresponding to the recommended resource i, w j represents the embedding vector of the recommended resource j, cosine(w i *w j ) represents the cosine similarity between recommended resource i and recommended resource j.

[0081] In a possible implementation, before determining the similarity weight between the first recommended resource and the second recommended resource according to the resource similarity, the following steps may also be included: ① Determine the resource similarity between the first recommended resource and other recommended resources among the above-mentioned M recommended resources, and sort the resource similarities from large to small, and determine the recommended resources ranked in the first S as the similar recommended resource set associated with the first recommended resource, where S is a positive integer less than or equal to M. Among them, the similar recommended resource set may include S recommended resources. The value of S may be a preset fixed value, such as S may be a value corresponding to a certain percentage of M, such as the value of S is 60% M. To determine the resource similarity between the first recommended resource and other recommended resources among the above-mentioned M recommended resources, the cosine similarity between the first recommended resource and each other recommended resource may be calculated, which will not be elaborated here. The method for determining the S recommended resources may also be the K nearest neighbor method, and the S recommended resources are the recommended resources closest to the first recommended resource. For example, the M recommended resources may include recommended resources A, B, C, D, and E. The resource similarity between recommended resource A and recommended resources B, C, D, and E can be calculated respectively to obtain the resource similarity {R AB , R AC , R AD , R AE}, and the ranking of the similarity of each resource is: R AB >R AD >R AC >R AE, if it is necessary to determine that the recommended resources ranked in the top 3 are the similar recommended resource set associated with recommended resource A, then the similar recommended resource set associated with recommended resource A includes recommended resources B, D, and C. ② Determine the resource similarity between the second recommended resource and other recommended resources in the above M recommended resources, and sort the similarities of each resource from large to small, and determine that the recommended resources ranked in the top S are the similar recommended resource set associated with the second recommended resource. The method for determining the similar recommended resource set associated with the second recommended resource here can refer to the relevant description in step ① here, and will not be repeated here. ③ If the similar recommended resource set associated with the first recommended resource includes the second recommended resource, or the similar recommended resource set associated with the second recommended resource includes the first recommended resource, then determine the similarity weight between the first recommended resource and the second recommended resource based on the resource similarity. The method for determining the similarity weight in step ③ here can refer to the above formula (2) to calculate the similarity weight, and will not be repeated here. ④ If the set of similar recommended resources associated with the first recommended resource does not include the second recommended resource, and the set of similar recommended resources associated with the second recommended resource does not include the first recommended resource, the similarity weight between the first recommended resource and the second recommended resource is determined to be zero. In other words, if the first recommended resource is a similar recommended resource of the second recommended resource, or the second recommended resource is a similar recommended resource of the first recommended resource, it means that the similarity between the two is high, and it is necessary to calculate the similarity weight based on the resource similarity. If the first recommended resource and the second recommended resource are not similar recommended resources of each other, it means that the similarity between the two is low. Therefore, the similarity weights between some recommended resources with low similarity can be directly ignored, thereby reducing the association between recommended resources with low similarity, and reducing the amount of calculation for determining the edge weights of each edge in the subsequent process, thereby improving the calculation rate.

[0082] S505: Determine the edge weight of the target edge according to the access co-occurrence weight and the similarity weight.

[0083] Among them, the edge weight of the target edge can be determined by directly adding the access co-occurrence weight and the similarity weight to obtain the corresponding edge weight. For example, the access co-occurrence weight between the recommended resource i and the recommended resource j is w1, and the similarity weight is w2, then the edge weight between the recommended resource i and the recommended resource j is w1+w2. In a possible implementation, the edge weight of the target edge can be determined by multiplying the access co-occurrence weight by a certain coefficient and the similarity weight by a certain coefficient and then adding them to obtain the corresponding edge weight. For example, the access co-occurrence weight between the recommended resource i and the recommended resource j is w1, and the similarity weight is w2, then the edge weight between the recommended resource i and the recommended resource j is k1*w1+k2*w2, where k1 and k2 are both constants. If k1 is greater than k2, it means that the access co-occurrence weight has a greater influence in the final calculated edge weight. If k1 is less than k2, it means that the similarity weight has a greater influence in the final calculated edge weight. This can achieve the customization of the influence between the access co-occurrence weight and the similarity weight in the edge weight.

[0084] In one possible implementation, the initial resource network may also include edges between recommended resources and recommended resources in an associated similar recommended resource set, that is, between two recommended resources that are not accessed by the same object, if the two recommended resources are similar to each other, then an edge may also be constructed between the resource nodes corresponding to the two recommended resources, and the access co-occurrence weight between the two recommended resources is zero, that is, the edge weight between the resource nodes corresponding to the two recommended resources is determined according to the resource similarity of the two recommended resources.

[0085] For example, in a possible scenario, resource i has just been released and has not been visited by the object, that is, there is no click co-occurrence relationship between it and any other recommended resources, but the resource similarity between the associated resource data of resource i and resource j is very high. Obviously, in this case, there is actually a certain association between resource i and resource j. If the object visits resource j, since resource i and resource j are similar, it is very likely that the object is also interested in resource i. When constructing the initial resource network, the edge between resource i and resource j can be constructed in the initial resource network. The edge weight of the edge is the similarity weight between resource i and resource j, so as to reflect the association between resource i and resource j. In this way, resources that lack object access behavior can also be associated with similar recommended resources, avoiding the recommended resources from not being recommended due to the lack of object access behavior, and reducing the popularity bias of recommended resources.

[0086] S506: Determine the initial resource network having the edge weight of the target edge as the target resource network, and recommend recommended resources to the N objects according to the target resource network.

[0087] Among them, step S506 can refer to the relevant description of step S204, which is not repeated here.

[0088] The embodiment of the present application can construct an initial resource network based on the object access behavior of the object to the recommended resources, and for any two recommended resources accessed by the same object, the edge weights of the edges corresponding to the two recommended resources in the initial resource network can be determined based on the resource access heat of the two recommended resources, the object activity of the objects that access both the two recommended resources, and the time interval for the same object to access the two recommended resources, so as to obtain the target resource network, and then recommend the recommended resources based on the target resource network. Therefore, when constructing the target resource network, the weight values ​​of the edges between the resource nodes corresponding to the two recommended resources can be determined by factors such as the resource access heat of the two recommended resources and the object activity of the objects that access both the recommended resources, which can improve the accuracy of the determined edge weights, and then recommend resources to the object through the target resource network with more accurate edge weights, thereby improving the accuracy and diversity of resource recommendations for the object.

[0089] See also Figure 6 , Figure 6 1 is a flow chart of a resource recommendation method provided in an embodiment of the present application. The method can be executed by the electronic device mentioned above. The resource recommendation method can include the following steps.

[0090] S601: Acquire an initial resource network.

[0091] S602. Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the time interval between the target object's access to the first recommended resource and the second recommended resource, and the object activity of the target object.

[0092] S603: Determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity.

[0093] S604: Determine an initial resource network having the edge weight of the target edge as the target resource network.

[0094] Among them, the relevant descriptions of steps S601-S604 can refer to steps S201-S204, and are not repeated here.

[0095] S605: Acquire a candidate recommended resource set associated with each recommended resource according to the target resource network.

[0096] For each recommended resource, the candidate recommended resource set can be understood as a collection of recommended resources that have a high degree of association with the recommended resource.

[0097] In a possible implementation, if any one of the M recommended resources is represented as the pth recommended resource, i is a positive integer and is less than or equal to M, then step S605 may specifically include the following steps: ① Determine a resource node in the target resource network that has an edge with the resource node of the pth recommended resource as at least one candidate resource node. The edge may be an edge that includes both access co-occurrence weights and similarity weights, or an edge that only includes access co-occurrence weights, or an edge that only includes similarity weights, which is not limited here. ② According to the edge weight of the edge between each candidate resource node and the resource node of the pth recommended resource, sort the recommended resources corresponding to each candidate resource node in order from large to small. ③ Determine the candidate recommended resource set associated with the pth recommended resource based on the recommended resources ranked in the top K, where K is a positive integer. Among them, the recommended resources ranked in the top K indicate that the correlation between these K recommended resources and the p-th recommended resource is relatively high, and thus the K recommended resources can be used as recommended resources in the candidate recommended resource set associated with the p-th recommended resource. In a possible implementation, the value of K can be a preset fixed value. If the number of candidate resource nodes is less than K, the recommended resources corresponding to all candidate resource nodes can be determined as recommended resources in the candidate recommended resource set. For example, the K value can be 10, and the recommended resources ranked in the top 10 can be determined as the candidate recommended resource set associated with the p-th recommended resource. If the number of candidate resource nodes is 5, and 5 is less than 10, the recommended resources corresponding to the 5 candidate resource nodes can all be determined as the candidate recommended resource set associated with the p-th recommended resource. In a possible implementation, the value of K can be determined according to the number of candidate resource nodes, such as determining the value of K according to a certain percentage of the number of candidate resource nodes. For example, the value of K is determined based on 40% of the number of candidate resource nodes. If the number of candidate resource nodes is 10, the value of K can be 10*40%, and K=4. If there is no edge between the pth recommended resource and other recommended resources, the candidate recommended resource set associated with the recommended resource can be empty. It can be understood that if the pth recommended resource is any one of the M recommended resources, the candidate recommended resource set corresponding to each of the M recommended resources can be determined by the above method.

[0098] For example, see Figure 7 , Figure 7 Schematic diagram of the effect of generating a candidate recommended resource set provided by an embodiment of the present application. Figure 7 As shown, Figure 7(1) in FIG. 1 shows a target resource network, which includes resource nodes corresponding to resource AL, and the edges between resource nodes have corresponding edge weights. Figure 7 When the candidate recommended resource set associated with the resource node B in (1) is selected, at least one candidate resource node connected to the resource node corresponding to resource B can be as follows Figure 7 As shown in (2) in the figure, the candidate resource nodes corresponding to resource B include nodes corresponding to the six resources: resource A, resource D, resource G, resource C, resource H, and resource I (hereinafter referred to as A, D, G, C, H, and I). Figure 7 From (2), we can see that the edge weights of the edges connecting each candidate resource node with the resource node corresponding to resource B can be sorted from large to small as follows: A(0.8)>C(0.7)>D(0.55)>H(0.38)>G(0.35)>I(0.3). If the preset K value is 3, the top three resources can be determined as the candidate recommended resource set associated with resource B, that is, the candidate recommended resource set associated with resource B includes resource A, resource B, and resource D.

[0099] S606: Acquire the historically visited recommended resources of the qth object, and determine the candidate recommended resource set associated with the historically visited recommended resources as the target recommended resource set.

[0100] Any one of the N objects can be represented as the qth object. As mentioned above, the historically visited recommended resource can be a recommended resource visited by the qth object in the historical period.

[0101] The target recommended resource set may be a candidate recommended resource set associated with the historically visited recommended resources. It is understandable that if the qth object has multiple historically visited recommended resources, then there are also multiple target recommended resource sets.

[0102] S607: Construct a training sample pair for the qth object according to the historically accessed recommended resources and the target recommended resource set.

[0103] A training sample pair may include a historically visited recommended resource and a recommended resource in a target recommended resource set associated with the historically visited recommended resource. If the historically visited recommended resource is denoted as source ad and a recommended resource in a target recommended resource set associated with the historically visited recommended resource is denoted as target ad, then the training sample pair may be <source ad，target ad>, thus, training sample pairs can be generated based on recommended resources with higher edge weights among recommended resources directly connected to historically visited recommended resources in the target resource network. In a possible implementation, corresponding training sample pairs can be generated based on each recommended resource in the target recommended resource set of historically visited recommended resources, or corresponding training sample pairs can be generated from randomly selected recommended resources in the target recommended resource set of historically visited recommended resources, which is not limited here.

[0104] In a possible implementation, each training sample pair may be associated with a sample label, and the sample label may be determined based on whether the recommended resource that the qth object visits next after visiting the historically visited recommended resource is another recommended resource in the training sample. If the recommended resource that the qth object visits next after visiting the historically visited recommended resource is another recommended resource in the training sample, the sample label associated with the training sample may be a positive sample label, such as the sample label may be determined as 1. If the recommended resource that the qth object visits next after visiting the historically visited recommended resource is not another recommended resource in the training sample, the sample label associated with the training sample may be a negative sample label, such as the sample label may be determined as 0.

[0105] S608: Train the recommendation model according to the training samples to obtain a trained recommendation model.

[0106] The trained recommendation model may determine a similarity score between two input recommendation resources, and the similarity score may be used to determine a recommendation index of each recommendation resource in the target recommendation resource set of the qth object for the qth object.

[0107] In a possible implementation, training the recommendation model according to the training sample pair may specifically include the following steps: ① Obtain resource features of two recommended resources in the training sample pair, and object features of the qth object. The resource features may include time features of the resource, such as the time when the resource is released, the time when the resource is accessed, etc.; the resource features may also include access features of the resource, such as the number of times the resource is accessed by the qth object in the historical period, the time difference between each time the resource is accessed by the qth object in the historical period, etc., and the feature information of the object is introduced from the side through the time difference between the accesses of the two recommended resources and the accessed places, so that the subsequent determination of the similarity score of the two recommended resources is associated with the access behavior of the object; the resource features may also include network features of the resource, such as the category to which the resource belongs (such as clothing advertisements, furniture advertisements, etc.), the object data label of the object accessing the resource (such as the label of the style of the resource preferred by the object, etc.), etc., which are not limited here. The object features may include basic attribute features of the qth object, such as the identity identification code, name, etc. of the object, which are not limited here. The object features of the object are all obtained with the consent and authorization of the user. ② Input the training sample pair, the resource features of the two recommended resources in the training sample pair, and the object features into the initial recommendation model to obtain the predicted similarity score of the training sample pair. The predicted similarity score can be understood as the score of the similarity of the resource features of the two recommended resources in the training sample pair calculated by the initial recommendation model. In a possible implementation, the initial recommendation model can be some common binary classification models, such as a neural network model, a decision tree model, a logistic regression model, etc., which are not limited here, and the initial recommendation model can learn the similarity score between the two recommended resources in the input training sample pair. ③ Adjust the model parameters of the initial recommendation model according to the predicted similarity score and the sample label to obtain a trained recommendation model. The purpose of the model parameters of the initial recommendation model is to make the gap between the similarity score that can be calculated by the trained recommendation model and the numerical value of the sample label smaller. For example, the target loss function can be called to calculate the loss value based on the predicted similarity score and the sample label, and then the model parameters of the initial recommendation model can be adjusted according to the loss value. The initial recommendation model is iteratively trained through sample training until the loss value of the loss function is less than a certain threshold or the training reaches a certain round, and then a trained recommendation model can be obtained. In this way, a recommendation model can be trained to quickly determine the similarity score between the historical access resources of the qth object and the recommended resources in the associated target recommendation resource set.

[0108] In a possible implementation, the target loss function used to calculate the loss value based on the predicted similarity score and the sample label can be shown as the following formula (3):

[0109] L=(α+β)ylogh+(1-y)log(1-h) (3)

[0110] Where y is the value of the sample label, h is the predicted similarity score, and L is the calculated loss value. α can be the popularity weight, and β is the positive sample weight.

[0111] Among them, the popularity weight can be a weight value preset according to the popularity of the recommended resources in the target recommended resource set in the training sample pair. For different popularity, different popularity weights can be determined as parameter values ​​in the target loss function. If the popularity of the recommended resources in the target recommended resource set in the training sample pair is higher, the corresponding popularity weight is lower, and vice versa, the popularity weight is higher. In a possible implementation, the popularity of the recommended resource can be determined based on the number of occurrences of the recommended resource in at least one target recommended resource set corresponding to the qth object. If the number of occurrences in the target recommended resource set is greater, the corresponding popularity is higher, and if the number of occurrences in the target recommended resource set is less, the corresponding popularity is lower. It can be understood that compared to not using different popularity weights for recommended resources of different popularity, the present application can use loss functions with different parameter values ​​for training sample pairs including recommended resources of different popularity during the training process of the recommendation model by setting the popularity weight. If the popularity weight in the loss function required for training the training sample pairs corresponding to the recommended resources with higher popularity in the target recommended resource set is lower, then the similarity score calculated by the final trained recommendation model between the historically visited recommended resources and the recommended resources with higher popularity in the target recommended resource set can be reduced, thereby subsequently reducing the chance of recommended resources with higher popularity being recommended, and increasing the chance of recommended resources with higher popularity being recommended, thereby reducing the popularity deviation of resource recommendations.

[0112] The positive sample weight may be a preset weight for influencing the predicted similarity score of the training sample pairs associated with the positive sample labels. By setting the positive sample weight, the parameter values ​​in different loss functions are affected, so that the recommendation model can use loss functions including different parameter values ​​for the training sample pairs associated with different sample labels during the training process, and then the similarity score calculated by the finally trained recommendation model between the historically visited recommended resources of the object and the recommended resources in the target recommended resource set that have also been visited by the object can be reduced or increased according to the positive sample weight.

[0113] S609: Call the trained recommendation model to determine the recommended resource for recommending the qth object from the target recommended resource set.

[0114] Among them, calling the trained recommendation model can directly determine the similarity score between the historical access recommendation resources of the qth object and each recommendation resource in the target recommendation resource set, and then determine the recommendation index of each recommendation resource in the target recommendation resource set according to the similarity score corresponding to each recommendation resource in the target recommendation resource set, so as to determine some recommendation resources with larger recommendation indexes to recommend the qth object in the future. The recommendation index is used to indicate the degree of recommendation for the recommended resource. If the recommendation index is higher, it means that the recommended resource is more likely to be recommended to the object.

[0115] In a possible implementation, if the number of historically visited recommended resources for the qth object is one, the similarity scores between each recommended resource in the target recommended resource set and the historically visited recommended resources can be directly used as the recommendation index of each recommended resource in the target recommended resource set for the qth object.

[0116] In a possible implementation, if the number of historically visited recommended resources of the qth object is multiple, and the same recommended resources may appear in the target recommended resource sets associated with different historically visited recommended resources, then the recommendation index of each recommended resource in the target recommended resource set is determined according to the similarity scores corresponding to each recommended resource in the target recommended resource set, which may specifically include the following steps: determining the recommended resources that appear multiple times in the target recommended resource sets associated with each historically visited recommended resource, and determining the average value of the similarity scores between the recommended resources that appear multiple times and the historically visited recommended resources associated with each target recommended resource set to which they belong, and determining the average value of the similarity scores as the recommendation index of the recommended resources that appear multiple times for the qth object. Among them, the recommended resources that appear multiple times are used to indicate the recommended resources included in at least two target recommended resource sets. For example, among multiple historically visited recommended resources for the qth object, the target recommended resource sets associated with historically visited recommended resources 1, 2, and 3 all include recommended resource r. It can be determined that the similarity score between recommended resource r and historically visited recommended resource 1 is G1, the similarity score between recommended resource r and historically visited recommended resource 2 is G2, and the similarity score between recommended resource r and historically visited recommended resource 3 is G3. The average value of G1, G2, and G3 can be determined as the recommendation index of recommended resource r for the qth object.

[0117] In a possible implementation, step S609 may specifically include the following steps: ① Calling a trained recommendation model to predict the recommendation index of each recommended resource in the target recommended resource set for the qth object. Among them, generating the recommendation index of each recommended resource in the target recommended resource set for the qth object can refer to the above method, which will not be repeated here. ② Sort each recommended resource in the target recommended resource set according to the recommendation index corresponding to each recommended resource in the target recommended resource set. It can be understood that if the number of target recommended resource sets is multiple, when sorting each recommended resource in the target recommended resource set, the union of each recommended resource in the target recommended resource set is sorted, that is, when sorting, each recommended resource appears only once. ③ Determine the recommended resources ranked in the top L in the target recommended resource set as the recommended resources recommended for the qth object; L is a positive integer. Among them, the recommended resources ranked in the top L are the recommended resources with the largest recommendation indexes, so that the recommended resources with larger recommendation indexes can be finally determined to be recommended to the object. Moreover, since the recommendation index is sorted based on each target recommendation resource set, and the loss function used in the training of the recommendation model for calculating the similarity score also introduces popularity weights, unpopular recommended resources can also be recommended, reducing the exposure opportunities of popular recommended resources, thereby reducing the popularity bias of resource recommendations and increasing the diversity of resource recommendations.

[0118] In one possible application scenario, see Figure 8 , Figure 8It is a flow chart of a resource recommendation method provided by an embodiment of the present application. Among them, the above-mentioned recommended resources can be advertisements. First, the click behavior of N objects for M recommended resources (i.e., a resource access behavior) can be obtained, and the access co-occurrence weights between advertisements with click co-occurrence relationships can be calculated (i.e., step S801). If there is a click co-occurrence relationship between two advertisements, it means that there is at least one object that has clicked on both advertisements; then an embedding vector of the associated resource data of the advertisement (such as the image, title, etc. of the advertisement) is generated (i.e., step S802) to calculate the similarity weights between each advertisement (i.e., step S803), and the edge weights of the edges between the resource nodes corresponding to each advertisement are determined based on the access co-occurrence weights and similarity weights between each advertisement (i.e., step S804), so as to obtain the target resource network according to the resource nodes corresponding to each advertisement and the edge weights of the edges between each resource node. The edge weights of the edges directly connected to any advertisement are sorted from large to small, and the advertisements connected to the edge weights ranked in the top K are determined as the candidate recommended resource set associated with any advertisement (i.e., step S805). Training sample pairs are generated based on the advertisements in the candidate recommended resource set and the historical clicked advertisements of the object (i.e., historical access resources), and a recommendation model is trained (i.e., step S806). The trained recommendation model is called to determine the recommendation index of the advertisements in the candidate recommended resource set associated with the historical clicked advertisements of the object, so as to sort the advertisements in the candidate recommended resource set associated with the historical clicked advertisements according to the recommendation index (step S807), and then determine the advertisements recommended for the object according to the sorting of the recommendation index (step S808).

[0119] The embodiment of the present application can construct an initial resource network based on the object access behavior of the object to the recommended resources, and for any two recommended resources accessed by the same object, the edge weights of the edges corresponding to the two recommended resources in the initial resource network can be determined based on the resource access heat of the two recommended resources, the object activity of the objects that access both the two recommended resources, and the time interval for the same object to access the two recommended resources, so as to obtain the target resource network, and then recommend the recommended resources based on the target resource network. Therefore, when constructing the target resource network, the weight values ​​of the edges between the resource nodes corresponding to the two recommended resources can be determined by factors such as the resource access heat of the two recommended resources and the object activity of the objects that access both the recommended resources, which can improve the accuracy of the determined edge weights, and then recommend resources to the object through the target resource network with more accurate edge weights, thereby improving the accuracy and diversity of resource recommendations for the object.

[0120] See also Fig. 9 , Fig. 9is a schematic diagram of the structure of a resource recommendation device provided in an embodiment of the present application. Optionally, the resource recommendation device may be provided in the above-mentioned electronic device. Fig. 9 As shown, the resource recommendation device described in this embodiment may include:

[0121] The acquisition unit 901 is used to acquire an initial resource network; the initial resource network is constructed according to the object access behaviors of N objects to M recommended resources, N and M are both positive integers, the initial resource network includes multiple resource nodes and multiple edges between the resource nodes, one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors to the recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is represented as a target edge, and the target edge is used to connect the resource node corresponding to the first recommended resource and the resource node corresponding to the second recommended resource;

[0122] Processing unit 902 is used to determine the object that has object access behavior to both the first recommended resource and the second recommended resource among the N objects as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval of the target object to the first recommended resource and the second recommended resource, and the object activity of the target object;

[0123] The processing unit 902 is further configured to determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0124] The processing unit 902 is further configured to determine the initial resource network having the edge weight of the target edge as the target resource network, and recommend recommended resources to the N objects according to the target resource network.

[0125] In one implementation, the resource access popularity of the first recommended resource refers to the number of times the first recommended resource is accessed by N objects in a historical period, and the resource access popularity of the second recommended resource refers to the number of times the second recommended resource is accessed by N objects in a historical period; object activity refers to the number of resource accesses of the target object to M recommended resources in a historical period.

[0126] In one implementation, the processing unit 902 is specifically configured to:

[0127] Calculate the access co-occurrence weight between the first recommended resource and the second recommended resource according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0128] Obtaining resource similarity between the first recommended resource and the second recommended resource, and determining a similarity weight between the first recommended resource and the second recommended resource according to the resource similarity;

[0129] The edge weight of the target edge is determined according to the access co-occurrence weight and the similarity weight.

[0130] In one implementation, the processing unit 902 is specifically configured to:

[0131] Acquire first associated resource data of the first recommended resource; the first associated resource data includes at least one of associated text data and associated image data of the first recommended resource;

[0132] Acquire second associated resource data of the second recommended resource; the second associated resource data includes at least one of associated text data and associated image data of the second recommended resource;

[0133] generating a first embedding vector for the first associated resource data, and generating a second embedding vector for the second associated resource data;

[0134] The resource similarity between the first recommended resource and the second recommended resource is determined according to the first embedding vector and the second embedding vector.

[0135] In one implementation, any one of the N objects is represented as the qth object, where q is a positive integer and q is less than or equal to N; the processing unit 902 is specifically configured to:

[0136] Obtain a candidate recommended resource set associated with each recommended resource according to the target resource network;

[0137] Obtain the historically visited recommended resources of the qth object, and determine the candidate recommended resource set associated with the historically visited recommended resources as the target recommended resource set;

[0138] Construct a training sample pair for the qth object based on the historical access recommendation resources and the target recommendation resource set;

[0139] The recommendation model is trained according to the training samples to obtain a trained recommendation model;

[0140] The trained recommendation model is called to determine the recommended resource for the qth object from the target recommendation resource set.

[0141] In one implementation, any one of the M recommended resources is represented as the p-th recommended resource, where p is a positive integer and is less than or equal to M; the processing unit 902 is specifically configured to:

[0142] Determine a resource node in the target resource network that has an edge connecting to the resource node of the p-th recommended resource as at least one candidate resource node;

[0143] According to the edge weight of the edge between each candidate resource node and the resource node of the pth recommended resource, the recommended resources corresponding to each candidate resource node are sorted in descending order;

[0144] The candidate recommended resource set associated with the p-th recommended resource is determined according to the recommended resources ranked in the top K; K is a positive integer.

[0145] In one implementation, the processing unit 902 is specifically configured to:

[0146] Call the trained recommendation model to predict the recommendation index of each recommendation resource in the target recommendation resource set for the qth object;

[0147] Sorting each recommended resource in the target recommended resource set according to the recommendation index corresponding to each recommended resource in the target recommended resource set;

[0148] The recommended resources ranked in the first L in the target recommended resource set are determined as the recommended resources recommended for the qth object; L is a positive integer.

[0149] See also Fig.10 , Fig.10 1 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device described in this embodiment includes: a processor 1001 and a memory 1002. Optionally, the electronic device may also include a network interface or a power supply module. The processor 1001 and the memory 1002 may exchange data.

[0150] The processor 1001 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0151] The network interface may include an input device and / or an output device, for example, the input device may be a control panel, a microphone, a receiver, etc., and the output device may be a display screen, a transmitter, etc., which are not listed here. For example, in an embodiment of the application, the network interface may include a receiver and a transmitter.

[0152] The memory 1002 may include a read-only memory and a random access memory, and provide program instructions and data to the processor 1001. A portion of the memory 1002 may also include a non-volatile random access memory. When the processor 1001 calls a program instruction, it is used to execute:

[0153] An initial resource network is obtained; the initial resource network is constructed according to the object access behaviors of N objects for M recommended resources, where N and M are both positive integers, and the initial resource network includes a plurality of resource nodes and edges between the plurality of resource nodes, where one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors for both recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is indicated as a target edge, and the target edge is used to connect a resource node corresponding to the first recommended resource and a resource node corresponding to the second recommended resource;

[0154] Determine the object among the N objects that has object access behavior to both the first recommended resource and the second recommended resource as the target object, and obtain the resource access heat of the first recommended resource, the resource access heat of the second recommended resource, the access time interval between the target object and the first recommended resource and the second recommended resource, and the object activity of the target object;

[0155] Determine the edge weight of the target edge according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0156] An initial resource network having the edge weight of the target edge is determined as the target resource network, and recommended resources are recommended for the N objects according to the target resource network.

[0157] In one implementation, the resource access popularity of the first recommended resource refers to the number of times the first recommended resource is accessed by N objects in a historical period, and the resource access popularity of the second recommended resource refers to the number of times the second recommended resource is accessed by N objects in a historical period; object activity refers to the number of resource accesses of the target object to M recommended resources in a historical period.

[0158] In one implementation, the processor 1001 is specifically configured to:

[0159] Calculate the access co-occurrence weight between the first recommended resource and the second recommended resource according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity;

[0160] Obtaining resource similarity between the first recommended resource and the second recommended resource, and determining a similarity weight between the first recommended resource and the second recommended resource according to the resource similarity;

[0161] The edge weight of the target edge is determined according to the access co-occurrence weight and the similarity weight.

[0162] In one implementation, the processor 1001 is specifically configured to:

[0163] Acquire first associated resource data of the first recommended resource; the first associated resource data includes at least one of associated text data and associated image data of the first recommended resource;

[0164] Acquire second associated resource data of the second recommended resource; the second associated resource data includes at least one of associated text data and associated image data of the second recommended resource;

[0165] generating a first embedding vector for the first associated resource data, and generating a second embedding vector for the second associated resource data;

[0166] The resource similarity between the first recommended resource and the second recommended resource is determined according to the first embedding vector and the second embedding vector.

[0167] In one implementation, any one of the N objects is represented as the qth object, where q is a positive integer and q is less than or equal to N; the processor 1001 is specifically configured to:

[0168] Obtain a candidate recommended resource set associated with each recommended resource according to the target resource network;

[0169] Obtain the historically visited recommended resources of the qth object, and determine the candidate recommended resource set associated with the historically visited recommended resources as the target recommended resource set;

[0170] Construct a training sample pair for the qth object based on the historical access recommendation resources and the target recommendation resource set;

[0171] The recommendation model is trained according to the training samples to obtain a trained recommendation model;

[0172] The trained recommendation model is called to determine the recommended resource for the qth object from the target recommendation resource set.

[0173] In one implementation, any one of the M recommended resources is represented as the p-th recommended resource, where p is a positive integer and is less than or equal to M; the processor 1001 is specifically configured to:

[0174] Determine a resource node in the target resource network that has an edge connecting to the resource node of the p-th recommended resource as at least one candidate resource node;

[0175] According to the edge weight of the edge between each candidate resource node and the resource node of the pth recommended resource, the recommended resources corresponding to each candidate resource node are sorted in descending order;

[0176] The candidate recommended resource set associated with the p-th recommended resource is determined according to the recommended resources ranked in the top K; K is a positive integer.

[0177] In one implementation, the processor 1001 is specifically configured to:

[0178] Call the trained recommendation model to predict the recommendation index of each recommendation resource in the target recommendation resource set for the qth object;

[0179] Sorting each recommended resource in the target recommended resource set according to the recommendation index corresponding to each recommended resource in the target recommended resource set;

[0180] The recommended resources ranked in the first L in the target recommended resource set are determined as the recommended resources recommended for the qth object; L is a positive integer.

[0181] Optionally, when the program instructions are executed by the processor, other steps of the method in the above embodiment may also be implemented, which will not be repeated here.

[0182] The present application also provides a computer-readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the above method, such as executing the method executed by the above electronic device, which is not repeated here.

[0183] Optionally, the storage medium involved in the present application, such as computer-readable storage medium, may be non-volatile or volatile.

[0184] Optionally, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of blockchain nodes, etc. Among them, the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods, each of which contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. The blockchain may include the underlying blockchain platform, the platform product service layer, and the application service layer.

[0185] It should be noted that, for the above-mentioned various method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0186] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0187] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction, and the computer instruction is stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device performs the steps performed in the embodiments of the above methods. For example, the computer device can be a terminal, or can be a server.

[0188] The above is a detailed introduction to a resource recommendation method, device, electronic device, product and medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A resource recommendation method, characterized in that: The method comprises: Acquire an initial resource network; the initial resource network is constructed according to object access behaviors of N objects for M recommended resources, N and M are both positive integers, the initial resource network includes multiple resource nodes and edges between the multiple resource nodes, one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors for both recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is represented as a target edge, and the target edge is used to connect a resource node corresponding to a first recommended resource and a resource node corresponding to a second recommended resource; Determine an object among the N objects that has object access behaviors to both the first recommended resource and the second recommended resource as a target object, and obtain the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval of the target object to the first recommended resource and the second recommended resource, and the object activity of the target object; Calculating an access co-occurrence weight between the first recommended resource and the second recommended resource according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity; Acquire resource similarity between the first recommended resource and the second recommended resource, and determine a similarity weight between the first recommended resource and the second recommended resource according to the resource similarity; Determine the edge weight of the target edge according to the access co-occurrence weight and the similarity weight; Determining the initial resource network having the edge weight of the target edge as a target resource network, and recommending recommended resources to the N objects according to the target resource network; Wherein, the access co-occurrence weight W between the first recommended resource and the second recommended resource ij Ways to obtain include: i represents the first recommended resource, j represents the second recommended resource, u represents the target object, δ(t ui -t uj ) represents the time t at which the target object accesses the first recommended resource ui The time t at which the second recommended resource is accessed uj The time interval between visits, q u represents the object activity of the target object, p i represents the resource access popularity of the first recommended resource, p j Indicates the resource access popularity of the second recommended resource.

2. The method according to claim 1, characterized in that: The resource access popularity of the first recommended resource refers to the number of times the first recommended resource is accessed by the N objects in the historical period, and the resource access popularity of the second recommended resource refers to the number of times the second recommended resource is accessed by the N objects in the historical period; the object activity refers to the number of times the target object accesses the M recommended resources in the historical period.

3. The method according to claim 1, characterized in that: The obtaining the resource similarity between the first recommended resource and the second recommended resource includes: Acquire first associated resource data of the first recommended resource; the first associated resource data includes at least one of associated text data and associated image data of the first recommended resource; Acquire second associated resource data of the second recommended resource; the second associated resource data includes at least one of associated text data and associated image data of the second recommended resource; generating a first embedding vector for the first associated resource data, and generating a second embedding vector for the second associated resource data; The resource similarity between the first recommended resource and the second recommended resource is determined according to the first embedding vector and the second embedding vector.

4. The method according to claim 1, characterized in that: Any one of the N objects is represented as the qth object, where q is a positive integer and is less than or equal to N; and recommending recommended resources for the N objects according to the target resource network includes: Acquire a candidate recommended resource set associated with each recommended resource according to the target resource network; Acquire the historically visited recommended resources of the qth object, and determine the candidate recommended resource set associated with the historically visited recommended resources as the target recommended resource set; Constructing a training sample pair for the qth object according to the historically accessed recommended resources and the target recommended resource set; Training the recommendation model according to the training samples to obtain a trained recommendation model; The trained recommendation model is called to determine a recommended resource for recommending the qth object from the target recommended resource set.

5. The method according to claim 4, characterized in that: Any one of the M recommended resources is represented as the p-th recommended resource, where p is a positive integer and is less than or equal to M; The step of acquiring a candidate recommended resource set associated with each recommended resource according to the target resource network includes: Determine a resource node in the target resource network that has an edge connected to the resource node of the p-th recommended resource as at least one candidate resource node; According to the edge weight of the edge between each candidate resource node and the resource node of the p-th recommended resource, the recommended resources corresponding to each candidate resource node are sorted in descending order; The candidate recommended resource set associated with the p-th recommended resource is determined according to the recommended resources ranked in the top K; K is a positive integer.

6. The method according to claim 4, characterized in that: The calling of the trained recommendation model to determine the recommended resource for recommending the qth object from the target recommended resource set includes: Calling the trained recommendation model to predict the recommendation index of each recommendation resource in the target recommendation resource set for the qth object; Sorting each recommended resource in the target recommended resource set according to the recommendation index corresponding to each recommended resource in the target recommended resource set; The recommended resources ranked in the first L in the target recommended resource set are determined as the recommended resources recommended for the qth object; L is a positive integer.

7. A resource recommendation device, characterized in that: The device comprises: An acquisition unit is used to acquire an initial resource network; the initial resource network is constructed according to object access behaviors of N objects for M recommended resources, N and M are both positive integers, the initial resource network includes multiple resource nodes and edges between the multiple resource nodes, one resource node corresponds to one recommended resource, and one edge is used to indicate that there is at least one object that has object access behaviors for both recommended resources corresponding to the two resource nodes connected by the edge; any edge of the initial resource network is represented as a target edge, and the target edge is used to connect a resource node corresponding to a first recommended resource and a resource node corresponding to a second recommended resource; a processing unit, configured to determine an object among the N objects that has object access behaviors to both the first recommended resource and the second recommended resource as a target object, and obtain the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval of the target object to the first recommended resource and the second recommended resource, and the object activity of the target object; The processing unit is further configured to calculate an access co-occurrence weight between the first recommended resource and the second recommended resource according to the resource access popularity of the first recommended resource, the resource access popularity of the second recommended resource, the access time interval, and the object activity; Acquire resource similarity between the first recommended resource and the second recommended resource, and determine a similarity weight between the first recommended resource and the second recommended resource according to the resource similarity; Determine the edge weight of the target edge according to the access co-occurrence weight and the similarity weight; The processing unit is further configured to determine the initial resource network having the edge weight of the target edge as a target resource network, and recommend recommended resources to the N objects according to the target resource network; Wherein, the access co-occurrence weight W between the first recommended resource and the second recommended resource ij Ways to obtain include: i represents the first recommended resource, j represents the second recommended resource, u represents the target object, δ(t ui -t uj ) represents the time t at which the target object accesses the first recommended resource uu The time t at which the second recommended resource is accessed uj The time interval between visits, q u represents the object activity of the target object, p i represents the resource access popularity of the first recommended resource, p j Indicates the resource access popularity of the second recommended resource.

8. A computer program product, characterized in that The method comprises a computer program / instruction, which implements the steps of the method according to any one of claims 1 to 6 when executed by a processor.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 6.

Images