Method and device for generating a recommendation list of preferred items, and non-volatile storage medium

Abstract

The present application discloses a method and device for generating a recommendation list of preferred items, and a non-volatile storage medium. The method includes: creating a popularity node, and embedding the popularity node into a first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent the grouping of items according to the popularity of the items; determining the preference relationship corresponding to the target object according to the target object's preference for the item, and embedding the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item; creating a heterogeneous graph conversion model according to the heterogeneous graph; converting the heterogeneous graph according to the heterogeneous graph conversion model to generate a conversion result, and determining the target object's recommendation list of preferred items based on the conversion result. The present application solves the technical problem of inaccurate recommendation lists of preferred items caused by popularity deviation.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a method and device for generating a recommendation list of preferred items, and a non-volatile storage medium. Background Art

[0002] In the Internet age, users can enjoy various services provided by electronic platforms. However, with the increasing number of users of electronic platforms, the problem of information overload has become more and more serious. Information overload makes it impossible for users to effectively search for the content they want. Recommender System (RS) is a powerful tool to solve information overload. It has been closely related to our lives and has been applied in e-commerce, audio and video, news and information and other fields. Recommender system helps users filter information according to their information needs, interests and hobbies, and recommends information and products of interest to users. With the rapid development of technology, personalized recommendation system has been successfully applied to various fields and is still being improved.

[0003] Collaborative filtering, as a successful recommendation system method, can predict the user's rating of an item and generate recommendations based on the user's preferences. However, collaborative filtering is affected by popularity bias when generating recommendations, and tends to recommend more popular items to users. Popularity bias can lead to the long tail phenomenon, making the cumulative rating of the top 20% of items much higher than the remaining 80% of long tail items; and after training on data with long tail effect, the model will inherit this bias and expand the popularity bias by recommending popular items, causing popular items to become more popular and niche items to be not recommended. Existing technical solutions to solve popularity bias include the following: improving the performance of discovering niche items by reweighting; correcting popularity bias by introducing regularization terms; or reducing the recommendation ratio of popular items by reducing the ratings of popular items; although the above technical solutions can reduce the impact of popularity bias, the generated recommendations do not meet the user's preferences.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a method and device for generating a preference item recommendation list, and a non-volatile storage medium, so as to at least solve the technical problem of inaccurate preference item recommendation list caused by popularity deviation.

[0006] According to one aspect of an embodiment of the present application, a method for generating a recommendation list of preferred items is provided, comprising: creating a popularity node, and embedding the popularity node into a first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent the grouping of items according to their popularity; determining a preference relationship corresponding to a target object according to the target object's preference for the item, and embedding the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents a relationship between the target object and the preference relationship and the popularity of the item; creating a heterogeneous graph conversion model according to the heterogeneous graph; converting the heterogeneous graph according to the heterogeneous graph conversion model to generate a conversion result, and determining a recommendation list of preferred items for the target object based on the conversion result.

[0007] Optionally, popularity nodes are created and embedded into the first knowledge graph to generate a second knowledge graph, including: dividing items into a first number of groups according to the popularity of the items; embedding each group of items as a popularity node into the first knowledge graph to generate a second knowledge graph, wherein the number of popularity nodes is the first number; the first knowledge graph includes at least one of the following: a target object, an item, and a first relationship between the target object and the item; the second knowledge graph includes at least one of the following: a target object, an item, a first relationship between the target object and the item, and the popularity of the item.

[0008] Optionally, the preference relations are embedded in the second knowledge graph to generate a heterogeneous graph, including: setting the number of preference relations to a first number; replacing the first relations in the second knowledge graph with the preference relations; and establishing a correspondence between the first number of popularity nodes in the second knowledge graph and the first number of preference relations.

[0009] Optionally, embedding the preference relationship into the second knowledge graph to generate a heterogeneous graph also includes: using popularity nodes as nodes of the heterogeneous graph; connecting the popularity nodes as edges of the heterogeneous graph; and generating a heterogeneous graph based on the nodes of the heterogeneous graph and the edges of the heterogeneous graph.

[0010] Optionally, the heterogeneous graph is converted according to a heterogeneous graph conversion model to generate a conversion result, including: determining a first projection of a node of the heterogeneous graph; determining first information of an edge of the heterogeneous graph through which the node of the heterogeneous graph passes and the mutual attention of the heterogeneous graph according to the first projection; and generating a conversion result according to the first information and the mutual attention of the heterogeneous graph.

[0011] Optionally, determining a target object's preferred item recommendation list based on the conversion result includes: determining a proportion of popular items in the target object's interaction data; calculating a first weight based on the proportion of popular items in the interaction data, wherein the first weight represents the degree of influence of the popularity of the item on the target object's preferred item recommendation list; calculating a relationship between the target object, the item, and popularity based on the first weight to obtain a calculation result; and determining the target object's preferred item recommendation list based on the calculation result.

[0012] Optionally, before converting the heterogeneous graph according to the heterogeneous graph conversion model and determining the recommended list of preferred items of the target object according to the conversion result, the method further includes: training the heterogeneous graph conversion model by a preset algorithm.

[0013] According to another aspect of an embodiment of the present application, a device for generating a recommendation list of preferred items is also provided, including: a first creation module, used to create popularity nodes, and embed the popularity nodes into a first knowledge graph to generate a second knowledge graph, wherein the popularity nodes are used to represent the grouping of items according to their popularity; an embedding module, used to determine the preference relationship corresponding to the target object according to the target object's preference for the item, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item; a second creation module, used to create a heterogeneous graph conversion model based on the heterogeneous graph; a determination module, used to convert the heterogeneous graph according to the heterogeneous graph conversion model, generate a conversion result, and determine the target object's recommendation list of preferred items based on the conversion result.

[0014] According to another aspect of an embodiment of the present application, a non-volatile storage medium is provided. The non-volatile storage medium includes a stored program. When the program is running, the device where the non-volatile storage medium is located is controlled to execute the above method for generating a recommendation list of preferred items.

[0015] According to another aspect of the embodiment of the present application, a processor is provided, the processor being used to run a program stored in a memory, and the above method for generating a recommendation list of preferred items is executed when the program is run.

[0016] In an embodiment of the present application, a popularity node is created and embedded into a first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent the grouping of items according to their popularity; a preference relationship corresponding to the target object is determined according to the target object's preference for the item, and the preference relationship is embedded into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item; a heterogeneous graph conversion model is created according to the heterogeneous graph; the heterogeneous graph is converted according to the heterogeneous graph conversion model to generate a conversion result, and a recommendation list of preferred items for the target object is determined based on the conversion result. By creating a knowledge graph containing item popularity information and target object preference information; and using the knowledge graph to make recommendations based on the user's preference for popular items, the purpose of reducing the impact of popularity deviation in the recommendation system is achieved, thereby achieving the technical effect of generating a recommendation list of preferred items according to the target object's preference and increasing the exposure rate of non-popular items, thereby solving the technical problem of inaccurate recommendation lists of preferred items caused by popularity deviation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 is a flow chart of a method for generating a recommendation list of preferred items according to the related art;

[0019] Figure 2 is a schematic diagram of creating a KGEPN according to an embodiment of the present application;

[0020] Figure 3 is a schematic structural diagram of a HGT according to an embodiment of the present application;

[0021] Figure 4 It is a schematic diagram of generating different recommendation lists of preferred items according to different preferences of users for popular items according to an embodiment of the present application;

[0022] Figure 5 is a structural diagram of a device for generating a recommendation list of preferred items according to an embodiment of the present application;

[0023] Figure 6 is a list of attributes of the datasets LastFM and DBbook-2014 provided according to the embodiments of the present application;

[0024] Figure 7 This is a list of experimental results of several algorithms provided according to the embodiments of this application. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] According to an embodiment of the present application, a method embodiment for generating a recommendation list of preferred items is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] In order to solve the technical problems mentioned in the background technology, the embodiment of the present application provides a framework for eliminating popularity bias (Adaptive Alleviation for Popularity Bias with Knowledge Graph, AWING) based on Knowledge Graph (KG). By using heterogeneous graph transformation on the knowledge graph embedded with popular nodes to explore the fine-grained preferences behind user-item interactions, such as popular preference items, explicit semantics are given to preferences, and the popularity bias problem is adaptively alleviated from the user's perspective. When dealing with the popularity bias problem in the recommendation system based on the method provided in the embodiment of the present application, the knowledge graph is introduced for the first time, which is not only conducive to recommendation, but also can effectively alleviate the popularity problem while utilizing the user's preference for popularity, taking into account the user's usage experience; compared with other methods, this method has certain advantages in reasonably alleviating the popularity bias problem.

[0029] Figure 1 is a flowchart of a method for generating a recommendation list of preferred items according to an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0030] Step S102, creating a popularity node, and embedding the popularity node into the first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent the grouping of items according to their popularity;

[0031] By executing step S102, a knowledge graph embedded with popular nodes is established. The role of the popular nodes is to enhance the entity relationship in the knowledge graph.

[0032] Step S104, determining the preference relationship corresponding to the target object according to the target object's preference for the item, and embedding the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item;

[0033] By executing step S104, the user's preference relationship is integrated with the second knowledge graph to generate a heterogeneous graph for representing the relationship between the user's preference, the item, and the popularity of the item.

[0034] Step S106, creating a heterogeneous graph conversion model according to the heterogeneous graph;

[0035] According to step S106, a heterogeneous graph conversion model is created based on the heterogeneous graph, and the heterogeneous graph conversion model learns representations of users, items, and fine-grained preferences based on the heterogeneous graph converter model.

[0036] Step S108, converting the heterogeneous graph according to the heterogeneous graph conversion model to generate a conversion result, and determining a recommended list of preferred items of the target object according to the conversion result;

[0037] Step S108 is executed to convert the heterogeneous graph using the heterogeneous graph conversion model. The conversion result is a personalized recommendation list for each user after removing the popularity deviation, which is generated based on the user's preference for popular items.

[0038] Through the above steps, it is possible to reduce the impact of popularity deviation when generating a recommendation list of preferred items based on the user's preference for popular items, improve the accuracy of the preferred items, and at the same time improve the user's experience when using a network platform based on a recommendation system.

[0039] According to another optional embodiment of the present application, a popularity node is created and embedded into a first knowledge graph to generate a second knowledge graph, wherein the method for creating the popularity node is as follows: items are divided into a first number of groups according to their popularity; each group of items is a popularity node, and therefore, there are a first number of popularity nodes in the second knowledge graph; the first knowledge graph contains the following information: a target object, an item, and a first relationship between the target object and the item; the second knowledge graph, in addition to containing the information of the first knowledge graph, also contains information on the popularity of the item.

[0040] Figure 2 : is a schematic diagram of creating a KGEPN according to this embodiment. First, the items are divided into K groups according to their popularity, and then K popular nodes called pi, i∈{1,…,K} are created to represent these groups; and the items are connected to their corresponding nodes. Before the popularity node is created, the relationship in the knowledge graph is (entity s, relationship between s and t, entity t). After the popularity node is created, a new relationship (i, popularity, pn) is introduced to the knowledge graph, thereby successfully integrating the popularity information into the knowledge graph; this knowledge graph integrating popularity information is represented as (Knowlege Graph Embedded with Popularity Nodes, KGEPN).

[0041] According to another optional embodiment of the present application, the preference relationship is embedded in the second knowledge graph to generate a heterogeneous graph, which also includes: using popularity nodes as nodes of the heterogeneous graph; connecting popularity nodes as edges of the heterogeneous graph; and generating a heterogeneous graph based on the nodes of the heterogeneous graph and the edges of the heterogeneous graph.

[0042] The preference relations are embedded in the second knowledge graph to generate a heterogeneous graph, which is achieved by the following method: setting the number of preference relations to the first number; replacing the first relations in the second knowledge graph with the preference relations; establishing a correspondence between the first number of popularity nodes in the second knowledge graph and the first number of preference relations. The preference relations are embedded in the second knowledge graph by setting the number of preference relations to the first number; replacing the first relations in the second knowledge graph with the preference relations; establishing a correspondence between the first number of popularity nodes in the second knowledge graph and the first number of preference relations.

[0043] According to another optional embodiment of the present application, before embedding the preference relationship into the second knowledge graph, it is necessary to model the user-item relationship at the granularity of preference; the specific modeling method includes: capturing multiple preference intuitive knowledge that affects user behavior; since preference intuitive knowledge can reflect the commonality of all user behaviors, this preference intuitive knowledge can be used as the reason for the user to choose an item, and based on this preference intuitive knowledge, the user-item relationship is modeled at the granularity of preference. Among them, the granularity of preference reflects the amount of captured preference intuitive knowledge, and the more the amount of captured preference intuitive knowledge, the larger the granularity of preference. Taking p as the preference set shared by all users, we can divide the unified user-item relationship into |p| preferences, and decompose each (u, interact_with, i) triple into the form of {(u, p, i)|p∈P} to create a preference graph. Since the preference is expressed as a latent vector, it is vague for a deeper understanding. We set the number of preferences to the number of relationships in KGEPN and convert the relationship information in KGEPN into preferences. Specifically, we utilize the Euclidean norm to align preference embeddings and relation embeddings in KGEPN.

[0044] According to another optional embodiment of the present application, the preference relationship is embedded in the second knowledge graph to generate a heterogeneous graph, wherein the method for generating the heterogeneous graph is: using popularity nodes as nodes of the heterogeneous graph; connecting popularity nodes as edges of the heterogeneous graph; and generating the heterogeneous graph based on the nodes of the heterogeneous graph and the edges of the heterogeneous graph.

[0045] In this optional embodiment, KGEPN and PG are combined into a new heterogeneous graph (Heterogeneous Graph, HG), and the specific method is: the nodes in KGEPN are used as nodes of HG, and any two nodes are connected as edges of HG.

[0046] According to some optional embodiments of the present application, a heterogeneous graph is converted according to a heterogeneous graph conversion model, which is achieved by the following method: determining a first projection of a node of the heterogeneous graph; determining first information of an edge of the heterogeneous graph through which the node of the heterogeneous graph passes and the mutual attention of the heterogeneous graph according to the first projection; and generating a conversion result according to the first information and the mutual attention of the heterogeneous graph.

[0047] Based on HG (i.e. heterogeneous graph), a heterogeneous graph transformer (Heterogeneous Graph Transformer, HGT) was developed. Figure 3 This is a schematic diagram of the HGT structure. Heterogeneous graph transformation is completed by message calculation and message aggregation. The message calculation part combines the message matrix To alleviate the distribution differences of different types of nodes and edges; given a source node s and an edge e, HGT can calculate the information of node s passing through edge e through the following equation:

[0048]

[0049] Among them, M τ(s) is a unique linear projection of node type τ(s).

[0050] In the message aggregation part, HGT is first used to calculate the heterogeneous mutual attention between the source node s and the target node t to control the influence of s on t. HGT uses a specific linear projection Q for each type of node. τ(t) or K τ(s) , and use a different edge-based matrix for each edge type (e) To maximize the difference modeling of the distribution. Let N(t) represent the one-hop neighbor node of node t, and then calculate the mutual attention Attention(s, e, t) of the heterogeneous graph:

[0051]

[0052] HGT then uses the attention vector as a weight to average the corresponding messages from the source node and updates the output transformation result:

[0053]

[0054] According to an optional embodiment of the present application, determining a target object's preferred item recommendation list based on a conversion result includes the following steps: determining a proportion of popular items in the target object's interaction data; calculating a first weight based on the proportion of popular items in the interaction data, wherein the first weight represents the degree of influence of the popularity of the item on the target object's preferred item recommendation list; calculating a relationship between the target object, the item, and popularity based on the first weight to obtain a calculation result, and determining the target object's preferred item recommendation list based on the calculation result.

[0055] With independent edge matrices corresponding to each edge type, users’ preferences can be quantified more finely. Figure 4 It is a schematic diagram of generating different recommendation lists of preferred items according to different preferences of users for popular items, such as Figure 4 , is the final representation of a given user u and item i. For the preference of user u, the corresponding scores between u and i are calculated to determine the possibility of u accepting i. In this calculation, a weight w is designed to control the impact of popularity on the recommendation results. According to the proportion of popular items in the profile of user u, a certain proportion of popular item recommendations are removed for u. Specifically, through the equation:

[0056]

[0057] To suppress the recommendation of popular items,

[0058] w=1-RP(u)

[0059] Among them, RP(u) is the proportion of popular items in the interaction data of user u. If the user is very interested in popular items, the user's popular item preferences in the preference recommendation list will rarely be deleted, thereby ensuring the user experience. On the contrary, if the user is not interested in popular items, more popular items in the preference recommendation list will be deleted. Different preference item recommendation lists are generated according to the different preferences of users for popular items.

[0060] According to an optional embodiment of the present application, before converting the heterogeneous graph according to the heterogeneous graph conversion model and determining the recommended list of preferred items of the target object based on the conversion result, the method also includes: training the heterogeneous graph conversion model through a preset algorithm.

[0061] In this embodiment, HGT (i.e., heterogeneous graph transformer) learns representations of users, items, and fine-grained preferences based on the heterogeneous graph transformer model; it aims to aggregate information from the neighbors of the target node t. The depth of HGT is represented by L. Pairwise BPR loss is selected to train HGT. By combining the alignment loss and the BPR loss, the pairwise BPR loss is minimized. Learn the model parameters, where Θ represents the model parameter set, and then use α and λ to constrain the alignment loss and L2 norm respectively.

[0062] Figure 5 is a structural diagram of a device for generating a recommendation list of preferred items according to an embodiment of the present application, such as Figure 5 , the device comprises:

[0063] The first creation module 50 is used to create popularity nodes and embed the popularity nodes into the first knowledge graph to generate a second knowledge graph, wherein the popularity nodes are used to represent grouping of items according to their popularity.

[0064] The embedding module 52 is used to determine the preference relationship corresponding to the target object according to the target object's preference degree for the item, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item.

[0065] The second creation module 54 is used to create a heterogeneous graph conversion model according to the heterogeneous graph.

[0066] The determination module 56 is used to transform the heterogeneous graph according to the heterogeneous graph transformation model, generate a transformation result, and determine a recommendation list of preferred items of the target object according to the transformation result.

[0067] It should be noted that Figure 5 The preferred implementation of the illustrated embodiment can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0068] According to the embodiment of the present application, a specific implementation method is provided. During the specific implementation, the public data sets LastFM and DBbook-2014 are used. Figure 6 The property list of the datasets LastFM and DBbook-2014 is provided; in order to ensure the quality of the dataset, a 5-core setting is used, i.e., at least 5 interacting users and items are retained to construct item knowledge for each dataset; 80% of the items related to each user are randomly selected to form a training set, and all the remaining 20% ​​are used as a test set to test an optional method provided in this application, and compare it with the recommendation algorithms: Bayesian Personalized Ranking Matrix Factorization (BPRMF), Light Graph Convolution Network (LightGCN), Knowledge-enhanced Translation-based User Preference (KTUP), IPS-CN, and Embedded Secure Access Module (ESAM); Figure 7 Experimental results of several algorithms. Figure 7AWING-APS in is a variant of AWING that removes a certain percentage of popular items when recommending. From the experimental results, it can be seen that the BPRMF model performs the worst on the two data sets, and the performance of LightGCN is better than that of BPRMF, but its average proportion of long-tail items, i.e., non-popular items (the Average Percentage of Tail items, APT) is also very low, indicating that there is a widespread popularity bias in the recommendation system of LightGCN; the performance of KTUP is better than that of LightGCN and BPRMF models in all indicators, which shows that the introduction of knowledge graphs is not only conducive to recommendation, but also helps to alleviate popularity bias; compared with the above three models, IPS-CN and ESAM have greatly improved in APT, but have not maintained the ratio of popular and non-popular items; the method AWING according to the embodiment of the present application is superior to all the comparison methods on the two data sets in terms of recall rate (Recall) and normalized discounted cumulative gain (Normalize Discounted Cumulative Gain, NDCG), which shows that identifying fine-grained preferences is helpful for recommendation; AWING-APS is superior to all the comparison methods on the two data sets in terms of user popularity preference consistency (the Change in the Group Average Popularity, ΔGAP), overall diversity (Aggregate Diversity, AD) and average percentage of tail items (Average Percentage of This shows that removing a certain proportion of popular item recommendations is of great significance in alleviating popularity bias. At the same time, this method maintains a similar ratio of popular and non-popular items, ensuring a good user experience.

[0069] According to another aspect of an embodiment of the present application, a non-volatile storage medium is provided. The non-volatile storage medium includes a stored program. When the program is running, the device where the non-volatile storage medium is located is controlled to execute the above method for generating a recommendation list of preferred items.

[0070] The above-mentioned non-volatile storage medium is used to store a program that performs the following functions: create a popularity node, and embed the popularity node into a first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent the grouping of items according to the popularity of the items; determine the preference relationship corresponding to the target object according to the target object's preference for the item, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item; create a heterogeneous graph conversion model based on the heterogeneous graph; convert the heterogeneous graph according to the heterogeneous graph conversion model to generate a conversion result, and determine a recommended list of preferred items for the target object based on the conversion result.

[0071] According to another aspect of the embodiment of the present application, a processor is provided, the processor being used to run a program stored in a memory, and the above method for generating a recommendation list of preferred items is executed when the program is run.

[0072] The above-mentioned processor is used to run a program that performs the following functions: create popularity nodes, and embed the popularity nodes into a first knowledge graph to generate a second knowledge graph, wherein the popularity nodes are used to represent the grouping of items according to the popularity of the items; determine the preference relationship corresponding to the target object according to the target object's preference for the items, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the items; create a heterogeneous graph conversion model based on the heterogeneous graph; convert the heterogeneous graph according to the heterogeneous graph conversion model to generate a conversion result, and determine a recommended list of preferred items for the target object based on the conversion result.

[0073] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0074] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0076] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0077] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0078] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0079] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for generating a list of recommended preferred items, characterized in that: include: Creating a popularity node and embedding the popularity node into the first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent grouping of the items according to their popularity; Determine the preference relationship corresponding to the target object according to the preference degree of the target object to the item, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item, embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, including: using the popularity nodes as nodes of the heterogeneous graph; connecting the popularity nodes as edges of the heterogeneous graph; generating the heterogeneous graph according to the nodes of the heterogeneous graph and the edges of the heterogeneous graph; Creating a heterogeneous graph conversion model according to the heterogeneous graph; The heterogeneous graph is converted according to the heterogeneous graph conversion model to generate a conversion result, and a recommendation list of preferred items of the target object is determined according to the conversion result.

2. The method according to claim 1, characterized in that Creating a popularity node and embedding the popularity node into the first knowledge graph to generate a second knowledge graph includes: dividing the items into a first number of groups according to popularity of the items; Embed each group of items as the popularity node into the first knowledge graph to generate the second knowledge graph, wherein the number of the popularity nodes is the first number; The first knowledge graph includes at least one of the following: the target object, the item, and a first relationship between the target object and the item; The second knowledge graph includes at least one of the following: the target object, the item, a first relationship between the target object and the item, and popularity of the item.

3. The method according to claim 2, characterized in that Embedding the preference relationship into the second knowledge graph to generate a heterogeneous graph includes: setting the number of the preference relations to the first number; Replacing the first relationship in the second knowledge graph with the preference relationship; A correspondence between the first number of popularity nodes and the first number of preference relationships in the second knowledge graph is established.

4. The method according to claim 1, characterized in that: The heterogeneous graph is converted according to the heterogeneous graph conversion model to generate a conversion result, including: Determining a first projection of a node of the heterogeneous graph; Determine, according to the first projection, first information that a node of the heterogeneous graph passes through an edge of the heterogeneous graph and a mutual attention of the heterogeneous graph; The conversion result is generated according to the mutual attention of the first information and the heterogeneous graph.

5. The method according to claim 1, characterized in that Determining a recommended list of preferred items for the target object based on the conversion result includes: Determining a proportion of popular items in the interaction data of the target object; Calculating a first weight according to the proportion of popular items in the interaction data, wherein the first weight represents the influence of the popularity of the item on the recommended list of preferred items of the target object; Calculate the relationship between the target object, the item, and the popularity according to the first weight to obtain a calculation result; A recommended list of preferred items for the target object is determined according to the calculation result.

6. The method according to claim 1, characterized in that Before converting the heterogeneous graph according to the heterogeneous graph conversion model and determining a recommended list of preferred items of the target object according to the conversion result, the method further includes: training the heterogeneous graph conversion model by a preset algorithm.

7. A device for generating a recommendation list of preferred items, characterized in that: include: A first creation module, configured to create a popularity node and embed the popularity node into the first knowledge graph to generate a second knowledge graph, wherein the popularity node is used to represent grouping of the items according to their popularity; An embedding module, used to determine the preference relationship corresponding to the target object according to the preference degree of the target object to the item, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, wherein the heterogeneous graph represents the relationship between the target object and the preference relationship and the popularity of the item, and embed the preference relationship into the second knowledge graph to generate a heterogeneous graph, including: using the popularity node as a node of the heterogeneous graph; connecting the popularity nodes as an edge of the heterogeneous graph; generating the heterogeneous graph according to the nodes of the heterogeneous graph and the edges of the heterogeneous graph; A second creation module, used for creating a heterogeneous graph conversion model according to the heterogeneous graph; A determination module is used to convert the heterogeneous graph according to the heterogeneous graph conversion model, generate a conversion result, and determine a recommended list of preferred items of the target object according to the conversion result.

8. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute the method for generating a recommendation list of preferred items as claimed in any one of claims 1 to 6.

9. A processor, characterized in that: The processor is used to run a program stored in the memory, wherein the program, when running, executes the method for generating a preferred item recommendation list according to any one of claims 1 to 6.

Images