Item recommendation method, device, storage medium and electronic device

Abstract

The present invention discloses an item recommendation method, device, storage medium and electronic device. It relates to the field of artificial intelligence, and in particular to the field of recommendation systems. The method includes: constructing an initial implicit collaborative knowledge graph; based on the initial implicit collaborative knowledge graph, constructing multiple embedding representations, the multiple embedding representations include account embedding representations, item embedding representations and entity embedding representations; based on the multiple embedding representations, constructing an adaptive implicit collaborative knowledge graph, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in items, as well as the implicit preferences and associations between items and entities; based on the adaptive implicit collaborative knowledge graph, using an item recommendation model, obtaining item recommendation results corresponding to a target account, wherein the target account is an account in the adaptive implicit collaborative knowledge graph. The present invention solves the technical problem of poor item recommendation results caused by the large sparsity of interaction data and knowledge graphs.

Description

Technical Field

[0001] The present invention relates to the fields of graph neural networks and recommendation systems, and more specifically, to an item recommendation method, device, storage medium, and electronic device. Background Art

[0002] Recommendation systems provide personalized recommendations based on user information, item features, and historical interaction data. Recommendation systems train deep learning models to build relationships between users, items, and entities, enabling cold start, cross-domain, and multimodal recommendations, improving recommendation effectiveness. In recent years, graph neural networks (graph learning) have developed rapidly. These models can learn high-level relationships between entities, integrate entity features, and construct association networks, enhancing the effectiveness of recommendation systems.

[0003] The effectiveness of recommendation systems is affected by the sparsity of user-item interaction data, user variability, product diversity, uneven distribution, and cold start issues. How to improve the personalization and accuracy of recommendation results is a core issue in recommendation system research. Interaction data and knowledge graphs are often very sparse, that is, there are very few edges in the graph, which makes it difficult for graph neural networks to fully function. Specifically: 1) There are a large number of unobserved behaviors between users and items. However, related technologies regard these unobserved behaviors as negative feedback information, ignoring that some of these behaviors may also reflect users' positive preferences, that is, unobserved positive feedback information. The above practices may reduce the quality of user preference modeling; 2) The incomplete user preference modeling in related technologies and the insufficient mining of entity relationships in knowledge graphs lead to low accuracy of item recommendation results.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] Embodiments of the present invention provide an item recommendation method, device, storage medium, and electronic device to at least solve the technical problem of poor item recommendation effect caused by the large sparsity of interaction data and knowledge graphs.

[0006] According to one aspect of an embodiment of the present invention, a method for item recommendation is provided, comprising: constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent explicit interactive relationships and implicit preferences and associations between accounts, items and entities, and the entity is attribute information of the item, or association information associated with the item; based on the initial implicit collaborative knowledge graph, constructing multiple embedding representations, wherein the multiple embedding representations include account embedding representations, item embedding representations and entity embedding representations; based on the multiple embedding representations, constructing an adaptive implicit collaborative knowledge graph, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preferences and associations between the item and the entity; based on the adaptive implicit collaborative knowledge graph, using an item recommendation model to obtain item recommendation results corresponding to a target account, wherein the target account is an account in the adaptive implicit collaborative knowledge graph.

[0007] According to another aspect of an embodiment of the present invention, an item recommendation device is also provided, including: a construction module for constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent the explicit interaction relationship and implicit preferences and associations between accounts, items and entities, and the entity is the attribute information of the item, or the association information associated with the item; a representation module for constructing multiple embedding representations based on the initial implicit collaborative knowledge graph, wherein the multiple embedding representations include account embedding representation, item embedding representation and entity embedding representation; an association module for constructing an adaptive implicit collaborative knowledge graph based on the multiple embedding representations, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preferences and associations between the item and the entity; a target module for obtaining the item recommendation result corresponding to the target account by adopting the item recommendation model based on the adaptive implicit collaborative knowledge graph, wherein the target account is an account in the adaptive implicit collaborative knowledge graph.

[0008] According to another aspect of an embodiment of the present invention, a non-volatile storage medium is provided, wherein the non-volatile storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executed by any one of the item recommendation methods.

[0009] According to another aspect of an embodiment of the present invention, an electronic device is also provided, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the item recommendation methods.

[0010] In an embodiment of the present invention, an initial implicit collaborative knowledge graph is constructed, wherein the initial implicit collaborative knowledge graph is used to represent the explicit interactive relationship and implicit preferences and associations between accounts, items and entities, and the entity is the attribute information of the item, or the association information associated with the item; based on the initial implicit collaborative knowledge graph, multiple embedding representations are constructed, wherein the multiple embedding representations include account embedding representations, item embedding representations and entity embedding representations; based on the multiple embedding representations, an adaptive implicit collaborative knowledge graph is constructed, wherein the adaptive implicit collaborative knowledge graph is used to represent the potential interest of the account in the item, as well as the implicit preferences and associations between the item and the entity; based on The adaptive implicit collaborative knowledge graph adopts an item recommendation model to obtain item recommendation results corresponding to the target account, wherein the target account is an account in the adaptive implicit collaborative knowledge graph. It achieves the purpose of more accurately predicting the user's potential interest in items by constructing and optimizing the implicit collaborative knowledge graph, combining explicit and implicit user preferences, and thus providing each target user with personalized and high-quality item recommendation results. It achieves the technical effect of more comprehensively capturing and utilizing the complex relationships between users, items and entities, and improving the accuracy and diversity of recommendations, thereby solving the technical problem of poor item recommendation effect caused by the large sparsity of interaction data and knowledge graphs. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0012] Figure 1 is a flowchart of an item recommendation method according to an embodiment of the present invention;

[0013] Figure 2 is an optional model framework diagram according to an embodiment of the present invention;

[0014] Figure 3 is a schematic diagram of an optional graph neural network module according to an embodiment of the present invention;

[0015] Figure 4 is a flowchart of an optional item recommendation method according to an embodiment of the present invention;

[0016] Figure 5 2 is a schematic diagram of an item recommendation device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention 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 numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes 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.

[0019] First, to facilitate understanding of the embodiments of the present invention, some of the terms or nouns involved in the present invention are explained below:

[0020] Graph Neural Networks (GNNs) are deep learning models designed to process graph-structured data. They use representations learned from graph nodes and edges to perform tasks such as node classification, graph classification, and link prediction. GNNs are capable of capturing complex relationships between nodes and processing non-Euclidean data, making them widely used in fields such as social network analysis, recommender systems, and bioinformatics.

[0021] Recommender systems, a key branch of computer science, artificial intelligence, and data mining, aim to provide personalized product or service recommendations by analyzing users' historical behavior, preferences, and other relevant information. Recommender systems are widely used in a variety of fields, including e-commerce, social media, online video, music, and news.

[0022] The Implicit Collaborative Knowledge Graph (ICKG) is a knowledge representation and reasoning framework that combines implicit collaborative filtering and knowledge graph technologies. It mines implicit collaborative relationships between users and associations between entities (such as products, services, and content) to construct a graph structure rich in semantics and relationships. In this graph, nodes represent entities, and edges represent relationships between entities, such as user preferences for entities, similarities between entities, and categorical relationships.

[0023] Embedding representation is a technique for mapping high-dimensional data (such as text, images, or sound) into a low-dimensional space (typically Euclidean space), preserving the data's inherent structure and relationships. In machine learning and deep learning, embedding representation is often used for feature extraction and dimensionality reduction to better perform tasks such as classification, clustering, or similarity measurement.

[0024] According to an embodiment of the present invention, an embodiment of a method for recommending 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.

[0025] Figure 1 is a flow chart of an item recommendation method according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0026] Step S102: constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent explicit interactive relationships and implicit preferences and associations between accounts, items, and entities, where an entity is attribute information of an item or association information associated with an item;

[0027] Optionally, the implicit collaborative knowledge graph does not rely on explicit user feedback (such as ratings and comments), but instead infers user interests and preferences through user behavioral data (such as purchase history and browsing history). Entities and their relationships can be organized and stored by utilizing the graph structure, enabling complex queries and reasoning. By analyzing user behavior and preferences, the implicit collaborative knowledge graph can be used to provide personalized recommendation services. The introduction of the knowledge graph enables the system to understand and reason about complex relationships between entities, improving the accuracy and relevance of recommendations. In addition, the implicit collaborative knowledge graph can add new entities and relationships to adapt to changing data and needs.

[0028] Optionally, the initial implicit collaborative knowledge graph not only includes explicit interaction data between users and items, such as purchase and browsing history, but also incorporates item attribute information and implicit relationships between items and entities. This integration of multi-source information can more comprehensively understand user preferences and improve the level of personalization of recommendations. By constructing an implicit collaborative knowledge graph, this embodiment can explore items and entities that users have not directly expressed but may be interested in. For example, if a user often browses a certain type of product but has never purchased it, the user's implicit preferences can be inferred by analyzing the association between the user's browsing behavior and the product attributes, thereby predicting items that the user may be interested in but has not yet explored. The nodes and edges in the graph make it possible to establish high-order associations and capture complex multi-hop relationships between users, items, and entities. For example, a user may be interested in a certain brand of product. This preference is not only based on the product itself, but also on the brand to which the product belongs, other brand-related products or entities (such as designers, places of origin, etc.).

[0029] Optionally, the item can be a commodity, and the corresponding entity can be the category, brand, attribute of the commodity, or other information related to the commodity, such as reviewers, places, events, etc. Entities exist in the form of nodes in the knowledge graph, and are connected to other entities or items through edges to express the relationship between them. The relationship between entities and items is diverse. It can be direct, such as the item belongs to a certain entity category (if the entity is a category), or indirect, such as the item and the entity are implicitly associated through user behavior or recommendation context. For example, if the entity is a brand, the item can be a product under a certain brand; or if the entity is a place, the item can be an activity or service related to the place. In the knowledge graph, these relationships are represented in the form of edges, forming a complex network structure, which helps to more comprehensively understand the characteristics of the items and the possible preferences of users.

[0030] In an optional embodiment, constructing an initial implicit collaborative knowledge graph includes: constructing multiple triples, wherein the triples include items, entities, and implicit relationships between items and entities; constructing a binary graph based on the interactive operation behavior between accounts and items; using items as a bridge connecting accounts and entities, fusing multiple triples into the binary graph to obtain an initial implicit collaborative knowledge graph.

[0031] Optional, Figure 2 It is an optional model framework diagram provided according to an embodiment of the present invention, which can construct an initial implicit collaborative knowledge graph G based on training samples:

[0032] 1) Organize items, entities, and the implicit relationships between them in the business system in the form of triples, and construct an item-entity implicit graph from multiple pairs of triples;

[0033] 2) Collect the interactive operations between users and items in the business system and construct a user-item binary graph;

[0034] 3) Integrate item-entity pairs into the user-item binary graph, use item nodes as bridges to connect user nodes and entity nodes, construct an initial implicit collaborative knowledge graph G, and store it in the graph database; assume there are M user nodes, N item nodes, and T entity nodes.

[0035] Alternatively, by constructing triples containing items, entities, and the implicit relationships between them, we can capture the complex associations between items and entities. These associations can stem from item attributes, brands, categories, or implicit similarities between items. This fine-grained representation of relationships provides a foundation for subsequent graph construction, enabling the recommendation system to not only understand users' explicit preferences but also explore implicit preferences that may not yet be expressed. Based on explicit user-item interactions, such as purchases, reviews, and favorites, a user-item binary graph is constructed. This step intuitively indicates which items users are interested in and the intensity of their interest, providing direct user feedback data for subsequent graph fusion. By fusing multiple triples into the binary graph, using items as a bridge between users and entities, implicit preferences and explicit interaction data can be combined to form a more complete and refined implicit collaborative knowledge graph. This fusion not only enhances the model's understanding of user preferences and item attributes, but also enriches the multidimensional features of items through the relationships between entities, providing more comprehensive information support for personalized recommendations.

[0036] Step S104: constructing multiple embedding representations based on the initial implicit collaborative knowledge graph, wherein the multiple embedding representations include account embedding representation, item embedding representation, and entity embedding representation;

[0037] Optionally, accounts, items, and entities can be mapped from a high-dimensional feature space to a low-dimensional vector space through an embedding representation. This process can utilize deep learning techniques, such as neural networks, which can effectively compress information while preserving key features. This approach can reduce computational complexity and improve model training and prediction efficiency. By constructing account embeddings, item embeddings, and entity embeddings, users' historical behavior, item attributes, and associations between entities can be integrated. This enriches the feature representations of each node, promotes complementarity between features, and enables the model to understand user preferences and item characteristics from multiple perspectives. While preserving the initial graph structure, the embedding representation uses a propagation algorithm to transmit and aggregate information within the graph, enhancing the associations and relationships within the graph structure. During the learning process, the model can better utilize the structural information in the graph to capture implicit preferences and associations between accounts, items, and entities.

[0038] In an optional embodiment, multiple embedding representations are constructed based on the initial implicit collaborative knowledge graph, including: mapping accounts, items and entities in the initial implicit collaborative knowledge graph to a vector space to obtain multiple embedding representations, wherein the multiple embedding representations are in the form of vector matrices.

[0039] Optionally, multiple embedding representations can be in the form of vector matrices. Initialize the user, item, and entity embedding vectors: map all user, item, and entity nodes in the initial implicit collaborative knowledge graph G to the vector space, and generate multiple embedding representations to be learned by random initialization, namely the user embedding vector matrix V u , item embedding vector matrix V i and entity embedding vector matrix V e The specific process is described as follows:

[0040]

[0041] in, and V i 0 Represent the embedded vector matrices of initialized users, items and entities respectively, represents the initialization embedding vector of user u1, represents the initialization embedding vector of item i1, Represents the initialization embedding vector of entity e1, and the dimension of the vector is D.

[0042] Optionally, the node information in the initial implicit collaborative knowledge graph can be converted into a low-dimensional vector matrix. This step is to represent non-numerical features (such as account ID, item ID, entity attributes, etc.) as numerical vectors, which is convenient for processing and analysis by machine learning models. In the vector space, by calculating the vector distance or similarity between nodes (i.e., accounts, items, and entities), the strength of the relationship between nodes can be quantified, including explicit interactions and implicit preferences. This quantification method helps the model understand the account's potential interest in items, as well as the implicit preferences and associations between items and entities. Constructing an embedded representation can be achieved through deep learning technology (such as neural networks), which can automatically learn the feature representation of nodes, rather than just a simple mapping. Representation learning can capture the intrinsic properties of nodes and the complex relationships in the graph structure, providing high-quality feature representations for subsequent model training and prediction.

[0043] Optionally, by converting to an embedding representation in the form of a vector matrix, the recommendation model can more accurately understand user preferences and item characteristics, thereby providing more personalized recommendations that better align with user interests. Embeddings preserve the relational information in the graph, enabling a multi-faceted understanding of accounts, items, and entities. While raw account, item, and entity features may be high-dimensional, converting them to a vector matrix significantly reduces computational resources and storage requirements, improving the overall efficiency and responsiveness of the recommendation system. For new accounts or items, even without direct interaction data, embeddings can leverage their position in the graph and associated entity information to generate preliminary feature representations, alleviating the cold start problem in the recommendation system. The constructed embeddings in the form of vector matrices can be dynamically updated. As user behavior changes and item attributes are updated, the model learns new embeddings, facilitating the updating of the adaptive implicit collaborative knowledge graph and maintaining the timeliness and relevance of recommendation results. By learning feature representations in vector space, the model can better handle unseen accounts, items, and entities, improving its generalization to new data and reducing the risk of overfitting.

[0044] Step S106: constructing an adaptive implicit collaborative knowledge graph based on the multiple embedded representations, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preferences and associations between the item and the entity;

[0045] Optionally, by constructing an adaptive implicit collaborative knowledge graph, the graph can be updated in real time or periodically based on changes in user behavior, updates to item attributes, and the addition of new entity association information, thereby maintaining the graph's timeliness and accuracy and providing the recommendation system with the latest and most user-friendly data structure. By leveraging the similarity between account embeddings and item embeddings, the construction of an adaptive implicit collaborative knowledge graph can identify potential interests between accounts and items that are not directly observed. For example, even if a user has not directly interacted with an item, if the item's embedding is similar to the embeddings of items the user has liked in the past, the system can infer that the user may be interested in this item. Introducing entity embeddings can further enhance item descriptions. Entity information covers item attributes, categories, brands, and more. This information is connected to item nodes through implicit associations, allowing for a more comprehensive consideration of the multi-dimensional characteristics of items when making recommendations, improving the relevance and diversity of recommendations. The adaptive implicit collaborative knowledge graph not only considers first-order relationships (direct account-item interactions), but can also establish high-order relationships through multi-hop propagation, capturing deep user preferences and item associations. For example, it can indirectly infer the similarity between users or the correlation between items through multiple entities (brands, types) that users like in common.

[0046] Optionally, an adaptive implicit collaborative knowledge graph can more accurately understand user interests and item characteristics by mining implicit preferences and high-order relationships, thereby generating more precise recommendations, reducing recommendations that users are not interested in, and improving user satisfaction and retention. When user-item interaction data is sparse, by constructing an adaptive implicit collaborative knowledge graph, entity information and implicit preferences can be used to supplement the feature representations of users and items, reducing the impact of data scarcity on recommendation effectiveness. For the cold start problem of new users or new items, the system can generate preliminary feature vectors by associating with existing entities, accelerating the recommendation system's adaptation to new data.

[0047] In an optional embodiment, an adaptive implicit collaborative knowledge graph is constructed based on multiple embedding representations, including: constructing an account-item potential interaction graph based on the account embedding representation and the item embedding representation, wherein the account-item potential interaction graph is used to represent the potential interaction relationship between the account and the item; constructing a knowledge graph entity implicit relationship graph based on the entity embedding representation, wherein the knowledge graph entity implicit relationship graph is used to represent the potential association relationship between entities; and fusing the initial implicit collaborative knowledge graph, the account-item potential interaction graph, and the knowledge graph entity implicit relationship graph to obtain an adaptive implicit collaborative knowledge graph.

[0048] A knowledge graph is a structured data storage method used to represent entities (such as people, places, and objects) and their relationships (such as "belongs to" and "is located in"). It typically exists in the form of a graph, where nodes represent entities and edges represent relationships. Graph neural networks are deep learning models that specialize in processing graph-structured data. They are able to learn embedding representations of nodes that capture both local and global neighborhood information about the nodes. In the context of adaptive implicit collaborative knowledge graphs, adaptability can refer to the model's ability to dynamically adjust its parameters or structure based on data to better capture relationships between entities. Implicit collaboration can refer to inferring similarities or collaborative relationships between users by analyzing their behavior patterns and preferences in the absence of explicit user interaction data. In graph neural networks, knowledge graphs can serve as additional structural information input, helping the model better understand the relationships and attributes between nodes, thereby improving prediction accuracy. Graph neural networks can be used to learn embedding representations of entities in knowledge graphs. These embeddings can be used for tasks such as link prediction, entity disambiguation, and relationship extraction, thereby enhancing the representational capabilities of knowledge graphs.

[0049] Optionally, constructing an account-item potential interaction graph and a knowledge graph entity implicit relationship graph essentially reveals the implicit relationships between accounts, items, and entities by analyzing the similarity or correlation between account embedding representations and item and entity embedding representations. Through fusion processing, the adaptive implicit collaborative knowledge graph not only incorporates the explicit interaction information in the initial graph but also incorporates the additional relationships discovered in the account-item potential interaction graph and the knowledge graph entity implicit relationship graph. This allows the graph to more comprehensively reflect the multi-dimensional relationships between accounts, items, and entities, thereby improving the expressive power of the recommendation model. Constructing an adaptive implicit collaborative knowledge graph is a dynamic process that adapts in real time to changes in user behavior and updates to entity association information. This adaptability not only enables timely capture of changes in user interests but also adjusts relationships between items based on market dynamics and the evolution of entity information, ensuring the timeliness and personalization of recommendation results.

[0050] In an optional embodiment, an account-item potential interaction graph is constructed based on the account embedding representation and the item embedding representation, including: performing an inner product operation on the account embedding representation and the item embedding representation to obtain an account-item preference similarity matrix; screening out K items with the highest similarity between the account and the item from the account-item preference similarity matrix, where K is an integer greater than or equal to 2; and constructing the account-item potential interaction graph based on the K items.

[0051] Optionally, an inner product operation can be performed on the account embeddings and item embeddings to generate an account-item preference similarity matrix, quantifying the potential interests between accounts and items. By converting the representations of accounts and items in vector space into similarity scores, it is possible to identify items that an account may be interested in, even if these interests are not directly recorded in explicit interaction data. By selecting the K most similar items in the account-item preference similarity matrix, a latent interaction graph can be constructed for each account—a graph containing the set of items that the account is most likely to be interested in. This helps the recommender system generate more personalized recommendations that better align with the user's potential interests, improving recommendation accuracy and user satisfaction. In real applications, user-item interaction data is often very sparse, meaning that most users have not interacted with most items. By constructing the account-item latent interaction graph, the recommender system can infer which items an account may be interested in based on the account and item embeddings, thereby alleviating data sparsity to some extent, especially in cold-start scenarios. The selected K items include not only items with which the account has directly interacted in the past, but also items that the account has not directly encountered but may be interested in. This helps the recommendation system provide diverse recommendations, avoids overly monotonous recommendation lists, and improves the user experience. When constructing the potential interaction graph, inner product and filtering operations can quickly process a large number of account-item pairs while simultaneously filtering out the most relevant items. This helps improve the computational efficiency of the recommendation system, especially when working with large datasets. Furthermore, the preference similarity matrix obtained through the inner product operation can be used to introduce a regularization mechanism to optimize model parameters, reduce the risk of overfitting, and thus improve the model's generalization ability.

[0052] In an optional embodiment, a knowledge graph entity implicit relationship graph can be constructed based on the entity embedding representation, including: performing an inner product operation on the entity embedding representation and the entity embedding representation to obtain an entity-entity implicit relationship similarity matrix; from the entity-entity implicit relationship similarity matrix, screening out the K entities with the highest similarity between entities, where K is an integer greater than or equal to 2; and constructing a knowledge graph entity implicit relationship graph based on the K entities.

[0053] Optionally, an entity-entity implicit relationship similarity matrix can be generated by performing an inner product operation on the entity embedding representations. This matrix quantifies the potential connections between entities and helps capture complex relationships between item attributes or related entities. For example, two entities may not be directly connected, but their embedding representations allow the system to identify potential connections or similarities between them. The K entities with the highest similarity between them are selected from the similarity matrix to construct an implicit entity relationship graph within the knowledge graph. The value of K can be optimized based on business needs and experimental results, as it determines the density and scope of entity relationships within the graph. Constructing an implicit relationship graph can enhance the understanding of recommendation systems, enabling them to make recommendations based on deeper relationships, beyond explicit user-item interactions. Fusion of this implicit relationship graph with an adaptive implicit collaborative knowledge graph enables the model to dynamically adjust its structure based on updates to entity embedding representations, capturing changes in user preferences and evolving item attributes, thereby improving the flexibility and adaptability of the recommendation system.

[0054] Optionally, the adaptive implicit collaborative knowledge graph G can be constructed in the following ways, but is not limited to: A :

[0055] (1) Embed the user into the vector matrix V u and the item embedding vector matrix V i Perform inner product operation to obtain user-item preference similarity matrix S ui , the formula is as follows:

[0056]

[0057] Among them, W1 and W2 represent weight matrices respectively.

[0058] For each user, we set the top K maximum values ​​(topK) from a set of data to obtain the user-item preference similarity matrix S. ui The K items with the highest similarity between users and items are selected to construct a user-item potential interaction list, and then the records in the list are represented as elements in the user-item adjacency matrix, and finally the user-item potential interaction graph G is obtained. ui ;

[0059] (2) Embed the entity into the vector matrix V e and entity embedding vector matrix V e Perform inner product operation to obtain entity-entity implicit relationship similarity matrix S ee , and set a minimum threshold for the cosine similarity between entities:

[0060]

[0061] For each entity e, we set the topK to get the entity-entity implicit relationship similarity matrix S. ee The K entities with the highest similarity are selected (excluding the entity itself), and finally a list of entity-entity implicit relationships is constructed, and the elements in it are converted into elements in the entity adjacency matrix, thereby generating the knowledge graph entity implicit relationship graph G e ;

[0062] (3) The user-item potential interaction graph G ui and the knowledge graph entity implicit relationship graph G e It is integrated with the constructed implicit collaborative knowledge graph G to construct an adaptive implicit collaborative knowledge graph G A .

[0063] Optionally, the high-level programming language Python can be used to build the graph neural network of this embodiment, and the adaptive implicit collaborative knowledge graph G A , pre-define the propagation depth L and obtain high-level feature representations of users, items, and entities:

[0064] (1) Adaptive implicit collaborative knowledge graph G A Input the L-layer neural network. In this embodiment, L is set to 4, and finally the final user vector matrix Q is output. u ∈R D×M , item vector matrix Q i ∈R D×N and entity vector matrix Q e ∈R D×T , each layer of the neural network has the same structure, but uses different parameters. The structure of the l-th layer of the neural network is as follows:

[0065] a) Take the target user node u l As the center, the information of neighboring nodes in the graph, that is, the information of item nodes and target nodes themselves, are gathered to the central node, and finally the feature representation of the user vector of the lth layer is obtained:

[0066]

[0067] Among them, N ul Represents user node u l The set of neighbor nodes, |N ul | represents user node u l The number of neighbor nodes, |N i | represents the number of neighbor nodes of item node i, ∝ represents the coefficient used to control popularity;

[0068] b) Take the target item node i lAs the center, the information of neighboring nodes in the graph, namely user nodes and entity nodes, as well as the target node’s own information, are gathered to the central node, and finally the feature representation of the item vector in the lth layer is obtained:

[0069]

[0070] in, Represents item node i l The set of neighboring user nodes, Represents item node i l The set of neighbor entity nodes, |N il | represents item node i l The number of neighbor nodes, |N u | represents the number of neighbor nodes of item node u;

[0071] c) Take the target entity node e l As the center, the information of neighboring nodes in the graph, namely entity nodes and item nodes, as well as the target node’s own information, are gathered to the central node, and finally the feature representation of the entity vector at the lth layer is obtained:

[0072]

[0073] in, Represents entity node e l A collection of neighbor item nodes, Represents entity node e l The set of neighbor entity nodes, |N el | represents entity node e l The number of neighbor nodes, |N e | represents the number of neighbor nodes of entity node e.

[0074] Step S108: Based on the adaptive implicit collaborative knowledge graph, an item recommendation model is used to obtain an item recommendation result corresponding to the target account, wherein the target account is an account in the adaptive implicit collaborative knowledge graph.

[0075] Optionally, the adaptive implicit collaborative knowledge graph contains explicit interactions and implicit preference information between accounts (i.e., users), items, and entities. The item recommendation model can leverage this information, including account preferences, item attributes, associations between entities, and implicit relationships between them, to predict which items the target account may be interested in. This process is key to achieving personalized recommendations, generating a customized list of recommendations for each account. After receiving recommendations, the system can collect user feedback on the recommended items, including clicks, purchases, and ratings. This feedback is used to further optimize the adaptive implicit collaborative knowledge graph, adjusting the embedding vectors of accounts, items, and entities, as well as the parameters of the item recommendation model, thereby forming a closed-loop learning and optimization process to continuously improve recommendation performance. For new accounts or new items, due to the lack of interaction data, the adaptive implicit collaborative knowledge graph can leverage the potential associations between accounts and items and entities, as well as the implicit relationships between entities, to generate recommendations for new accounts, alleviating the cold start problem. Furthermore, for accounts with sparse data, the entity information and implicit preferences in the graph can supplement the lack of direct interaction information and improve the accuracy of recommendations.

[0076] Optionally, after obtaining the item recommendation results corresponding to the target account, the item recommendation results may also be sent to the target user's corresponding client interface for display.

[0077] In an optional embodiment, the item recommendation result corresponding to the target account can be obtained by adopting an item recommendation model based on an adaptive implicit collaborative knowledge graph, including: based on the adaptive implicit collaborative knowledge graph, adopting the graph neural network in the item recommendation model to obtain a first feature representation corresponding to the account vector and a second feature representation corresponding to the item vector; performing an inner product operation on the first feature representation and the second feature representation to obtain the probability of interaction between the account and the item included in the adaptive implicit collaborative knowledge graph; and determining the item recommendation result corresponding to the target account based on the probability of interaction between the account and the item included in the adaptive implicit collaborative knowledge graph.

[0078] Optionally, high-order feature representations of accounts and items can be learned by executing a graph neural network on an adaptive implicit collaborative knowledge graph. These feature representations not only contain explicit interaction information, but also incorporate the account's potential interest in items and the implicit preferences and associations between items and entities, thereby forming a more comprehensive and in-depth representation of users and items. Through the inner product operation, the model is able to calculate the similarity between the first feature representation of the account vector and the second feature representation of the item vector, and then predict the probability of interaction between the account and the item. This probability prediction directly determines which items will be recommended to a specific account. Based on the interaction probability between the account and the item, a list of item recommendations can be generated for each target account. This list will be sorted from high to low according to the predicted interaction probability, ensuring that the items most likely to be of interest to the account are at the front of the list.

[0079] Optionally, the recommendation model can be optimized in the following way. Specifically, the output representation of each layer vector of user u can be obtained by iteratively propagating the graph convolution of the L layer in the recommendation model, that is, Similarly, the output representation of each layer vector of item i can be obtained And the output representation of each layer vector of entity e The final vector of user u is expressed as The final vector of user i is expressed as This embodiment adopts a layer selection mechanism, that is, it uses the user vector and item vector output by the last two layers to predict the probability of user-item interaction, where f is the inner product function:

[0080]

[0081]

[0082] Optionally, the recommendation model can be optimized using the following joint learning objective function optimization model:

[0083]

[0084] Among them, θ refers to all learnable parameters in the model, namely [W1,W2,V u ,V i ,V e ], λ is the regularization coefficient; Figure 3 It is an optional graph neural network module diagram according to an embodiment of the invention.

[0085] a)L CF The loss function (Bayesian Personalized Ranking, BPR) loss commonly used in recommendation systems and ranking learning problems in the recommendation field is adopted:

[0086]

[0087] Among them, Y={(u,i,j)(u,i)∈Y + ,(u,j)∈Y -}, Y represents the training sample of user-item interaction data, Y+ represents the positive sample, that is, it means that there is an interaction sample between user u and item i, Y - represents a negative sample, i.e., a sample with no interaction between user u and item j; σ(·) is the sigmoid activation function;

[0088] b)L KG BPR loss can be used:

[0089]

[0090] Among them, Z={(i,e,v)(i,e)∈Z + ,(i,v)∈Z -}, z represents the training sample of the item-entity implicit knowledge graph, Z + represents a positive sample, that is, a sample that indicates interaction between item i and entity e, Z - Represents a negative sample, that is, a sample in which there is no interaction between item i and entity v.

[0091] c) is the objective function of the enhanced graph learning optimization designed in this embodiment, which can minimize the difference between the actual user-item bipartite graph G1 and the user-item potential interaction graph G obtained after similarity calculation. ui To ensure that the predicted results are consistent with the actual results.

[0092]

[0093] Optionally, initialize the user, item, and entity embedding vector dimensions to D, with an empirical value of 64, and use a Gaussian distribution with a mean of 0 and a standard deviation of 0.01 to initialize the embedding vector matrix. ui Filter out the K items with the highest similarity between users and items, and the empirical value is [1,10]. By setting topK, we can get the similarity matrix S of entity-entity implicit relationship. eeThe K entities with the highest similarity are selected from the dataset, with an empirical value of [1,5]. A minimum threshold is set for the cosine similarity between entities, with an empirical value of 0.5. The number of graph neural network layers, L, is empirically set to 4. The popularity bias coefficient α in the graph neural network is empirically adjusted between {0.3, 0.4, 0.5}. The training samples serve as data samples for model fitting. Since this embodiment trains the model based on the BPR loss function, negative samples with the same number of interaction histories are randomly sampled for each user in the training set. The model predicts and outputs the predicted probability of whether the target user will click on the item. The items are sorted and displayed to the user.

[0094] Optionally, training samples for training and optimizing the recommendation model can be: user-item interaction data, which records explicit interactions between users and items, such as clicks, purchases, and ratings. Item-entity triplet data, which contains the relationship between items and related entities (item attribute information or information associated with items). Through these triplet data, an implicit item-entity graph can be constructed, which is then integrated into the user-item binary graph to form the initial implicit collaborative knowledge graph G. In addition to direct interaction data, indirect user behavior data such as search history, browsing behavior, and favorites on the platform are also considered part of the training samples. These data can provide clues to users' implicit preferences.

[0095] Optionally, negative samples in a recommender system refer to samples where no interaction between the user and the item is observed. In this method, negative samples refer to data points where there is no interaction record between the user and the item, which can be items that the user has never clicked, purchased, or rated.

[0096] Through the above steps S102 to S108, the purpose of better capturing the missing information of the graph neural network nodes can be achieved, thereby achieving a technical effect of higher accuracy and better effect of recommendation, and then solving the technical problem of poor item recommendation effect caused by the large sparsity of interaction data and knowledge graph.

[0097] Based on the above embodiments and optional embodiments, the present invention proposes an optional implementation mode: Figure 4 is a flow chart of an optional item recommendation method according to an embodiment of the present invention. Figure 4 As shown, the method includes:

[0098] S1, constructs an implicit collaborative knowledge graph G based on training samples, specifically including:

[0099] S11. Constructing implicit object-entity graph;

[0100] S12. Construct a user-item binary graph;

[0101] S13. Fuse item-entity pairs into the user-item binary graph to construct an implicit collaborative knowledge graph G;

[0102] S2, initialize the user, item and entity embedding vectors to obtain the user embedding vector matrix V u , item embedding vector matrix V i , and the entity embedding vector matrix V e ;

[0103] S3, building an adaptive implicit collaborative knowledge graph, specifically including:

[0104] S31. Perform an inner product operation on the user embedding vector matrix and the item embedding vector matrix to obtain a user-item preference similarity matrix;

[0105] S32. Perform an inner product operation on the entity embedding vector matrix and the entity embedding vector matrix to obtain an entity-entity implicit relationship similarity matrix;

[0106] S33. Transform the user-item potential interaction graph G ui and the knowledge graph entity implicit relationship graph G e Fuse with the implicit collaborative knowledge graph G constructed in step S1 to construct an adaptive implicit collaborative knowledge graph;

[0107] S4, the obtained adaptive implicit collaborative knowledge graph is input into the enhanced graph neural network module. After L layers of graph convolution iterative propagation, the output representation of each layer vector of user u is obtained, that is, Similarly, the output representation of each layer vector of item i can be obtained And the output representation of each layer vector of entity e Using the user vector and item vector output by the last two layers, the inner product operation is performed to predict the probability of user and item interaction.

[0108] S5, constructing a joint learning objective function to optimize model parameters, wherein the specific construction method of the objective function is the same as described above and will not be repeated here;

[0109] S6, output the probability of the user clicking on the item.

[0110] It should be noted that this embodiment constructs an adaptive implicit collaborative knowledge graph, utilizing the currently learned node (user, item, and entity) embedding vectors to guide the construction of the user-item potential interaction graph and the knowledge graph entity implicit relationship graph, thereby updating the implicit collaborative knowledge graph structure. The updated implicit collaborative knowledge graph also better serves the representation learning of node embedding vectors, further improving recommendation performance. Furthermore, the embodiment designs corresponding graph enhancement functions during the optimization process to better optimize model parameters and more fully capture the potential interaction relationships between users and items and the implicit relationships between entities.

[0111] In this embodiment, an item recommendation device is also provided, which is used to implement the above-mentioned embodiments and preferred implementations. The details that have been described will not be repeated here. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0112] According to an embodiment of the present invention, there is also provided an embodiment of a device for implementing the above-mentioned item recommendation method. Figure 5 FIG. 1 is a schematic diagram of the structure of an item recommendation device according to an embodiment of the present invention. Figure 5 As shown, the above-mentioned item recommendation device includes: a construction module 200, a representation module 202, an association module 204, and a target module 206, wherein:

[0113] A construction module 200 is used to construct an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent explicit interactions and implicit preferences and associations between accounts, items, and entities, and entities are attribute information of items or association information associated with items;

[0114] A representation module 202, connected to the construction module 200, is configured to construct a plurality of embedding representations based on the initial implicit collaborative knowledge graph, wherein the plurality of embedding representations include an account embedding representation, an item embedding representation, and an entity embedding representation;

[0115] An association module 204, connected to the representation module 202, is configured to construct an adaptive implicit collaborative knowledge graph based on the multiple embedded representations, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in items, as well as implicit preferences and associations between items and entities;

[0116] The target module 206 is connected to the association module 204 and is used to obtain the item recommendation results corresponding to the target account based on the adaptive implicit collaborative knowledge graph and the item recommendation model, wherein the target account is an account in the adaptive implicit collaborative knowledge graph.

[0117] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0118] It should be noted that the construction module 200, representation module 202, association module 204, and target module 206 correspond to steps S102 to S108 in the embodiment. The examples and application scenarios implemented by these modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment. It should be noted that these modules, as part of the device, can be run on a computer terminal.

[0119] It should be noted that the optional or preferred implementation of this embodiment can be found in the relevant description in the embodiment, which will not be repeated here.

[0120] The above-mentioned item recommendation device may also include a processor and a memory. The above-mentioned construction module 200, representation module 202, association module 204, target module 206, etc. are all stored in the memory as program modules, and the processor executes the above-mentioned program modules stored in the memory to realize corresponding functions.

[0121] The processor includes a core, which retrieves corresponding program modules from memory. There can be one or more cores. Memory may include non-permanent memory in a computer-readable medium, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0122] According to an embodiment of the present application, an embodiment of a non-volatile storage medium is further provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein when the program is executed, the device containing the non-volatile storage medium is controlled to execute any of the above-mentioned item recommendation methods.

[0123] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group, and the non-volatile storage medium includes a stored program.

[0124] Optionally, when the program is running, the device where the non-volatile storage medium is located is controlled to perform the following functions: constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent the explicit interaction relationship and implicit preferences and associations between accounts, items and entities, and the entity is the attribute information of the item, or the association information associated with the item; based on the initial implicit collaborative knowledge graph, constructing multiple embedding representations, wherein the multiple embedding representations include account embedding representation, item embedding representation and entity embedding representation; based on the multiple embedding representations, constructing an adaptive implicit collaborative knowledge graph, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preferences and associations between the item and the entity; based on the adaptive implicit collaborative knowledge graph, using the item recommendation model to obtain the item recommendation results corresponding to the target account, wherein the target account is the account in the adaptive implicit collaborative knowledge graph.

[0125] According to an embodiment of the present application, an embodiment of a processor is further provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-mentioned item recommendation methods when it is run.

[0126] According to an embodiment of the present application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program that, when executed by a processor, implements any of the steps of the above-mentioned item recommendation method.

[0127] Optionally, the above-mentioned computer program product, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent the explicit interactive relationship and implicit preferences and associations between accounts, items and entities, and the entity is the attribute information of the item, or the association information associated with the item; based on the initial implicit collaborative knowledge graph, constructing multiple embedding representations, wherein the multiple embedding representations include account embedding representations, item embedding representations and entity embedding representations; based on the multiple embedding representations, constructing an adaptive implicit collaborative knowledge graph, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preferences and associations between the item and the entity; based on the adaptive implicit collaborative knowledge graph, using the item recommendation model to obtain the item recommendation results corresponding to the target account, wherein the target account is the account in the adaptive implicit collaborative knowledge graph.

[0128] An embodiment of the present invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are implemented: constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent explicit interactive relationships and implicit preferences and associations between accounts, items, and entities, and the entity is the attribute information of the item, or the association information associated with the item; based on the initial implicit collaborative knowledge graph, constructing multiple embedding representations, wherein the multiple embedding representations include account embedding representations, item embedding representations, and entity embedding representations; based on the multiple embedding representations, constructing an adaptive implicit collaborative knowledge graph, wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preferences and associations between the item and the entity; based on the adaptive implicit collaborative knowledge graph, using an item recommendation model, obtaining item recommendation results corresponding to the target account, wherein the target account is an account in the adaptive implicit collaborative knowledge graph. The order of the above embodiments of the present invention is only for description and does not represent the advantages or disadvantages of the embodiments.

[0129] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0130] 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 exemplary. For example, the division of the above modules can be a logical function division. In actual implementation, there may be other division methods, such as multiple modules 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, modules or indirect coupling or communication connection of modules, which can be electrical or other forms.

[0131] The modules described above as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0132] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or software functional modules.

[0133] If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a non-volatile storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned non-volatile storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, and other media that can store program codes.

[0134] The above are only preferred embodiments of the present invention. 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 invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An item recommendation method, characterized in that: include: Constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent explicit interactions and implicit preferences and associations between accounts, items, and entities, where the entities are attribute information of items or association information associated with items; Based on the initial implicit collaborative knowledge graph, construct a plurality of embedding representations, wherein the plurality of embedding representations include an account embedding representation, an item embedding representation, and an entity embedding representation; Based on the multiple embedding representations, an adaptive implicit collaborative knowledge graph is constructed, including: based on the account embedding representation and the item embedding representation, an account-item potential interaction graph is constructed, wherein the account-item potential interaction graph is used to represent the potential interaction relationship between the account and the item; based on the entity embedding representation, a knowledge graph entity implicit relationship graph is constructed, wherein the knowledge graph entity implicit relationship graph is used to represent the potential association relationship between entities; the initial implicit collaborative knowledge graph, the account-item potential interaction graph and the knowledge graph entity implicit relationship graph are fused to obtain the adaptive implicit collaborative knowledge graph; wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preference and association between the item and the entity; Based on the adaptive implicit collaborative knowledge graph, an item recommendation model is adopted to obtain item recommendation results corresponding to a target account, wherein the target account is an account in the adaptive implicit collaborative knowledge graph.

2. The method according to claim 1, characterized in that The construction of the initial implicit collaborative knowledge graph includes: Constructing a plurality of triples, wherein the triples include items, entities, and implicit relationships between items and entities; Based on the interactive operation behavior between accounts and items, a binary graph is constructed; Using items as a bridge to connect accounts and entities, the multiple triples are integrated into the binary graph to obtain the initial implicit collaborative knowledge graph.

3. The method according to claim 1, characterized in that The step of constructing multiple embedding representations based on the initial implicit collaborative knowledge graph includes: The accounts, items, and entities in the initial implicit collaborative knowledge graph are mapped to a vector space to obtain the multiple embedding representations, wherein the multiple embedding representations are in the form of vector matrices.

4. The method according to claim 1, wherein The constructing of an account-item potential interaction graph based on the account embedding representation and the item embedding representation includes: Performing an inner product operation on the account embedding representation and the item embedding representation to obtain an account-item preference similarity matrix; Filtering the K items with the highest similarity between the account and the item from the account-item preference similarity matrix, where K is an integer greater than or equal to 2; Based on the K items, the account-item potential interaction graph is constructed.

5. The method according to claim 1, wherein The step of constructing a knowledge graph entity implicit relationship diagram based on the entity embedding representation includes: Performing an inner product operation on the entity embedding representation and the entity embedding representation to obtain an entity-entity implicit relationship similarity matrix; Filtering out K entities with the highest similarity between entities from the entity-entity implicit relationship similarity matrix, where K is an integer greater than or equal to 2; Based on the K entities, an implicit relationship graph of the knowledge graph entities is constructed.

6. The method according to any one of claims 1 to 5, characterized in that The method of using an item recommendation model based on the adaptive implicit collaborative knowledge graph to obtain an item recommendation result corresponding to the target account includes: Based on the adaptive implicit collaborative knowledge graph, the graph neural network in the item recommendation model is used to obtain a first feature representation corresponding to the account vector and a second feature representation corresponding to the item vector; Performing an inner product operation on the first feature representation and the second feature representation to obtain a probability of interaction between accounts and items included in the adaptive implicit collaborative knowledge graph; Based on the probability of interaction between accounts and items included in the adaptive implicit collaborative knowledge graph, the item recommendation result corresponding to the target account is determined.

7. An item recommendation device, characterized in that: include: A construction module for constructing an initial implicit collaborative knowledge graph, wherein the initial implicit collaborative knowledge graph is used to represent explicit interactions and implicit preferences and associations between accounts, items, and entities, and the entities are attribute information of items or association information associated with items; A representation module, configured to construct a plurality of embedding representations based on the initial implicit collaborative knowledge graph, wherein the plurality of embedding representations include an account embedding representation, an item embedding representation, and an entity embedding representation; An association module is configured to construct an adaptive implicit collaborative knowledge graph based on the multiple embedding representations, including: constructing an account-item potential interaction graph based on the account embedding representation and the item embedding representation, wherein the account-item potential interaction graph is used to represent the potential interaction relationship between the account and the item; constructing a knowledge graph entity implicit relationship graph based on the entity embedding representation, wherein the knowledge graph entity implicit relationship graph is used to represent the potential association relationship between entities; fusing the initial implicit collaborative knowledge graph, the account-item potential interaction graph, and the knowledge graph entity implicit relationship graph to obtain the adaptive implicit collaborative knowledge graph; wherein the adaptive implicit collaborative knowledge graph is used to represent the account's potential interest in the item, as well as the implicit preference and association between the item and the entity; The target module is used to obtain the item recommendation results corresponding to the target account based on the adaptive implicit collaborative knowledge graph and using the item recommendation model, wherein the target account is an account in the adaptive implicit collaborative knowledge graph.

8. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the item recommendation method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: The invention comprises one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the item recommendation method according to any one of claims 1 to 6.

Images