Intelligent recommendation method and device, equipment and storage medium

Abstract

The invention provides an intelligent recommendation method which can be applied to the technical field of artificial intelligence. The method comprises the following steps: processing input multi-dimensional data by utilizing a pre-trained first recommendation model to obtain an optimal recommendation result; the processing process of the first recommendation model comprises the following steps: strengthening multi-dimensional data based on a pre-trained agent to obtain an enhanced knowledge graph; inputting the intelligent agent based on the enhanced knowledge graph to reversely adjust the parameters of the intelligent agent to obtain the adjusted intelligent agent; based on the enhanced knowledge graph, retrieving the multi-dimensional data, and obtaining an initial recommendation result by using a pre-trained second recommendation model according to the retrieved data; and integrating the adjusted intelligent agent with the initial recommendation result to generate an optimal recommendation result. The invention further provides an intelligent recommendation device, equipment and a storage medium.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, specifically the field of recommendation algorithms, and more specifically to an intelligent recommendation method, apparatus, device, and storage medium. Background Art

[0002] Continuous learning is a key branch of personalized recommendation system research. However, existing continuous learning processes face numerous challenges when dealing with complex user behavior patterns and product attributes. On the user side, shifting user interests and timeliness lead to delayed recommendation effectiveness. On the product side, long-tail products lack value responsiveness. Some products may have low exposure but appeal to specific user groups, resulting in their market potential not being fully tapped. On the model side, there is a need to balance the needs of personalization and diversity, but models often overly rely on patterns in the training data and fail to generalize well.

[0003] Therefore, existing recommendation systems have obvious deficiencies in diversity, depth of understanding, and interpretability, and cannot accurately provide users with high-quality recommendation results. Summary of the Invention

[0004] In view of the above problems, the present application provides an intelligent recommendation method, apparatus, device and storage medium for improving recommendation accuracy.

[0005] According to the first aspect of the present application, an intelligent recommendation method is provided, comprising: using a pre-trained first recommendation model to process input multidimensional data to obtain a preferred recommendation result; the method of using the pre-trained first recommendation model to process the input multidimensional data to obtain a preferred recommendation result comprises: strengthening the multidimensional data based on a pre-trained intelligent agent to obtain an enhanced knowledge graph; inputting the intelligent agent based on the enhanced knowledge graph to reversely adjust the intelligent agent parameters to obtain an adjusted intelligent agent; retrieving the multidimensional data based on the enhanced knowledge graph, and obtaining an initial recommendation result based on the retrieved data using a pre-trained second recommendation model; and integrating the adjusted intelligent agent with the initial recommendation result to generate the preferred recommendation result.

[0006] According to an embodiment of the present application, the strengthening of the multidimensional data based on a pre-trained intelligent agent to obtain an enhanced knowledge graph includes: based on the intelligent agent, performing retrieval enhancement on the multidimensional data to obtain retrieval-enhanced multidimensional data; based on the intelligent agent, performing relationship enhancement on the first multidimensional data to obtain relationship-enhanced multidimensional data, wherein the first multidimensional data includes the multidimensional data and / or the retrieval-enhanced multidimensional data; based on the intelligent agent, performing reasoning enhancement on the second multidimensional data to obtain reasoning-enhanced multidimensional data, wherein the second multidimensional data includes one or more of the multidimensional data, the retrieval-enhanced multidimensional data and the relationship-enhanced multidimensional data; and constructing the enhanced knowledge graph based on the third multidimensional data, wherein the third multidimensional data includes one or more of the multidimensional data, the retrieval-enhanced multidimensional data, the relationship-enhanced multidimensional data and the reasoning-enhanced multidimensional data.

[0007] According to an embodiment of the present application, the relationship enhancement of the first multidimensional data based on the intelligent agent to obtain the relationship-enhanced multidimensional data includes: based on the intelligent agent, improving the knowledge nodes extracted from the first multidimensional data; and based on the intelligent agent, vectorizing the representation of the extracted knowledge nodes to obtain the relationship-enhanced multidimensional data.

[0008] According to an embodiment of the present application, the improved extraction of knowledge nodes of the first multidimensional data based on the intelligent agent includes: performing entity extraction on the first multidimensional data to obtain entities; outputting an entity sequence based on a preset knowledge node tuple; rewriting the first multidimensional data; performing entity extraction on the rewritten first multidimensional data to obtain rewritten entities; and adjusting the entity sequence according to the rewritten entities to output the extracted knowledge nodes.

[0009] According to an embodiment of the present application, the vectorized representation of the extracted knowledge nodes based on the intelligent agent to obtain the relationship-enhanced multidimensional data includes: obtaining a graph structure element vector based on the first multidimensional data; performing text encoding on the extracted knowledge nodes based on the intelligent agent to obtain a text element vector; fusing the graph structure element vector and the text element vector to obtain node features; and aggregating node features of neighboring nodes based on the extracted knowledge nodes to obtain a node feature matrix to obtain the relationship-enhanced multidimensional data corresponding to the extracted knowledge nodes.

[0010] According to an embodiment of the present application, the method further includes: in the process of training the first recommendation model, constructing a cross-entropy loss corresponding to the enhanced knowledge graph, and constructing a ranking loss corresponding to the second recommendation model; determining the joint loss of the first recommendation model based on the cross-entropy loss and the ranking loss; and minimizing the joint loss.

[0011] According to an embodiment of the present application, the method further includes: reversely adjusting the enhanced knowledge graph based on the preferred recommendation result.

[0012] The second aspect of the present application provides an intelligent recommendation device, comprising: a recommendation module, used to use a pre-trained first recommendation model to process input multidimensional data to obtain a preferred recommendation result; the recommendation module comprises: a graph enhancement submodule, used to enhance the multidimensional data based on a pre-trained intelligent agent to obtain an enhanced knowledge graph; a feedback fine-tuning submodule, used to input the intelligent agent based on the enhanced knowledge graph to reversely adjust the intelligent agent parameters to obtain an adjusted intelligent agent; an initial recommendation submodule, used to retrieve the multidimensional data based on the enhanced knowledge graph, and obtain an initial recommendation result based on the retrieved data using a pre-trained second recommendation model; and a result generation submodule, used to integrate the adjusted intelligent agent with the initial recommendation result to generate the preferred recommendation result.

[0013] The third aspect of the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0014] The fourth aspect of the present application further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the computer program or instructions are executed by a processor.

[0015] The fifth aspect of the present application further provides a computer program product, comprising a computer program or instructions, which implement the steps of the above method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above contents and other objects, features and advantages of the present application will become more apparent through the following description of the embodiments of the present application with reference to the accompanying drawings, in which:

[0017] Figure 1 The following schematically illustrates an application scenario of the intelligent recommendation method according to an embodiment of the present application;

[0018] Figure 2 Schematically shows a first recommendation model processing flow chart of the intelligent recommendation method according to an embodiment of the present application;

[0019] Figure 3 A model interaction diagram of an intelligent recommendation method according to an embodiment of the present application is schematically shown;

[0020] Figure 4The following schematically shows a flow chart for constructing a large model of the intelligent recommendation method according to an embodiment of the present application;

[0021] Figure 5 The following schematically shows a flow chart of enhancing the knowledge graph of the intelligent recommendation method according to an embodiment of the present application;

[0022] Figure 6 Schematically shows a relationship enhancement flow chart of the intelligent recommendation method according to an embodiment of the present application;

[0023] Figure 7 A schematic diagram of a vectorized representation structure of an intelligent recommendation method according to an embodiment of the present application is shown;

[0024] Figure 8 A schematic diagram of a structure of an intelligent recommendation device according to an embodiment of the present application is shown; and

[0025] Figure 9 A block diagram of an electronic device suitable for implementing the intelligent recommendation method according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0026] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present application.

[0027] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "comprise," "include," etc. used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0029] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0030] With the development of intelligence, recommendation system models are diverse, but the following defects still exist:

[0031] 1. Information cocoon effect: Recommendation systems rely on historical user behavior data, which can easily limit the results to the user's known areas of interest, ignore the user's potential interests, and reduce the novelty and exploratory nature of the recommendation list.

[0032] 2. Limited understanding: The model's understanding of user preferences and product features often remains superficial, lacking effective exploration of deeper relationships (such as semantic connections and contextual associations). This shallow understanding fails to capture the essence of user needs, thus affecting the relevance and accuracy of recommendations.

[0033] 3. Lack of explainability: The internal operating mechanisms of recommendation models present a "black box" problem. This creates barriers to understanding and trust for business decision makers. This lack of transparency can undermine their confidence, especially when making critical decisions based on recommendation results.

[0034] In summary, in the process of continuous learning, existing recommendation systems have obvious deficiencies in diversity, depth of understanding, and interpretability, and cannot accurately provide users with high-quality recommendation results. Therefore, this application combines the advantages of large models and knowledge graphs to design a mutually beneficial and two-way enhanced product recommendation solution, which can effectively solve the above problems and further improve the overall performance and user experience of the recommendation system. By combining the advantages of both, not only can the diversity and relevance of recommendations be improved, but also the interpretability and reliability of the system can be enhanced, providing users with more accurate and personalized recommendation services.

[0035] Therefore, an embodiment of the present application provides an intelligent recommendation method, including: using a pre-trained first recommendation model to process input multidimensional data to obtain a preferred recommendation result; using the pre-trained first recommendation model to process input multidimensional data to obtain a preferred recommendation result, including: strengthening the multidimensional data based on the pre-trained intelligent agent to obtain an enhanced knowledge graph; inputting the intelligent agent based on the enhanced knowledge graph to reversely adjust the intelligent agent parameters to obtain an adjusted intelligent agent; retrieving the multidimensional data based on the enhanced knowledge graph, and obtaining an initial recommendation result based on the retrieved data using a pre-trained second recommendation model; integrating the adjusted intelligent agent with the initial recommendation result to generate a preferred recommendation result. By constructing an enhanced knowledge graph based on the intelligent agent, the enhanced knowledge graph reversely adjusts the intelligent agent, and combining it with the second recommendation model, a dual-drive intelligent recommendation model is formed, namely the first recommendation model. Using the first recommendation model, a mutually beneficial and two-way enhanced product recommendation solution is formed, which can effectively solve the problems of the existing recommendation system, improve the accuracy of recommendations, and enhance the overall performance and user experience of the recommendation solution. Moreover, by integrating the advantages of intelligent agents, knowledge graphs, and recommendation models, the two-way drive can significantly increase the freshness and exploratory nature of recommendation results, which not only improves the diversity and relevance of recommendations, but also enhances the interpretability and reliability of the recommendation system, providing users with more accurate and personalized recommendation services.

[0036] Figure 1 The following schematically illustrates an application scenario diagram of the intelligent recommendation method according to an embodiment of the present application.

[0037] like Figure 1 As shown, the application scenario 100 according to this embodiment may include, but is not limited to, recommendation scenarios related to e-commerce shopping, educational products, travel, learning and education, financial products, etc. The network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0038] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0039] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0040] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0041] It should be noted that the intelligent recommendation method provided in the embodiment of the present application can generally be executed by the server 105. Accordingly, the intelligent recommendation device provided in the embodiment of the present application can generally be set in the server 105. The intelligent recommendation method provided in the embodiment of the present application can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the intelligent recommendation device provided in the embodiment of the present application can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0042] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0043] The following will be based on Figure 1 The scene described by Figures 2 to 6 The intelligent recommendation method according to the embodiment of the present application is described in detail.

[0044] Figure 2 The following schematically shows a first recommendation model processing flow chart of the intelligent recommendation method according to an embodiment of the present application. Figure 3 A model interaction diagram of the intelligent recommendation method according to an embodiment of the present application is schematically shown.

[0045] According to an embodiment of the present application, the intelligent recommendation method includes using a pre-trained first recommendation model to process input multi-dimensional data to obtain a preferred recommendation result.

[0046] This intelligent recommendation method can be automated, acquiring multidimensional data based on a recommendation request, inputting the multidimensional data into a pre-trained first recommendation model, and automatically outputting a preferred recommendation result. The multidimensional data includes one or more heterogeneous multidimensional data types, such as product information, behavioral information, and user attributes. Product information includes basic information, categories, and prices, while behavioral information includes records of users' browsing, purchases, and reviews. User attributes include personal information and preferences.

[0047] It should be noted that in the technical solution of this application, the user information involved (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0048] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided in the embodiments of the present application all provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0049] The first recommendation model is a recommendation model based on the bidirectional drive of intelligent agents and knowledge graphs, which adopts a mutually beneficial dual-drive recommendation framework, such as Figure 3 As shown, the intelligent agent strengthens the data to enhance the knowledge graph, and the enhanced knowledge graph reversely adjusts the intelligent agent to build a mutually beneficial and bidirectional enhancement model.

[0050] The first recommendation model combines the advantages of knowledge graphs and large-scale language models to enable the graph to provide precise entity relationships and domain knowledge, while the intelligent agent provides general knowledge and content generation capabilities. While ensuring data verifiability and traceability, it can perform more complex multi-step reasoning and contextual reasoning tasks, thereby providing more intelligent and personalized recommendations and meeting the specific needs of different industries and fields. Based on multi-dimensional data such as leads, opportunities, product information, users and their activity records, and implicit content information of large models, and with the design goal of maximizing the benefits of multilateral recommendations, the intelligent agent and knowledge graph are integrated into the first recommendation model's "generation (data), production (knowledge), supply (model), and sales (results)" at multiple levels of breadth and depth to address the shortcomings of existing recommendation models in terms of diversity, depth of understanding, and explainability.

[0051] The first recommendation model uses historical multidimensional data as training data. The historical multidimensional data records historical behaviors, attributes, and related multidimensional data from multiple dimensions around entities such as users, products, and scenarios. The training process of the first recommendation model includes: strengthening historical multidimensional data based on an intelligent agent to obtain a historically enhanced knowledge graph; inputting the intelligent agent based on the historically enhanced knowledge graph to reversely adjust the intelligent agent parameters to obtain a historically adjusted intelligent agent; searching the historical multidimensional data based on the historically enhanced knowledge graph, and using the second recommendation model based on the historically retrieved data to obtain historical initial recommendation results; integrating the historically adjusted intelligent agent with the historical initial recommendation results to generate historically preferred recommendation results; and reversely adjusting the historically enhanced knowledge graph based on the historically preferred recommendation results.

[0052] According to an embodiment of the present application, in the process of training the first recommendation model, the cross-entropy loss corresponding to the enhanced knowledge graph is constructed, and the ranking loss corresponding to the second recommendation model is constructed; based on the cross-entropy loss and the ranking loss, the joint loss of the first recommendation model is determined; and the joint loss is minimized.

[0053] The enhanced knowledge graph after agent reinforcement is embedded as an additional feature into the training process of the second recommendation model to construct the cross entropy loss of the graph for:

[0054] (1)

[0055] in, is the sample size, It is The actual labels of the samples, It is the probability that the spectrum estimates the sample to be positive.

[0056] Constructing the ranking loss of the relative ranking of the second recommendation model for:

[0057] (2)

[0058] in, is the true distribution User's The probability that an item is selected, M is the number of items, is the probability estimated by the second recommendation model.

[0059] The joint loss of the first recommendation model is Expressed as:

[0060] (3)

[0061] in, is the weight coefficient, , used to balance the importance of each task, and finally establish a closed-loop recommendation model so that each recommendation result can become the basis for the next learning, and continuously iterate to optimize the recommendation quality.

[0062] In an embodiment of the present application, based on the cross-entropy loss of the enhanced knowledge graph and the ranking loss of the recommendation model, a joint loss is determined, and the knowledge embedding and recommendation tasks are jointly optimized to balance the importance of the two tasks and improve the generalization and recommendation accuracy of the dual-drive recommendation model.

[0063] Intelligent agents are deep learning-based AI models, such as large language models (LLMs), capable of understanding natural language, generating responses, and making inferences and decisions. Instructions are fed into the agent, which then outputs corresponding results. Through training on large-scale text data, the agent acquires language understanding and generation capabilities. Combined with domain-specific knowledge or tool invocations, it enables interactive conversations, information processing, and task execution. These agents are widely used in scenarios such as intelligent question-and-answering, content creation, and automated processes. By collecting historical multidimensional data such as financial transactions and user behavior, and cleaning and preprocessing it, a corpus containing graph information is generated. This graph information corpus and historical multidimensional data are then used to train the agent. Through supervised learning and reinforcement learning algorithms, the agent can learn patterns and associations within the data, becoming an agent well-suited for recommendation tasks.

[0064] Taking the agent as a large language model (hereinafter referred to as the large model) as an example, Figure 4 The following schematically shows a flowchart for constructing a large model of the intelligent recommendation method according to an embodiment of the present application.

[0065] Figure 4As shown, the scheduling system parses recommendation requests to obtain multidimensional heterogeneous data such as user and attribute data. This multidimensional heterogeneous data is then used as input for the large model. First, the large model performs data padding and expansion on this input data to achieve data expansion. The large model then extracts and improves useful knowledge nodes from this multidimensional heterogeneous data to achieve improved extraction. The large model then vectorizes the knowledge nodes and embedded text to obtain vectorized representations of entities and relationships, achieving optimized representation. The large model then deeply analyzes unstructured data such as user behavior logs in contextual learning using an attention mechanism to obtain a fused knowledge graph and achieve enhanced modeling. Secondly, the knowledge graph corpus is fed into the large model for enhancement, outputting enhanced knowledge graphs corresponding to retrieval enhancement, relationship enhancement, and reasoning enhancement. The enhanced knowledge graph is then used to fine-tune the parameters of the large model. This process generates a graph-enhanced large model, which can then be used to further organize and output recommended content.

[0066] The second recommendation model is based on graph relationships. It takes the output graph data of the knowledge graph as input and outputs recommendation results. This model can include models based on knowledge graph embedding, graph neural network models, deep learning models, and other models. The training process for the second recommendation model involves inputting historical search data into the model and optimizing the model parameters using a loss function (such as pairwise ranking loss). This allows the model to learn the mapping between entity relationships in the graph and user preferences. Through continuous iteration, the model parameters are adjusted to enhance its ability to capture user interests and item associations, ultimately resulting in a model that can be used for recommendations.

[0067] like Figure 2 、 3 As shown, the above-mentioned process of processing the input multi-dimensional data using the pre-trained first recommendation model to obtain the preferred recommendation result includes operations S210 to S240.

[0068] In operation S210, multidimensional data is enhanced based on the pre-trained agent to obtain an enhanced knowledge graph.

[0069] Based on pre-trained intelligent agents, the input multi-dimensional data is enhanced and processed to output an enhanced knowledge graph. This enhanced processing includes sample expansion, knowledge extraction, knowledge representation, and enhanced modeling. The enhanced knowledge graph is a structured fusion knowledge graph that has been processed by intelligent agents and enhanced for retrieval, relationships, and reasoning.

[0070] In operation S220, the agent is input based on the enhanced knowledge graph to reversely adjust the agent parameters to obtain an adjusted agent.

[0071] When using the enhanced knowledge graph as an agent input for reverse tuning, the entities and relationships in the graph can be first encoded as vectors and concatenated with the text. Through methods such as prefix fine-tuning or LORA (introducing low-rank matrices at the model level), the main model parameters are fixed, and only a small number of newly added parameters are optimized. Leveraging the enhanced knowledge graph, the agent learns to integrate structured knowledge with language sequences, improving factuality and reasoning accuracy, reducing hallucinations, and enhancing expertise in specific fields (such as healthcare and finance).

[0072] The enhanced knowledge graph constructed can not only serve as an important resource for fine-tuning the intelligent agent and promote its application effect in specific fields, but also provide strong support for the first recommendation model. By integrating various types of information sources, it helps to comprehensively improve the user experience and achieve more personalized and accurate recommendation services.

[0073] In operation S230 , multidimensional data is retrieved based on the enhanced knowledge graph, and an initial recommendation result is obtained based on the retrieved data using a pre-trained second recommendation model.

[0074] Multidimensional data is deeply retrieved in the enhanced knowledge graph, and the retrieved data is then input into a pre-trained second recommendation model, which outputs initial recommendation results.

[0075] By leveraging the enhanced entity relationships of the knowledge graph, we traverse user nodes and their associated attribute nodes (such as age and consumption history) and relationship nodes (such as purchases and followers). Leveraging the efficient query language of graph databases, combined with keyword matching, path search, or semantic reasoning, we retrieve data related to user queries, including entities, attributes, and relationships. We quickly locate target user nodes and associated information, integrating multi-dimensional data to form a complete user profile, providing precise data support for recommendations.

[0076] In operation S240, the adjusted agent is integrated with the initial recommendation result to generate a preferred recommendation result.

[0077] The adjusted intelligent agent is combined with the initial recommendation results to generate preferred recommendation results, which include personalized recommendation copy, recommended product information, related product lists, etc.

[0078] According to an embodiment of the present application, after generating the preferred recommendation result in operation S240, it also includes reversely adjusting the enhanced knowledge graph based on the preferred recommendation result.

[0079] As part of the feedback mechanism in the first recommendation model, based on the preferred recommendation results, the knowledge graph is reversely enhanced and strengthened, and the preferred recommendation results are passed to the enhanced knowledge graph again to form a feedback closed-loop model to enhance the performance of the first recommendation model. In this way, a closed-loop recommendation system is established, so that each recommendation result can become the basis for the next learning, and the recommendation quality is continuously iterated and optimized.

[0080] Figure 5 The figure schematically shows a flow chart of enhanced knowledge graph of the intelligent recommendation method according to an embodiment of the present application.

[0081] In operation S210, multi-dimensional data is enhanced based on a pre-trained intelligent agent to obtain an enhanced knowledge graph, which includes operations S510 to S540.

[0082] In operation S510, retrieval enhancement is performed on multidimensional data based on an agent to obtain retrieval-enhanced multidimensional data.

[0083] Retrieval enhancement involves expanding multidimensional data to enhance the retrieval capabilities of the knowledge graph. Agents can be used to perform data infill and expansion on portions of the multidimensional data, such as using large models to generate similar contextual content and extend semantics. This results in retrieval-enhanced multidimensional data, creating more complete and rich data and improving the accuracy of subsequent analysis.

[0084] In operation S520, relationship enhancement is performed on the first multidimensional data based on the agent to obtain relationship-enhanced multidimensional data, wherein the first multidimensional data includes multidimensional data and / or retrieval-enhanced multidimensional data.

[0085] Relationship enhancement is to improve the knowledge nodes in the extracted data. It can extract and improve entities based on the intelligent agent to extract useful knowledge nodes from multidimensional data and / or retrieval-enhanced multidimensional data. For example, it can use the powerful semantic understanding ability of the large model to identify candidate knowledge nodes such as entities, relationships and events, extract entities and improve entities from candidate knowledge nodes, and extract key information from unstructured text to obtain relationship-enhanced multidimensional data.

[0086] In operation S530, based on the agent, the second multidimensional data is subjected to reasoning enhancement to obtain reasoning enhanced multidimensional data, wherein the second multidimensional data includes one or more of multidimensional data, retrieval enhanced multidimensional data, and relationship enhanced multidimensional data.

[0087] Reasoning enhancement is to optimize the representation of knowledge nodes. It can perform relationship vectorization on multidimensional data and / or retrieval-enhanced multidimensional data and / or relationship-enhanced multidimensional data based on intelligent agents to obtain reasoning-enhanced multidimensional data and achieve a clearer representation of entities. For example, knowledge nodes and embedded texts can be vectorized based on a large model to obtain vectorized representations between entities and relationships. The large model can be used as a text and graph structure encoder embedded in the knowledge graph to solve the problem of limited structural connectivity and improve the ability of knowledge representation.

[0088] Inference algorithms can be used to predict implicit relationships in knowledge graphs and build enhanced knowledge graphs. For example, deep relationship analysis can be performed on unstructured data such as user behavior logs in situational learning based on large models combined with attention mechanisms to further enhance graph capabilities.

[0089] In contextual learning, using intelligent agents to deeply analyze unstructured data can be abstracted as follows:

[0090] (4)

[0091] in, Representative User The scenario feature vector, represents the vectorized model, It is the user's historical behavior data. Refers to contextual information, is a model parameter. Based on each user's contextual feature vector and the specific environment they are currently in, a personalized recommendation list is generated. This can be achieved using the following formula:

[0092] (5)

[0093] Among them, r(.) represents the recommendation model, Represents a set of recommended items. is the set of all possible recommended items, Refers to the weight parameter in the recommendation algorithm.

[0094] In operation S540, an enhanced knowledge graph is constructed based on the third multidimensional data, wherein the third multidimensional data includes one or more of multidimensional data, retrieval-enhanced multidimensional data, relationship-enhanced multidimensional data, and reasoning-enhanced multidimensional data.

[0095] With entities as nodes and relationships as edges, we build a graph structure based on multidimensional data, retrieval-enhanced multidimensional data, relationship-enhanced multidimensional data, and reasoning-enhanced multidimensional data, and merge the same entity from different sources to form an enhanced knowledge graph.

[0096] In the embodiments of this application, the graph capabilities are enhanced in three ways based on agents, resulting in a structured and integrated knowledge graph that is enhanced for retrieval, relationships, and reasoning. This enhanced knowledge graph, based on the rich entity relationships in the graph, supplements sparse data and integrates agents to achieve a more comprehensive and detailed data representation. This improves the richness and completeness of the data while providing deeper information support for the recommendation model.

[0097] Figure 6 The following schematically shows a relationship enhancement flow chart of the intelligent recommendation method according to an embodiment of the present application.

[0098] In operation S520, relationship enhancement is performed on the first multidimensional data based on the agent to obtain relationship-enhanced multidimensional data, which includes operations S610 to S620.

[0099] The improved extraction process is aimed at the text content in the first multidimensional data , each Represents a word or character. The entity extraction task based on the intelligent agent is to find all possible entity sequences for the text content. , where each entity , Indicates the entity's starting position, Indicates length, Represents an entity type.

[0100] In operation S610, knowledge nodes of first multi-dimensional data are extracted based on an agent.

[0101] According to an embodiment of the present application, operation S610 of improving and extracting knowledge nodes of the first multi-dimensional data based on an intelligent agent includes operations S11 to S14.

[0102] In operation S11 , entity extraction is performed on the first multi-dimensional data to obtain entities.

[0103] According to the text content in the first multidimensional data , identify all possible entities and label them.

[0104] In operation S12, an entity sequence is output based on a preset knowledge node tuple.

[0105] List all found entity sequences E in a preset format, each entity e is a knowledge node tuple

[0106] In operation S13, the first multi-dimensional data is rewritten.

[0107] Rewrite text content , create new text content. For example, the original text content is: The thin and light notebook is equipped with an X-type processor, which is suitable for students to work and study. The rewritten text content is: The thin and light notebook is equipped with an X-type chip, which is specially designed for students and office workers to create an efficient learning and office experience.

[0108] In operation S14, entity extraction is performed on the rewritten first multi-dimensional data to obtain rewritten entities.

[0109] Find the entities in the rewritten text content and annotate them again.

[0110] In operation S15 , the entity sequence is adjusted according to the rewritten entity to output the extracted knowledge node.

[0111] Based on the rewritten entity, the entity sequence generated by the entity before the rewriting is adjusted, and the adjusted entity sequence and the rewritten entity are used as the extracted knowledge nodes.

[0112] For example, to perform improved extraction based on a large model, the following prompt instruction example can be used:

[0113] In this task, we need to deal with an entity annotation problem and expand the annotated entity based on the annotation results. Each entity definition has a starting position (b), a length ( ) and type (t), where entity type can be user id, product id, date and time, number, etc. The specific steps are as follows: (1) Entity extraction: First, identify all possible entities based on the text content T. Provide a brief description of why this type of annotation is chosen. (2) Standard output: List all found entity sequences E in JSON format. Each entity e should be a triple (3) Try to rewrite T and create new sentences while keeping the meaning of the original entities unchanged, and re-label the entities in the new sentences. (4) Implementation and verification: Implement the above content and ensure that the results generated each time can be automatically adjusted according to the different input data and can correctly affect the final output, output entities and entity sequences.

[0114] In the embodiments of the present application, by rewriting data to extract new entities and adjusting the original entity sequence to obtain knowledge nodes, information can be supplemented, errors and omissions can be corrected, and knowledge nodes can be made more accurate, complete and orderly.

[0115] In operation S620, based on the agent, the extracted knowledge nodes are vectorized to obtain relationship-enhanced multidimensional data.

[0116] Vectorized representation provides a clearer representation of entities. For example, a large model can be used as an encoder for text and graph structures embedded in a knowledge graph to vectorize the extracted knowledge nodes.

[0117] In an embodiment of the present application, based on intelligent agents, the extraction and characterization of extracted knowledge nodes are improved and optimized, and useful knowledge nodes are extracted by intelligent agents to improve entity extraction and characterization, enhance the relationship between knowledge nodes, improve data understanding, and improve the integrity of knowledge topology structure and the accuracy of semantic association.

[0118] In operation S620 , based on the agent, the extracted knowledge nodes are vectorized and represented to obtain relationship-enhanced multidimensional data, which includes operations S21 to S24 .

[0119] In operation S21 , a graph structure element vector is obtained based on the first multi-dimensional data.

[0120] Figure 7 The vectorized representation structure diagram of the intelligent recommendation method according to the embodiment of the present application is schematically shown as follows: Figure 7 As shown, when the first multidimensional data is multidimensional data, which is not expanded data, the normal path in the middle of the optimization representation graph is realized, and the relationship vector is extracted under this path; when the first multidimensional data is retrieval-enhanced multidimensional data, which is expanded multidimensional data, the retrieval-enhanced path on the left or right side of the optimization representation graph is realized, and the entity vector is extracted under this path. Based on the first multidimensional data, a knowledge graph can be established through graph neural network (such as GNN) processing to capture the structural information of the graph. The vectors of entities and relationships obtained from the graph structure of knowledge nodes based on different paths are graph structure element vectors, and the graph structure element vectors corresponding to the entities of the retrieval-enhanced path are The graph structure element vector corresponding to the relationship with the normal path , since the left and right paths have the same processing process, It only represents the graph structure element vectors corresponding to different entities. The processing methods are the same. To explain, I won’t go into too much detail.

[0121] In operation S22, text encoding is performed on the extracted knowledge nodes based on the agent to obtain text element vectors.

[0122] Using the agent as a text encoder to implement text encoding, entity sequence R .in, , in and It is an entity. is connected and Based on the text encoder ( ) to obtain the vector representation of the text description of entities and relations, specifically the following formula:

[0123] (6)

[0124] (7)

[0125] in, and are the text element vectors of entities and relations respectively. Similarly A vector of text elements for the entity.

[0126] In operation S23 , the graph structure element vector and the text element vector are fused to obtain node features.

[0127] To integrate text and graph structure information, we can use the attention mechanism to weightedly integrate text information and graph structure information to obtain node features. Specifically, the following formula is used:

[0128] (8)

[0129] (9)

[0130] in, and are the node features of entities and relationships respectively, and is a parameter that controls the contribution ratio of text and graph structure. It is the node feature of the entity.

[0131] In operation S24 , node features of neighboring nodes are aggregated based on the extracted knowledge nodes to obtain a node feature matrix, thereby obtaining relationship-enhanced multidimensional data corresponding to the extracted knowledge nodes.

[0132] For each knowledge node (entity), its final representation can be obtained by aggregating the information of neighboring nodes, specifically as follows:

[0133] (10)

[0134] in, It is The node feature matrix of the layer, It is The node feature matrix of the layer, is the normalized adjacency matrix plus self-loops, is the weight matrix, is the activation function.

[0135] In an embodiment of the present application, graph structure element vectors and text element vectors are integrated, and through intelligent agents, entities are more clearly represented based on graph structure and text, solving the problem of limited structural connectivity and improving the ability of knowledge representation.

[0136] In the embodiment of the present application, a mutually beneficial dual-driven recommendation scheme is implemented based on the first recommendation model, which also has the following advantages:

[0137] Enhanced data representation capabilities: We supplement sparse data with rich entity relationships in the graph and integrate intelligent agents to achieve more comprehensive and detailed data representation. This improves the richness and completeness of data while providing deeper information support for recommendation models.

[0138] Improve recommendation diversity: Based on the graph association network, it effectively discovers indirect potential product groups. At the same time, it combines with intelligent agents to generate diverse content descriptions. The two-way drive can significantly increase the freshness and exploratory nature of recommendation results, thereby breaking the "information cocoon" effect and improving user experience.

[0139] Deepening Understanding: The collaboration between the graph and the agent enables multi-dimensional analysis of user interests and deeply explores the subtle connections between product features. This multi-layered understanding helps provide more precisely matched recommendations to meet users' personalized needs.

[0140] Improved system transparency: By clearly demonstrating the logical chain behind recommendations—that is, the key factors that trigger a specific recommendation—the entire process is more explainable. This not only helps build user trust but also supports refined operational adjustments, enabling business decision makers to better understand and optimize recommendation strategies.

[0141] This mutually beneficial dual-driven recommendation framework uses graphs to provide precise entity relationships and domain knowledge, while agents provide general knowledge and content generation capabilities. While ensuring data verifiability and traceability, it enables more complex multi-step reasoning and contextual reasoning tasks, providing more intelligent and personalized recommendations and meeting the specific needs of different industries and fields.

[0142] Based on the above intelligent recommendation method, this application also provides an intelligent recommendation device. Figure 8 The device is described in detail.

[0143] Figure 8 The following schematically shows a structural block diagram of an intelligent recommendation device according to an embodiment of the present application.

[0144] like Figure 8 As shown, the intelligent recommendation device 800 of this embodiment includes a recommendation module 810;

[0145] The recommendation module 810 includes: a graph enhancement submodule 8101, a feedback fine-tuning submodule 8102, an initial recommendation submodule 8103 and a result generation submodule 8104.

[0146] The recommendation module 810 is used to process the input multidimensional data using a pre-trained first recommendation model to obtain a preferred recommendation result. In one embodiment, the recommendation module 810 can be used to execute the method described in the above intelligent recommendation method to process the input multidimensional data using a pre-trained first recommendation model to obtain a preferred recommendation result, which will not be repeated here.

[0147] The graph enhancement submodule 8101 is used to enhance the multidimensional data based on the pre-trained agent to obtain an enhanced knowledge graph. In one embodiment, the graph enhancement submodule 8101 can be used to perform the operation S210 described above, which will not be repeated here.

[0148] The feedback fine-tuning submodule 8102 is used to input the agent based on the enhanced knowledge graph to reversely adjust the agent parameters to obtain the adjusted agent. In one embodiment, the feedback fine-tuning submodule 8102 can be used to perform the operation S220 described above, which will not be repeated here.

[0149] The initial recommendation submodule 8103 is used to search the multidimensional data based on the enhanced knowledge graph and obtain initial recommendation results based on the searched data using the pre-trained second recommendation model. In one embodiment, the initial recommendation submodule 8103 can be used to perform the operation S230 described above, which will not be repeated here.

[0150] The result generation submodule 8104 is used to integrate the adjusted agent with the initial recommendation result to generate a preferred recommendation result. In one embodiment, the result generation submodule 8104 can be used to perform the operation S240 described above, which will not be repeated here.

[0151] According to an embodiment of the present application, any multiple modules among the graph enhancement submodule 8101, the feedback fine-tuning submodule 8102, the initial recommendation submodule 8103, and the result generation submodule 8104 can be combined into a single module, or any one of them can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to an embodiment of the present application, at least one of the graph enhancement submodule 8101, the feedback fine-tuning submodule 8102, the initial recommendation submodule 8103, and the result generation submodule 8104 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware by any other reasonable means of integrating or packaging circuits, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the graph enhancement submodule 8101, the feedback fine-tuning submodule 8102, the initial recommendation submodule 8103 and the result generation submodule 8104 can be at least partially implemented as a computer program module, which can perform the corresponding function when it is executed.

[0152] Figure 9 A block diagram of an electronic device suitable for implementing the intelligent recommendation method according to an embodiment of the present application is schematically shown.

[0153] like Figure 9 As shown, an electronic device 900 according to an embodiment of the present application includes a processor 901, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 902 or programs loaded from a storage unit 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or related chipsets and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present application.

[0154] Various programs and data required for the operation of the electronic device 900 are stored in the RAM 903. The processor 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. The processor 901 performs various operations of the method flow according to the embodiment of the present application by executing the programs in the ROM 902 and / or the RAM 903. It should be noted that the programs may also be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 may also perform various operations of the method flow according to the embodiment of the present application by executing the programs stored in one or more memories.

[0155] According to an embodiment of the present application, electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to bus 904. Electronic device 900 may also include one or more of the following components connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 908 including a hard disk; and a communication section 909 including a network interface card such as a LAN card or modem. Communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. Removable media 911, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 910 as needed, so that computer programs read from the removable media can be installed into storage section 908 as needed.

[0156] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the method according to the embodiments of this application is implemented.

[0157] According to an embodiment of the present application, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present application, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0158] The embodiments of the present application also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to enable the computer system to implement the intelligent recommendation method provided in the embodiments of the present application.

[0159] The computer program executes the above functions defined in the system / device of the embodiment of the present application when the processor 901 executes the computer program. According to the embodiment of the present application, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0160] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0161] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from a removable medium 911. When the computer program is executed by the processor 901, the above-mentioned functions defined in the system of the embodiment of the present application are performed. According to the embodiment of the present application, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0162] According to an embodiment of the present application, the program code for executing the computer program provided by the embodiment of the present application can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0163] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of the boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0164] Those skilled in the art will appreciate that the features described in the various embodiments of this application may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in this application. In particular, the features described in the various embodiments of this application may be combined and / or coupled in various ways without departing from the spirit and teachings of this application. All such combinations and / or couplings fall within the scope of this application.

Claims

1. An intelligent recommendation method, comprising: Using a pre-trained first recommendation model to process the input multi-dimensional data to obtain a preferred recommendation result; The method of processing the input multi-dimensional data using the pre-trained first recommendation model to obtain a preferred recommendation result includes: Strengthening the multidimensional data based on a pre-trained intelligent agent to obtain an enhanced knowledge graph; Inputting the agent based on the enhanced knowledge graph to reversely adjust the agent parameters to obtain an adjusted agent; Retrieving the multidimensional data based on the enhanced knowledge graph, and obtaining initial recommendation results using a pre-trained second recommendation model based on the retrieved data; The adjusted intelligent agent is integrated with the initial recommendation result to generate the preferred recommendation result.

2. The method according to claim 1, wherein the enhancing the multidimensional data based on the pre-trained agent to obtain an enhanced knowledge graph comprises: Based on the intelligent agent, performing retrieval enhancement on the multidimensional data to obtain retrieval enhanced multidimensional data; Based on the agent, performing relationship enhancement on first multidimensional data to obtain relationship-enhanced multidimensional data, wherein the first multidimensional data includes the multidimensional data and / or the retrieval-enhanced multidimensional data; Based on the agent, performing reasoning enhancement on the second multidimensional data to obtain reasoning-enhanced multidimensional data, wherein the second multidimensional data includes one or more of the multidimensional data, the retrieval-enhanced multidimensional data, and the relationship-enhanced multidimensional data; and Based on the third multidimensional data, the enhanced knowledge graph is constructed, wherein the third multidimensional data includes one or more of the multidimensional data, the retrieval-enhanced multidimensional data, the relationship-enhanced multidimensional data, and the reasoning-enhanced multidimensional data.

3. The method according to claim 2, wherein the step of performing relationship enhancement on the first multidimensional data based on the agent to obtain relationship-enhanced multidimensional data comprises: Based on the intelligent agent, improving and extracting knowledge nodes of the first multidimensional data; as well as Based on the intelligent agent, the extracted knowledge nodes are vectorized to obtain the relationship-enhanced multidimensional data.

4. The method according to claim 3, wherein the step of improving and extracting knowledge nodes of the first multidimensional data based on the agent comprises: Performing entity extraction on the first multidimensional data to obtain entities; Output entity sequence based on preset knowledge node tuple; rewriting the first multidimensional data; Performing entity extraction on the rewritten first multidimensional data to obtain rewritten entities; as well as According to the rewritten entity, the entity sequence is adjusted to output the extracted knowledge node.

5. The method according to claim 3, wherein the vectorizing representation of the extracted knowledge nodes based on the agent to obtain the relationship-enhanced multidimensional data comprises: Based on the first multidimensional data, obtaining a graph structure element vector; Based on the intelligent agent, text encoding is performed on the extracted knowledge nodes to obtain text element vectors; fusing the graph structure element vector and the text element vector to obtain node features; as well as Based on the extracted knowledge nodes, node features of neighboring nodes are aggregated to obtain a node feature matrix, thereby obtaining relationship-enhanced multidimensional data corresponding to the extracted knowledge nodes.

6. The method according to claim 1, further comprising: During the training of the first recommendation model, a cross entropy loss corresponding to the enhanced knowledge graph is constructed, and a ranking loss corresponding to the second recommendation model is constructed; Determining a joint loss of the first recommendation model based on the cross entropy loss and the ranking loss; and Minimize the joint loss.

7. The method according to claim 1, further comprising: Based on the preferred recommendation result, the enhanced knowledge graph is adjusted in reverse.

8. An intelligent recommendation device, comprising: A recommendation module, configured to process the input multidimensional data using a pre-trained first recommendation model to obtain a preferred recommendation result; The recommendation module includes: A graph enhancement submodule, configured to enhance the multidimensional data based on a pre-trained agent to obtain an enhanced knowledge graph; A feedback fine-tuning submodule, configured to input the agent based on the enhanced knowledge graph to reversely adjust the agent parameters to obtain an adjusted agent; an initial recommendation submodule, configured to retrieve the multidimensional data based on the enhanced knowledge graph, and obtain initial recommendation results based on the retrieved data using a pre-trained second recommendation model; and The result generation submodule is used to integrate the adjusted intelligent agent with the initial recommendation result to generate the preferred recommendation result.

9. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, The one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images