Library literature intelligent recommendation system generated based on knowledge graph model

Abstract

The invention belongs to the technical field of text generation, and particularly relates to a library literature intelligent recommendation system based on knowledge graph model generation, which comprises a model construction module, a recommendation module, an output module and a user interaction module, according to the library literature intelligent recommendation system generated based on the knowledge graph model, the model construction module is set to construct the knowledge graph model, and the knowledge graph model deeply understands complex semantic relationships among entities such as concepts, topics and authors in library literatures through an entity-relationship-entity structure instead of simple matching based on words, so that the knowledge graph model has the advantages that the knowledge graph model is simple in structure and easy to implement, and the library literature recommendation efficiency is improved. In addition, the knowledge graph model can mine abundant association relationships among the literatures, such as reference relationships, cooperation relationships and theme hierarchical relationships, and the knowledge graph model shows that the knowledge graph can find that an author of one literature and a manuscript reviewer of another literature are the same potential association, so that related literatures are recommended for the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of text generation, and specifically to an intelligent library document recommendation system generated based on a knowledge graph model. Background Art

[0002] Manually searching for interesting documents in a vast library of documents often takes a lot of time and effort. However, intelligent recommendation systems can automatically filter and recommend relevant documents based on user preferences, significantly reducing user search costs.

[0003] Chinese invention patent publication number CN118152552A discloses a big data analysis-based intelligent document recommendation system and method. This system obtains a user profile (including basic information, reading history, and preferences) and document search input, and then uses data processing and semantic understanding algorithms based on big data and artificial intelligence technologies on the back end to perform semantic association analysis between the user profile and document search input. The system then uses this semantic association between the user profile and document search input to perform semantic matching with the semantics of candidate documents extracted from the document library. This is used to determine the matching degree of the candidate documents and, based on the matching results, determine whether to recommend the candidate documents. The key technical solution is the ability to automatically filter out documents that meet the user's needs from a vast amount of document resources based on the user's basic information, reading history, preferences, and other information, and then present these documents to the user, enabling intelligent document recommendation and improving user reading efficiency and satisfaction.

[0004] While existing intelligent document recommendation systems based on big data analysis can make recommendations based on user profiles and semantic matching, they are less capable of mining these complex relationships. Furthermore, these systems often struggle to provide clear and intuitive explanations of recommendations. In other words, users may only know that the system has matched recommendations based on their historical behavior and document characteristics, but they may not understand the specific semantic associations or document relationships that led to the recommendations.

[0005] To this end, the present invention provides a library document intelligent recommendation system generated based on a knowledge graph model. Summary of the Invention

[0006] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0007] The technical solution adopted by the present invention to solve its technical problem is: the library document intelligent recommendation system based on knowledge graph model generation described in the present invention includes a model construction module, a recommendation module, an output module and a user interaction module; The model building module is used to organize the documents and related entities within the library in a graph structure, expressing the citation relationship between documents, the collaboration network of authors, and the hierarchical structure of topics to form a knowledge graph model; The recommendation module obtains recommended content based on semantic reasoning and association analysis performed by the knowledge graph model; The output module is used to output the recommended relevant documents in the library to the user and provide an explanation of the recommendation; The user interaction module is used to provide an interactive interface between the user and the system, verify the user's identity, record the user's borrowing record and browsing record in the library, and allow the user to evaluate and provide feedback on the recommendation results.

[0008] Furthermore, the construction process of the knowledge graph model includes data collection and preprocessing, entity and relationship extraction, knowledge fusion and graph storage, and the collection and preprocessing are reflected in the collection of data sources and data cleaning. The collected content includes the titles, abstracts and keywords of documents in library databases, academic databases and open data sources. At the same time, the missing metadata information is supplemented and improved by comparing with other data sources and extracting it from the full text of the document using natural language processing technology.

[0009] Furthermore, the entity extraction in the entity and relationship extraction is reflected in the use of NLP technology to extract entities from document titles and abstracts, and the relationship extraction is reflected in the use of rule matching and deep learning models to extract relationships from texts. At the same time, the performance of the extraction model is improved by combining manual annotation and deep learning model training.

[0010] Furthermore, the knowledge fusion and graph storage are manifested in aligning the same entities in different data sources and resolving data conflicts. Graph storage is manifested in using a graph database to store knowledge graphs, and at the same time, based on ontology alignment and entity disambiguation fusion technology, solving data conflicts and duplicate entity problems.

[0011] Furthermore, the processing flow of the ontology alignment technology is: ontology alignment, concept alignment, relationship alignment and result alignment; the processing flow of the entity disambiguation is: entity feature extraction, similarity calculation and formulation of disambiguation decisions; the processing flow of fusion is: merging entity data identified as duplicates, updating the relationship information related to the fused entities, and formulating conflict resolution strategies based on the results of the ontology alignment and the reliability factors of the data source.

[0012] Furthermore, the natural language processing technology also includes text mining and summary generation, wherein the text mining algorithm extracts keywords from the full text of the document through keyword extraction, and adds these keywords as the subject information of the document to the knowledge graph to help describe the content of the document.

[0013] Furthermore, the semantic reasoning and association analysis based on the knowledge graph model in the recommendation module adopts a hybrid recommendation method of path-based recommendation and embedding-based recommendation, that is, first find potential related documents through path analysis, and then recommend them through embedding similarity ranking. The formula is expressed as: in, is the recommendation score obtained by the path recommendation method, is the recommendation score obtained by the embedding-based recommendation method, The comprehensive score of mixed recommendation is By weighted summation, It is a weight parameter used to balance the influence of the two recommendation methods. The weight parameter in the hybrid recommendation formula is adjusted based on the user's historical behavior and interest preferences.

[0014] Furthermore, the recommendation module introduces a semantic reasoning model, namely a graph neural network, to deeply explore the semantic associations and potential relationships between documents and improve the accuracy of recommendations. At the same time, the principle of the graph neural network is that in the knowledge graph, documents, authors, and subject entities are regarded as nodes in the graph, and the citation relationships and author cooperation relationships between them are regarded as edges. The graph neural network updates the representation of the node by aggregating the information of the node neighbors, thereby capturing the complex semantic associations between documents and combining the knowledge graph. , the formula of graph neural network is expressed as: Aggregate neighbor information:

[0015] Update node representation:

[0016] in, For nodes, nodes , its neighbor node set is , is an aggregate function, For the update function For the Layer nodes .

[0017] Furthermore, the output module explains the reasons and basis for recommendations to users through examples and visual paths, introduces multiple evaluation indicators such as accuracy, recall rate, F1 value, and user satisfaction survey, monitors the recommendation effect in real time, and adjusts the recommendation algorithm and strategy based on the evaluation results.

[0018] Furthermore, the user interaction module monitors and classifies user evaluations and feedback in real time by establishing a rapid feedback processing mechanism, assigns issues to relevant personnel for processing in a timely manner, and feeds back the processing results to the user.

[0019] The beneficial effects of the present invention are as follows: 1. The knowledge graph model, through its entity-relationship-entity structure, deeply understands the complex semantic relationships between entities such as concepts, topics, and authors in library documents, rather than simply matching words. Furthermore, the knowledge graph model can mine rich associations between documents, such as citations, collaborations, and thematic hierarchies. This is reflected in the knowledge graph being able to identify potential connections, such as when the author of one document and the reviewer of another are the same person, thereby recommending relevant documents to users. Systems based on big data analysis cannot capture these indirect connections. 2. The knowledge graph model can use a graph structure to display the specific association path between recommended documents and user interests, such as through citation relationships, topic similarities, etc., to provide users with more explainable recommendation reasons and enhance users' trust in the recommendation results. At the same time, the knowledge graph model has a certain reasoning ability and can derive new knowledge or associations based on existing entities and relationships. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below with reference to the accompanying drawings.

[0021] Figure 1 This is the overall structure diagram of the library document intelligent recommendation system generated based on the knowledge graph model of the present invention; Figure 2 This is a structural diagram of the model building module of the library document intelligent recommendation system generated based on the knowledge graph model of the present invention; Figure 3 It is a structural diagram of the recommendation module of the library document intelligent recommendation system generated based on the knowledge graph model of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments 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 are within the scope of protection of the present invention.

[0023] See also Figure 1-3 , Example 1: This embodiment provides: a library document intelligent recommendation system generated based on a knowledge graph model, including a model building module, a recommendation module, an output module, and a user interaction module; The model building module is used to organize the documents and related entities within the library in a graph structure, expressing the citation relationship between documents, the author's collaboration network, and the hierarchical structure of topics to form a knowledge graph model. For example, it can not only obtain the borrowing data within the library, but also combine the document citation relationship in the academic database to build a more comprehensive knowledge graph.

[0024] The knowledge graph model consists of entities, relationships, and attributes: Entity: includes documents, authors, topics, institutions, and keywords; for example, a paper is an entity, and its authors, publication journals, and keywords are also entities.

[0025] Relationship: describes the association between entities, such as "Paper A cites Paper B", "Author C wrote Paper D", "Paper E belongs to Topic F".

[0026] Attributes: describe the characteristics of an entity, such as the publication year of a paper, the author's affiliation, and the classification label of a topic; The construction process of the knowledge graph model includes data collection and preprocessing, entity and relationship extraction, knowledge fusion and graph storage. The collection and preprocessing are reflected in the collection of data sources and data cleaning. The collected content includes the titles, abstracts and keywords of documents in library databases, academic databases and open data sources. At the same time, the missing metadata information is supplemented and improved by comparing with other data sources and extracting it from the full text of the document using natural language processing technology.

[0027] The extraction of entities in entity and relationship extraction is reflected in the use of NLP technology to extract entities from document titles and abstracts. For example, suppose there is a text: "The paper "Happy Learning" was written by Zhang San from the University of Technology." Named entity recognition process based on deep learning (with For example) its entity extraction process: Input layer: Convert each word in the text into a corresponding word vector representation; assuming the text length is n and the word vector dimension is d, the input can be represented as a matrix ; layer: Encode the input word vector to generate context-related word representation. The output dimension is h, then the new representation matrix is obtained ; Layer: Bidirectional right The output of the sequence model is used to capture context information. The output can be expressed as ; CRF layer: The CRF layer performs label prediction based on the output of BiLSTM. The label set can be defined as The above respectively represent the beginning of the paper title, the inside of the paper title, the beginning of a person's name, the inside of a person's name, the beginning of an organization name, the inside of an organization name, and others; By calculating the probabilities of all possible label sequences, the sequence with the highest probability is selected as the final entity label.

[0028] Entity extraction results: Through the above process, we can identify "Happy Learning" as the paper title entity, "Zhang San" as the person name entity, and "Technical University" as the organization name entity; Relationship extraction is reflected in the use of rule matching and deep learning models to extract relationships from text. At the same time, the performance of the extraction model is improved by combining manual annotation and deep learning model training. At the same time, relationship extraction is to identify the relationship between entities from the text, such as "reference", "cooperation", "belonging", etc.

[0029] For example, if you need to extract relations from the sentence "Paper A cites Paper B." The relationship extraction process is: Input representation: Convert each word in the sentence into a word vector, and encode the entity position information (such as the position of the entity in the sentence) into a vector, and concatenate it with the word vector. Let the word vector dimension be , the position vector dimension is , then the input is represented as ,in is the sentence length; Layer: Bidirectional Encode the input to capture contextual information, The output is ,in is the hidden layer dimension.

[0030] Pooling layer (optional): The output of is pooled (such as maximum pooling, average pooling) to obtain a fixed length representation of the sentence .

[0031] Classification layer: The pooled representation is input into a fully connected layer, and then Function to classify relations, let the relation category set be , then the classification layer output is ,in , , is the number of relationship categories.

[0032] Relationship extraction results: Through the above process, it can be determined that the relationship between "Paper A" and "Paper B" is "cite", that is, citation.

[0033] In knowledge fusion and graph storage, knowledge fusion is reflected in aligning the same entities in different data sources and resolving data conflicts. Graph storage is reflected in using graph databases to store knowledge graphs, and at the same time, based on ontology alignment and entity disambiguation fusion technology, it resolves data conflicts and duplicate entity problems.

[0034] The processing flow of ontology alignment technology is: ontology alignment, concept alignment, relationship alignment and result alignment; the processing flow of entity disambiguation is: entity feature extraction, similarity calculation and formulation of disambiguation decisions; the processing flow of fusion is: merging entity data identified as duplicates, updating the relationship information related to the fused entities, and formulating conflict resolution strategies based on the results of ontology alignment and the reliability factors of the data source.

[0035] For example, there may be inconsistent expressions of the same author name in different data sources. Entity disambiguation can accurately identify and merge them, making the knowledge graph more accurate.

[0036] Natural language processing technology also includes text mining and summary generation. The text mining algorithm extracts keywords from the full text of the document through keyword extraction, and adds these keywords as the subject information of the document to the knowledge graph to help describe the content of the document.

[0037] It should be noted that the knowledge graph model can deeply understand the complex semantic relationships between concepts, topics, authors and other entities in the document through the entity-relationship-entity structure, rather than just simple word matching. For example, the knowledge graph can understand the specific application scenarios and related technologies of the topic of "deep learning" in different documents.

[0038] The recommendation module obtains recommended content based on semantic reasoning and association analysis based on the knowledge graph model; The semantic reasoning and association analysis based on the knowledge graph model in the recommendation module adopts a hybrid recommendation method of path-based recommendation and embedding-based recommendation. That is, first find potential related documents through path analysis, and then recommend them by embedding similarity ranking. Its formula is expressed as:

[0039] in, is the recommendation score obtained by the path recommendation method, is the recommendation score obtained by the embedding-based recommendation method, The comprehensive score of mixed recommendation is By weighted summation, It is a weight parameter used to balance the influence of the two recommendation methods. The weight parameter in the hybrid recommendation formula is adjusted based on the user's historical behavior and interest preferences.

[0040] The recommendation module incorporates a semantic reasoning model, namely a graph neural network, to deeply explore the semantic connections and potential relationships between documents, improving recommendation accuracy. The principle of a graph neural network is as follows: In a knowledge graph, documents, authors, and topic entities are considered nodes in the graph, and citation relationships and author collaborations between them are considered edges. The graph neural network updates node representations by aggregating information about node neighbors, thereby capturing the complex semantic connections between documents. For example, suppose there is an academic knowledge graph containing multiple documents on "natural language processing." Document A focuses on "text classification," and Document B studies "named entity recognition." Both cite Document C (a review article on "basic technologies in natural language processing"). Traditional recommendation systems might simply recommend related documents based solely on citation relationships. However, with the introduction of a GNN, the GNN considers the neighboring information of document nodes. For example, if Document A's neighbors include not only Document C but also other documents related to "text classification," the GNN will aggregate the feature information of these neighboring nodes and update the original features of Document A. Similarly, the representation of Document B will be updated based on its neighboring nodes. In this way, when a user reads document A, the system can not only recommend directly related documents (such as the cited document C), but also recommend other documents with similar semantics to "text classification" based on the semantic representation learned by GNN, such as "Text classification algorithm optimization based on deep learning". Even if these documents have no direct citation relationship with document A, the semantic reasoning of GNN makes it easier to discover the potential correlation between them, thereby improving the accuracy of the recommendation.

[0041] Combined with knowledge graph , the formula of graph neural network is expressed as: Aggregate neighbor information:

[0042] Update node representation:

[0043] in, For nodes, nodes , its neighbor node set is , is an aggregate function, For the update function For the Layer nodes .

[0044] It is important to note that the knowledge graph model can use the rich semantic information and association relationships to recommend documents related to entities that new users may be interested in, or find existing documents related to their topics, authors, etc. for recommendation, thus alleviating the cold start problem to a certain extent. Furthermore, knowledge graphs can construct a rich knowledge network through various types of entities and relationships, more comprehensively representing knowledge in documents and related fields. Knowledge graph models possess a certain degree of reasoning ability, allowing them to derive new knowledge or associations based on existing entities and relationships. For example, if a knowledge graph contains the relationship "Paper A cites Paper B, and Paper B and Paper C are similar in topic," then it can be inferred that Paper A may also have a certain relevance to Paper C.

[0045] The output module is used to output the recommended relevant documents in the library to the user and provide the recommendation explanation; The output module explains the reasons and basis for recommendations to users through examples and visual paths, improving their understanding and acceptance of the recommendation results. It also introduces multiple evaluation metrics, such as precision, recall, F1 value, and user satisfaction surveys, to monitor the effectiveness of recommendations in real time and adjust the recommendation algorithm and strategy based on the evaluation results. It should be noted that the output module explains the reasons and basis for recommendations to users through examples and visual paths, improving their understanding and acceptance of the recommendations. For example, by showing users the citation relationship or thematic association path between the recommended document and the documents they have borrowed, users can understand why the recommendation was made.

[0046] The user interaction module is used to provide an interactive interface between the user and the system, verify the user's identity, record the user's borrowing and browsing records in the library, and allow the user to evaluate and provide feedback on the recommendation results.

[0047] The user interaction module monitors and classifies user evaluations and feedback in real time by establishing a rapid feedback processing mechanism, promptly assigns issues to relevant personnel for processing, and feeds back the processing results to users, thereby improving user satisfaction.

[0048] Example 2: The path-based recommendation principle is to discover potential connections between documents through paths in the knowledge graph. For example, if a user is interested in document X, the system finds that document X is cited by document Y, and the author Z of document Y has a cooperative relationship with the author of the document the user has read in the past, so document Y is recommended. The formula based on the path recommendation principle is described as follows: Assume that the knowledge graph is ,in is a collection of entities (including documents, authors), It is a set of relationships (such as reference relationships, cooperation relationships, etc.) for user interest entities and document entities If there is a path ,in , ,and , , then it is believed that the literature and user interests There is a potential correlation and a recommendation can be made.

[0049] For example: User's Interested, systematically discovered literature Literature Citations and literature Author Have a cooperative relationship with the author of the literature that the user has read in the past, so recommend literature .

[0050] The principle of embedding-based recommendation is to map entities and relationships in the knowledge graph into a low-dimensional vector space and calculate recommendations based on vector similarity, which can be described by formulating: Suppose the entity set in the knowledge graph is , the relationship set is , using embedding algorithms such as ) Mapping to low-dimensional vector ,relation Mapping to low-dimensional vector , for users , its interest vector is expressed as ,literature The vector representation of , by calculating the similarity (such as cosine similarity ,like Greater than a certain threshold , then recommended literature To users .

[0051] The above-mentioned front, back, left, right, up and down are all based on the Figure 1 As a benchmark, according to the person's observation perspective, the side of the device facing the observer is defined as the front, the left side of the observer is defined as the left, and so on.

[0052] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the scope of protection of the present invention.

[0053] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. The intelligent library document recommendation system generated based on the knowledge graph model is characterized by: Includes model building module, recommendation module, output module and user interaction module; The model building module is used to organize the documents and related entities within the library in a graph structure, expressing the citation relationship between documents, the collaboration network of authors, and the hierarchical structure of topics to form a knowledge graph model; The recommendation module obtains recommended content based on semantic reasoning and association analysis performed by the knowledge graph model; The output module is used to output the recommended relevant documents in the library to the user and provide an explanation of the recommendation; The user interaction module is used to provide an interactive interface between the user and the system, verify the user's identity, record the user's borrowing record and browsing record in the library, and allow the user to evaluate and provide feedback on the recommendation results.

2. The library document intelligent recommendation system based on the knowledge graph model according to claim 1 is characterized by: The construction process of the knowledge graph model includes data collection and preprocessing, entity and relationship extraction, knowledge fusion and graph storage, and the collection and preprocessing are reflected in the collection of data sources and data cleaning. The collected content includes the titles, abstracts and keywords of documents in library databases, academic databases and open data sources. At the same time, the missing metadata information is supplemented and improved by comparing with other data sources and extracting it from the full text of the document using natural language processing technology.

3. The library document intelligent recommendation system based on the knowledge graph model according to claim 2 is characterized by: In the entity and relationship extraction, entity extraction is reflected in the use of NLP technology to extract entities from document titles and abstracts, and relationship extraction is reflected in the use of rule matching and deep learning models to extract relationships from texts. At the same time, the performance of the extraction model is improved by combining manual annotation and deep learning model training.

4. The library document intelligent recommendation system based on knowledge graph model generation according to claim 3 is characterized by: In the knowledge fusion and graph storage, knowledge fusion is reflected in aligning the same entities in different data sources and resolving data conflicts, and graph storage is reflected in using a graph database to store knowledge graphs. At the same time, based on ontology alignment and entity disambiguation fusion technology, data conflicts and duplicate entity problems are resolved.

5. The library document intelligent recommendation system based on knowledge graph model generation according to claim 4 is characterized by: in, The processing flow of ontology alignment is: ontology alignment, concept alignment, relationship alignment and result alignment; the processing flow of entity disambiguation is: entity feature extraction, similarity calculation and formulation of disambiguation decisions; the processing flow of fusion is: merging entity data identified as duplicates, updating the relationship information related to the fused entities, and formulating conflict resolution strategies based on the results of ontology alignment and the reliability factors of the data source.

6. The library document intelligent recommendation system based on knowledge graph model generation according to claim 2 is characterized by: Natural language processing technology also includes text mining and summary generation. The text mining algorithm extracts keywords from the full text of the document through keyword extraction, and adds these keywords as the subject information of the document to the knowledge graph.

7. The library document intelligent recommendation system based on knowledge graph model generation according to claim 1 is characterized by: In the recommendation module, semantic reasoning and association analysis based on the knowledge graph model adopt a hybrid recommendation method of path-based recommendation and embedding-based recommendation. That is, potential related documents are first found through path analysis, and then recommendations are made through embedding similarity ranking. The formula is expressed as: in, The recommendation score obtained based on the path recommendation method, is the recommendation score obtained by the embedding-based recommendation method, The comprehensive score of mixed recommendation is By weighted summation, It is a weight parameter used to balance the influence of the two recommendation methods. The weight parameter in the hybrid recommendation formula is adjusted based on the user's historical behavior and interest preferences.

8. The library document intelligent recommendation system based on knowledge graph model generation according to claim 1 is characterized by: The recommendation module introduces a semantic reasoning model, namely a graph neural network, to deeply explore the semantic associations and potential relationships between documents. In the knowledge graph, documents, authors, and subject entities are regarded as nodes in the graph, and the reference relationships and author cooperation relationships between nodes are regarded as edges. The graph neural network updates the representation of nodes by aggregating the information of node neighbors, thereby capturing the complex semantic associations between documents. At the same time, combined with the knowledge graph , the formula of graph neural network is expressed as: Aggregate neighbor information: , Update node representation: in, For nodes, nodes , its neighbor node set is , is an aggregate function, For the update function For the Layer nodes .

9. The library document intelligent recommendation system based on knowledge graph model generation according to claim 1 is characterized by: The output module explains the reasons and basis for recommendations to users through examples and visual paths, introduces evaluation indicators, and adjusts recommendation algorithms and strategies based on evaluation results.

10. The library document intelligent recommendation system based on knowledge graph model generation according to claim 1 is characterized by: The user interaction module monitors and classifies the user's evaluation and feedback in real time by establishing a fast feedback processing mechanism, and feeds back the processing results to the user.

Images