Information recommendation method and system based on brain science knowledge extraction and graph network

Abstract

The invention belongs to the technical field of data processing, particularly relates to an information recommendation method and system based on brain science knowledge extraction and a graph network, and aims at solving the problems that an existing retrieval recommendation method is insufficient in operation mechanization, convenience and accuracy. The method comprises the following steps: extracting a plurality of keywords in text information and a text relationship among the keywords through a semantic analysis method based on brain science; constructing a knowledge graph based on the plurality of keywords and the text relationship among the keywords; constructing a knowledge relationship network based on the knowledge graph, and training through a graph data learning algorithm to obtain a graph neural network; the graph neural network comprises a representation vector of each keyword; and according to a retrieval instruction input by a user, determining the recommended content by calculating the semantic similarity between the retrieval instruction and the representation vector of the keyword. According to the method, the knowledge engine is automatically constructed from the text, and integration, recommendation and prediction of knowledge and automatic processing from the text information to the knowledge engine are realized.

Description

Background Art

[0002] The existing methods of extracting knowledge from scientific literature to build knowledge graphs and integrate and display knowledge are mainly to build a domain keyword table, extract knowledge from scientific literature using keywords, and then build a knowledge graph. In terms of displaying knowledge graphs, it mainly provides downloads of knowledge graph files or public database access ports, rather than providing a user interface that is convenient for non-knowledge graph professional users to access.

[0003] When users conduct searches, they can only mechanically enter search terms, and the accuracy of related content recommendations is insufficient. Summary of the invention

[0004] In order to solve the above problems in the prior art, namely, the problems that the existing retrieval recommendation methods are mechanized in operation, lack convenience and accuracy, the present invention provides an information recommendation method based on brain science knowledge extraction and graph network, including:

[0005] Extract multiple keywords and textual relationships between keywords from text information through semantic analysis methods based on brain science;

[0006] Constructing a knowledge graph based on the plurality of keywords and textual relationships between the keywords;

[0007] Based on the knowledge graph, a knowledge relationship network is constructed, and trained through a graph data learning algorithm to obtain a graph neural network; the graph neural network contains a representation vector of each keyword;

[0008] According to the search command input by the user, the recommended content is determined by calculating the semantic similarity between the search command and the representation vector of the keyword;

[0009] The recommended words and the representational relationships associated with the recommended words are visualized and displayed.

[0010] Further, the knowledge graph includes a first knowledge graph and a second knowledge graph;

[0011] The first knowledge graph is a knowledge graph of document-related information;

[0012] The second knowledge graph is a relationship-related knowledge graph.

[0013] Furthermore, keywords are used as nodes of the knowledge relationship network, and the relationships between keywords are used as edges of the knowledge relationship network.

[0014] Furthermore, the recommended content is determined by calculating the semantic similarity between the retrieval instruction and the representation vector of the keyword, specifically:

[0015] The backend system searches for the corresponding representation vector of the retrieval instruction in the graph neural network;

[0016] The semantic similarity between the corresponding representation vector and the representation vector of the keyword in the graph neural network is calculated, and the keyword corresponding to the representation vector with the highest semantic similarity and its corresponding related knowledge are selected as the recommended content.

[0017] Furthermore, if there are multiple corresponding characterization vectors at the same time, the average value is taken as the final corresponding characterization vector.

[0018] Furthermore, the graph neural network is visualized by using force-directed layout to highlight the representational relationship between the recommendation words and their associations.

[0019] Furthermore, it also includes performing angle transformation and / or position transformation on the visually displayed graph neural network according to the operation instructions.

[0020] Furthermore, it also includes visually displaying advanced recommended content and the relationship chain between the advanced recommended content and the recommended content after receiving the advanced search instruction.

[0021] Furthermore, the advanced search instruction includes selecting corresponding keywords from the graph neural network after angle transformation and / or position transformation.

[0022] Another aspect of the present invention provides an information recommendation system based on brain science knowledge extraction and graph network, the system comprising:

[0023] A keyword-relation extraction system configured to extract multiple keywords and textual relations between keywords in text information by using a semantic analysis method based on brain science;

[0024] A knowledge graph construction module, configured to construct a knowledge graph based on a plurality of the keywords and textual relations between the keywords;

[0025] A graph neural network construction module is configured to construct a knowledge relationship network based on the knowledge graph, and train it through a graph data learning algorithm to obtain a graph neural network; the graph neural network contains a representation vector of each keyword;

[0026] The recommended content selection module is configured to determine the recommended content by calculating the semantic similarity between the search instruction and the representation vector of the keyword according to the search instruction input by the user;

[0027] The visualization module is configured to visualize the recommendation words and the representational relationships associated with the recommendation words.

[0028] Beneficial effects of the present invention:

[0029] (1) The present invention realizes the automatic construction of a knowledge engine from text and realizes the integration, recommendation and prediction of knowledge;

[0030] (2) It realizes the automated processing from text information to interactive knowledge engine;

[0031] (3) The present invention addresses the problem that the latest information or technology is usually disclosed in the form of text, such as news, papers or comments, and that unstructured text is difficult to process automatically. The present invention automatically organizes time-sensitive text information into a knowledge engine and provides a convenient access interface for users, which can greatly improve the timeliness of knowledge services and expand the audience from a small number of knowledge engineering experts to all Internet users.

[0032] (4) In order to ensure the automation and high efficiency of the information extraction process, a simple relational extraction logic and a dictionary tree-based extraction method are adopted. In industries such as document information services and journalism, the efficiency and effectiveness of information collection, integration and prediction can be improved, and the manpower consumption can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0034] Figure 1 is a flowchart of an information recommendation method based on brain science knowledge extraction and graph network in the first embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of the effect of visual display in the first embodiment of the present invention;

[0036] Figure 3 It is a principle block diagram of a recommendation method combining simple retrieval with semantic retrieval in the second embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0038] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0039] The purpose of this solution is to develop a fully automatic knowledge graph construction process, and to achieve full process automation from knowledge collection, knowledge integration based on graph neural networks, knowledge recommendation to knowledge display. Solve the existing problems of insufficient knowledge timeliness and incomplete knowledge coverage caused by the inability to automatically construct knowledge graphs. At the same time, by building a user-friendly knowledge acquisition website interface, it solves the problem that users who are not knowledge graph professionals cannot easily obtain information in knowledge graphs and graph neural networks. This patent can realize automated knowledge graph construction and display, and provide users with real-time, comprehensive, and easy-to-access knowledge graph information services. At the same time, this solution allows users to find the knowledge they need in a variety of ways (semantic relevance, through text retrieval, through relationship retrieval), and continuously improve the search content, gradually approach the knowledge they need, and review the route of thinking.

[0040] In order to more clearly explain the information recommendation method based on brain science knowledge extraction and graph network of the present invention, Figure 1 Each step in the embodiment of the present invention is described in detail.

[0041] The information recommendation method based on brain science knowledge extraction and graph network of the first embodiment of the present invention includes steps S1 to S5, each of which is described in detail as follows:

[0042] Extract multiple keywords and textual relationships between keywords from text information through semantic analysis methods based on brain science;

[0043] In this embodiment, the semantic analysis method based on brain science can be performed by pre-building a knowledge extraction and graph network instruction engine, specifically, automatically extracting a domain keyword list or phrase list based on public information;

[0044] Arranging the keyword list and / or phrase list into a trie;

[0045] Keyword matching and knowledge extraction are performed based on the text information to be processed or recommended in the dictionary tree; more specifically, if a series of brain science literature needs to be processed, if two brain science concepts are extracted in the same sentence after keyword matching, it is determined that there is a relationship between the two brain science concepts. That is, the text relationship in a sentence represents its internal relationship, based on which the relationship between feature vectors can be constructed later.

[0046] Constructing a knowledge graph based on the plurality of keywords and textual relationships between the keywords;

[0047] Based on the knowledge graph, a knowledge relationship network is constructed, and trained through a graph data learning algorithm to obtain a graph neural network; the graph neural network contains a representation vector of each keyword;

[0048] The knowledge graph includes: concepts (keywords), relationships between concepts (keywords), concepts related to the relationships, sentences where the relationships are located, URLs of related documents, numbers of related documents, titles of related documents, and species related to the documents.

[0049] In this embodiment, the knowledge graph includes a first knowledge graph and a second knowledge graph;

[0050] The first knowledge graph is a knowledge graph of document-related information;

[0051] The knowledge graph related to the document includes the following information: document number, document publication date, document citation information and document-related species (if any).

[0052] The second knowledge graph is a relationship-related knowledge graph.

[0053] The relationship-related knowledge graph includes the following information: relationship-related documents, relationship-related keywords, and the time of relationship extraction.

[0054] Both the first knowledge graph and the second knowledge graph are products of keyword extraction from text information.

[0055] The relationship between the concepts of the first knowledge graph and the second knowledge graph is statistically represented by specific statistical variables, that is, a knowledge relationship network, including: the number of relationships between the two concepts, the names and URIs of the two concepts, and PAB and PBA.

[0056] In this embodiment, the knowledge relationship network constructed uses keywords as nodes of the knowledge relationship network and the relationships between keywords as edges of the knowledge relationship network. A highly summarized and simplified knowledge relationship network can be formed by summarizing the number of relationships between all keywords.

[0057] The obtained knowledge relationship network is trained through the node2vec algorithm, and then the representation vector of each keyword (node) is obtained.

[0058] According to the search command input by the user, the recommended content is determined by calculating the semantic similarity between the search command and the representation vector of the keyword;

[0059] In this embodiment, the method of determining the recommended content by calculating the semantic similarity between the retrieval instruction and the representation vector of the keyword is specifically as follows:

[0060] The backend system searches for the corresponding representation vector of the retrieval instruction in the graph neural network;

[0061] If there are multiple corresponding characterization vectors at the same time, the average value is taken as the final corresponding characterization vector;

[0062] When users send search instructions, they usually just select from pre-stored keywords, or extract the entered keywords based on the pre-built dictionary tree; after receiving the search instruction, the nodes and representation vectors corresponding to the keywords in the search instruction are found in the graph neural network, which are the corresponding representation vectors. The search instruction may contain multiple keywords, that is, there will be multiple corresponding representation vectors. At this time, the average value of the representation vector is selected as the corresponding representation vector, and the recommended content is selected by calculating the similarity between the corresponding representation vector after averaging and other representation vectors. Keywords in all libraries including non-current field keyword libraries can be matched in the search text submitted by the user, and keywords in non-current field keyword libraries can be searched.

[0063] The semantic similarity between the corresponding representation vector and the representation vector of the keyword in the graph neural network is calculated, and the keyword corresponding to the representation vector with the highest semantic similarity and its corresponding related knowledge are selected as the recommended content.

[0064] The specific operation is: the user issues a search command containing the keywords to be searched on the network interface through browsing and searching methods according to the content he wants to search. The website backend converts the keywords to be searched into representation vectors in the graph neural network through the search command, and the representation vector that is most similar to the representation vector is selected as the recommended content and recommended to the user.

[0065] The recommended words and the representational relationships associated with the recommended words are visualized and displayed.

[0066] The force-guided layout is used to visualize the graph neural network in a way that highlights the representation relationship between the recommendation words and the recommendation word associations. The force-guided layout can be implemented using methods such as GEM. The visualized graph neural network is shown in FIG. Figure 2 shown.

[0067] In the prior art, expert systems based on large language models can realize automatic integration of knowledge and construction of knowledge engines. However, knowledge engines based on pure large language models cannot track the source of knowledge, which is a disadvantage compared with this solution. Moreover, the computational cost of training graph neural networks in this solution is relatively low, while the cost of retraining large language models is relatively high, which makes this solution have the advantages of low cost and short update cycle. Knowledge engines based on RAG or AGENT technical routes can realize automatic integration of knowledge and construction of knowledge engines, and can also track the source of knowledge. However, if you want to have the same effect on knowledge recommendation and knowledge prediction as the functions based on graph neural networks in this solution, you also need to integrate the technical routes in this solution.

[0068] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art can understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0069] The second embodiment of the present invention is a method for performing advanced search through a graph neural network with visual display, specifically:

[0070] After obtaining the recommended content by the method of the first embodiment, the relationship related to the keywords in the recommended content and the keywords indirectly related to the keywords in the recommended content are displayed at the same time;

[0071] Users can adjust the perspective of the visually displayed graph neural network through the human-computer interaction interface, such as sliding the mouse, swiping across the human-computer interaction screen, VR, AR or MR, and can intuitively see the recommended content and its directly and indirectly related paths (relationships).

[0072] When users conduct searches, they often do not have accurate control over keywords, and there may be a gap between the searched content and the content they really want to search for, or they may not have accurate control over the keywords they want to search for. The search platforms of the prior art usually give words that are similar to the input content or only adjust individual characters. The present invention displays the recommended content and its related (not just similar) keywords (concepts) and relationships for users to choose from, thereby further improving the accuracy of users' acquisition of information.

[0073] Furthermore, in the prior art, only advanced keywords similar to the concept to be searched can be given when searching. When the user wants to obtain further information, he needs to further input keywords. The keywords input by the user again may not have a relationship with the previous keywords due to their parts of speech. Through the method of the present application, keywords of a whole chain or multiple relationship chains can be provided for users to choose from, allowing users to obtain the required information completely and coherently, and to view their own thinking paths, especially in the academic field, which can help users sort out and review their thoughts.

[0074] The third embodiment of the present invention is to combine the conventional retrieval method with the semantic similarity retrieval method, such as Figure 3 As shown, in addition to obtaining recommended content through semantic similarity, it is also possible to search directly through text or direct relationships; users can simultaneously obtain recommended content obtained through retrieval and recommended content obtained through semantic similarity as needed; further, the relationship path and concepts between the two recommended contents can be obtained, which facilitates the acquisition of more complete domain information.

[0075] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0076] The fourth embodiment of the present invention is an information recommendation system based on brain science knowledge extraction and graph network, comprising:

[0077] A keyword-relation extraction system configured to extract multiple keywords and textual relations between keywords in text information by using a semantic analysis method based on brain science;

[0078] A knowledge graph construction module, configured to construct a knowledge graph based on a plurality of the keywords and textual relations between the keywords;

[0079] A graph neural network construction module is configured to construct a knowledge relationship network based on the knowledge graph, and train it through a graph data learning algorithm to obtain a graph neural network; the graph neural network contains a representation vector of each keyword;

[0080] The recommended content selection module is configured to determine the recommended content by calculating the semantic similarity between the search instruction and the representation vector of the keyword according to the search instruction input by the user;

[0081] The visualization module is configured to visualize the recommendation words and the representational relationships associated with the recommendation words.

[0082] It should be noted that the information recommendation system based on brain science knowledge extraction and graph network provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not regarded as improper limitations of the present invention.

[0083] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the storage device and processing device described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0084] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0085] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.

[0086] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.

[0087] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An information recommendation method based on brain science knowledge extraction and graph network, characterized in that: include: Extract multiple keywords and textual relationships between keywords from text information through semantic analysis methods based on brain science; Constructing a knowledge graph based on the plurality of keywords and textual relationships between the keywords; Based on the knowledge graph, a knowledge relationship network is constructed, and trained through a graph data learning algorithm to obtain a graph neural network; the graph neural network contains a representation vector of each keyword; According to the search command input by the user, the recommended content is determined by calculating the semantic similarity between the search command and the representation vector of the keyword; The recommended words and the representational relationships associated with the recommended words are visualized and displayed.

2. The information recommendation method based on brain science knowledge extraction and graph network according to claim 1 is characterized in that: The knowledge graph includes a first knowledge graph and a second knowledge graph; The first knowledge graph is a knowledge graph of document-related information; The second knowledge graph is a relationship-related knowledge graph.

3. The information recommendation method based on brain science knowledge extraction and graph network according to claim 1 is characterized in that: Keywords are used as nodes of the knowledge relationship network, and the relationships between keywords are used as edges of the knowledge relationship network.

4. The information recommendation method based on brain science knowledge extraction and graph network according to claim 1 is characterized in that: The recommended content is determined by calculating the semantic similarity between the retrieval instruction and the representation vector of the keyword, specifically: The backend system searches for the corresponding representation vector of the retrieval instruction in the graph neural network; The semantic similarity between the corresponding representation vector and the representation vector of the keyword in the graph neural network is calculated, and the keyword corresponding to the representation vector with the highest semantic similarity and its corresponding related knowledge are selected as the recommended content.

5. The information recommendation method based on brain science knowledge extraction and graph network according to claim 4 is characterized in that: If there are multiple corresponding characterization vectors at the same time, the average value is taken as the final corresponding characterization vector.

6. The information recommendation method based on brain science knowledge extraction and graph network according to claim 1 is characterized in that: The graph neural network is visualized by using force-directed layout to highlight the representational relationship between the recommendation words and their associations.

7. The information recommendation method based on brain science knowledge extraction and graph network according to claim 1 is characterized in that: It also includes transforming the angle and / or position of the visually displayed graph neural network according to the operation instructions.

8. The information recommendation method based on brain science knowledge extraction and graph network according to claim 7 is characterized in that: It also includes visually displaying advanced recommended content and the relationship chain between the advanced recommended content and the recommended content after receiving advanced search instructions.

9. The information recommendation method based on brain science knowledge extraction and graph network according to claim 8 is characterized in that: The advanced search instructions include selecting corresponding keywords from the graph neural network after angle transformation and / or position transformation.

10. An information recommendation system based on brain science knowledge extraction and graph network, characterized in that: The system comprises: A keyword-relation extraction system configured to extract multiple keywords and textual relations between keywords in text information by using a semantic analysis method based on brain science; A knowledge graph construction module, configured to construct a knowledge graph based on a plurality of the keywords and textual relations between the keywords; A graph neural network construction module is configured to construct a knowledge relationship network based on the knowledge graph, and train it through a graph data learning algorithm to obtain a graph neural network; the graph neural network contains a representation vector of each keyword; The recommended content selection module is configured to determine the recommended content by calculating the semantic similarity between the search instruction and the representation vector of the keyword according to the search instruction input by the user; The visualization module is configured to visualize the recommendation words and the representational relationships associated with the recommendation words.

Images