A method and system for automatically constructing review literature mind map

Abstract

The present invention discloses a method and system for automatically constructing a review literature mind map, which relates to the field of data processing technology. The method comprises: parsing target review literature, extracting text, reference and hierarchical object information, and outputting structured text; using a large language model to analyze hierarchical objects and extract macro semantic information; interacting with a literature database, matching and extracting references, parsing and obtaining original reference information; inputting the text and original reference information into a large language model, parsing micro semantic information; associating and integrating micro and macro semantic information, and obtaining mind map construction information; importing a mind map construction tool, and automatically generating a literature mind map. Thus, the technical effect of deep semantic recognition and improving the quality of mind maps is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for automatically constructing a review literature mind map. Background Art

[0002] Mind mapping, as an intuitive graphical display method, extracts important information from the literature and presents it in a graphical form, which helps to sort out the knowledge framework and identify research trends.

[0003] In the prior art, methods for literature mind mapping mainly involve structured analysis and information extraction of literature. Some technical solutions propose to determine the outline information of the mind map by parsing the chapter names of review literature, and construct node content based on this outline information; however, the chapter titles in the literature mostly only provide general information, resulting in relatively brief analysis results, insufficient analysis, lack of deep semantic extraction and integration, and inaccurate reflection of the mind map. Summary of the invention

[0004] The present invention provides a method and system for automatically constructing a review literature mind map, so as to solve the technical problems in the prior art of insufficiently detailed analysis, lack of deep semantic extraction and integration, and inaccurate mind map reflection, and achieve the technical effect of deep semantic recognition and improving the quality of mind maps.

[0005] In a first aspect, the present invention provides a method for automatically constructing a review literature mind map, wherein the method comprises:

[0006] Parse the target review literature, extract the main text object information, reference object information and hierarchical object information, and output them as structured text information.

[0007] The hierarchical object information is analyzed according to a pre-built large language model to extract macro semantic information.

[0008] Based on the text object information and the reference object information, an interactive document database is used to match and extract corresponding references, and the original reference information is parsed and obtained.

[0009] The text object information and the original reference information are input into a pre-built large language model, and parsed to obtain micro semantic information.

[0010] Identify and associate micro semantic information with macro semantic information, and fuse the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information.

[0011] The mind map construction information is imported into the target map construction tool to automatically construct a mind map of the target review literature.

[0012] In a second aspect, the present invention also provides a system for automatically constructing a review literature mind map, wherein the system comprises:

[0013] The document information parsing module is used to parse the target review document, extract the main text object information, reference object information and hierarchical object information, and output it as structured text information.

[0014] The macro semantic analysis module is used to analyze the hierarchical object information according to a pre-built large language model and extract macro semantic information.

[0015] The reference matching module is used to interact with the document database based on the text object information and the reference object information, match and extract the corresponding references, and parse to obtain the original reference information.

[0016] The micro semantic parsing module is used to input the text object information and the original reference information into a pre-built large language model to parse and obtain micro semantic information.

[0017] The semantic information fusion module is used to identify and associate micro semantic information with macro semantic information, and fuse the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information.

[0018] The mind map generation module is used to import the mind map construction information into the target map construction tool to automatically construct the mind map of the target review literature.

[0019] The present invention discloses a method and system for automatically constructing a review document mind map, comprising: parsing a target review document, extracting text object information, reference object information and hierarchical object information, and outputting them as structured text information; analyzing hierarchical object information according to a pre-constructed large language model, and extracting macro semantic information; interacting with a document database based on the text object information and the reference object information, matching and extracting corresponding reference documents, and parsing to obtain original reference information; inputting the text object information and the original reference information into a pre-constructed large language model, parsing to obtain micro semantic information; identifying and associating micro semantic information with macro semantic information, and fusing the micro semantic information with the macro semantic information according to the association result, and obtaining mind map construction information; importing the mind map construction information into a target map construction tool, and automatically constructing a mind map of the target review document. The method and system for automatically constructing a review document mind map disclosed by the present invention solve the technical problems of insufficiently detailed analysis, lack of deep semantic extraction and integration, and inaccurate mind map reflection, and achieves the technical effect of deep semantic recognition and improved mind map quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1A schematic diagram of a process of automatically constructing a review literature mind map according to the present invention;

[0021] Figure 2 The present invention is a schematic diagram of the structure of a system for automatically constructing a review literature mind map.

[0022] Explanation of the accompanying drawings: document information parsing module 11, macro semantic analysis module 12, reference matching module 13, micro semantic parsing module 14, semantic information fusion module 15, mind map generation module 16. DETAILED DESCRIPTION

[0023] The technical solution provided in the embodiments of the present invention is to solve the technical problems of the prior art that the analysis is not detailed enough, the deep semantic extraction and integration is lacking, and the mind map reflects inaccurately. The overall idea adopted is as follows:

[0024] First, the target review literature is parsed, the main text object information, reference object information and hierarchical object information are extracted, and output as structured text information; then, the hierarchical object information is analyzed according to the pre-constructed large language model to extract macro-semantic information; then, based on the main text object information and the reference object information, the literature database is interacted with to match and extract the corresponding references, and the original reference information is parsed to obtain the original reference information; then, the main text object information and the original reference information are input into the pre-constructed large language model to parse and obtain the micro-semantic information; then, the micro-semantic information and the macro-semantic information are identified and associated, and the micro-semantic information and the macro-semantic information are fused according to the association result to obtain the mind map construction information; finally, the mind map construction information is imported into the target map construction tool to automatically construct the mind map of the target review literature.

[0025] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0026] Embodiment 1

[0027] Figure 1 The present invention is a flowchart of a method for automatically constructing a review literature mind map, wherein the method comprises:

[0028] S100: Parse the target review literature, extract the main text object information, reference object information and hierarchical object information, and output them as structured text information.

[0029] Specifically, the text object information is used to store the paragraph text of the target review document, the chapter number and chapter name to which the paragraph belongs, and the reference object information is the information of the extracted reference object, including document ID, title, author, publication year, journal information, etc., among which the reference is the unique identifier of each reference; the journal information involves volume number and issue number, journal name, page number of the reference in the journal, etc.

[0030] Specifically, the hierarchical object information refers to the hierarchical structure information of the document, such as chapters, sub-chapters, etc., to ensure that the hierarchical structure of the document can be correctly preserved and presented.

[0031] Specifically, the above-mentioned main text object information, reference object information and hierarchical object information are output as structured text information, which is convenient for subsequent data processing, document analysis and citation management.

[0032] In some embodiments, before parsing the target review document, extracting the main text object information, reference object information and hierarchical object information, and outputting them as structured text information, the method further includes:

[0033] Obtain the original document file of the target review document; perform format conversion of the original document file to obtain a standard document file, wherein the information of the standard document file includes the document title, document author, abstract, text, and references; and output the standard document file as a parsing object.

[0034] Specifically, first, obtain the original file of the target review document from an electronic database or other source. Preferably, the original document file is in PDF format. Then, use a specialized format conversion tool (such as PDF converter, OCR technology, Word to LaTeX converter, etc.) to convert the original document into a standard format. The standard format may be LaTeX, BibTeX, XML or JSON format to ensure that all content is machine readable.

[0035] Preferably, the acquired single original document file in PDF format is converted into an XML format file through grobid, and the converted XML format file retains the title, author, abstract, text, and reference information of the document.

[0036] Specifically, the standard document file is the basis for parsing and extracting structured text information. The standard document file obtained through the above steps provides a clear document basis for subsequent parsing.

[0037] In some implementations, the text object information includes text information and citation information, and the citation information is associated with the reference object information via a reference ID number.

[0038] Specifically, the text object information includes the text information and citation information associated by the reference ID number to ensure that the citation and the corresponding reference can be accurately matched; wherein, the citation information refers to the relevant information of other documents or references mentioned in the text. Exemplarily, the citation information includes the position of the citation in the text (i.e., the starting position and the ending position mentioned in the citation), the reference ID involved, etc.

[0039] By associating each citation information in the text with a specific reference ID number, we ensure that the citation points to the correct reference, provide a clear link between the text and the cited reference, and facilitate subsequent data analysis, document management and citation tracking.

[0040] In some implementations, the body object information and the reference object information are stored in a json file format, and the hierarchy object information is stored in a markdown file format.

[0041] Specifically, each main body object contains the main text, chapter information, and citation information, and the JSON format has a hierarchical structure and can clearly organize data, so it is suitable for storing main body object information; similarly, reference information is also stored in JSON format, because references often have unified fields (such as title, author, publication year, etc.), and the JSON format can easily store this data as key-value pairs.

[0042] Specifically, the hierarchical object information reflects the chapter, section and subsection structure of the document, and is organized by titles, subtitles, etc. The Markdown format can define the hierarchy of chapters through title tags, and is therefore suitable for storing hierarchical structures such as hierarchical object information.

[0043] The above storage method makes data easy to manage and query, and also facilitates subsequent data processing, analysis and display.

[0044] S200: Analyze the hierarchical object information according to a pre-built large language model to extract macro semantic information.

[0045] Specifically, the macro-semantic information involves the review topics, research questions, and research answers covered by the target review literature, among which the macro-semantic information may include multiple sets of interrelated research questions and research answers. Through the above-mentioned analysis and extraction, it is possible to identify the main research issues, main technical methods, main achievements and contributions, future development directions, research hotspots in the field, and other macro-semantic viewpoints proposed by the authors that reflect the review topics, and output them as macro-semantic information.

[0046] In some embodiments, analyzing the hierarchical object information according to a pre-built large language model to extract macro semantic information includes:

[0047] Define macro-semantic recognition words, including review topic recognition prompt words, research question recognition prompt words and answer recognition prompt words, wherein the review topic recognition prompt words, research question recognition prompt words and answer recognition prompt words have a cascade relationship; according to the macro-semantic recognition words, call a large language model to perform hierarchical recognition of the hierarchical object information, and obtain the review topic, research question set and corresponding question answer set of the target review document; based on the cascade relationship, structured store the review topic, the research question set and the question answer set, and output them as the macro-semantic information.

[0048] Specifically, in order to achieve macro-semantic analysis of hierarchical object information and extract review topics, research questions and answers, first, a series of macro-semantic identification words need to be defined. Macro-semantic identification words are used to help the model understand the key information in the literature and classify it into review topics, research questions and answers; among them, macro-semantic identification words include review topic identification prompt words for identifying the core topics or research fields of the literature, research question identification prompt words for identifying research questions raised in the literature, and answer identification prompt words for identifying research conclusions or answers in the literature.

[0049] Specifically, there is a cascading relationship between the review topic identification prompts, the research question identification prompts and the answer identification prompts. In other words, the review topic identification prompts include multiple different levels of research question identification prompts, and each different level of research question identification prompts has a corresponding answer identification prompt. Through the above-mentioned cascading relationship, each research question can be gradually analyzed to obtain the relevant answers, and the obtained research questions and related answers can be organized in a hierarchical and step-by-step manner. In other words, the answers are cascaded extracted based on the in-depth analysis of the research questions and related answers. For each research question, the corresponding answer is the content that follows it, which is determined by the contextual relationship in the literature.

[0050] Through the above-mentioned cascading relationship, we ensure that the various research questions and answers in the obtained literature are not isolated, but form a hierarchical and progressive relationship network, that is, each answer corresponds to a relevant research question, which helps to deeply understand the information structure in the review literature and efficiently extract key information, ultimately providing support for the construction of mind maps and knowledge graphs.

[0051] Specifically, a pre-built large language model (such as the Qwen2-7B-Instruct large model) is used to hierarchically identify the documents according to their chapter titles and content paragraphs. The content of each level (such as chapters, subsections, etc.) is extracted and organized according to the cascading relationship. The extracted macro-semantic information is structured and stored, that is, each review topic and related research questions and answers are organized in a hierarchical relationship, and output as structured data for subsequent query and display.

[0052] S300: Based on the text object information and the reference object information, interact with the literature database, match and extract corresponding references, and parse to obtain original reference information.

[0053] Specifically, the reference ID, document title and other information are extracted from the text and reference object information as query conditions to query the reference records in the database, wherein the above-mentioned document database refers to a document management system or document data platform that stores reference metadata.

[0054] Optionally, the references extracted from the literature database may be in a standardized format, such as document metadata in XML format, or may be other non-standardized data formats, such as PDF, CAJ, etc.; if the references extracted from the literature database are non-standardized, the corresponding original reference information is parsed into a standardized XML format.

[0055] S400: Input the text object information and the original reference information into a pre-built large language model, and parse and obtain micro semantic information.

[0056] In some embodiments, the text object information and the original reference information are input into a pre-built large language model, and parsed to obtain micro semantic information, including:

[0057] A highlight dimension set is defined, and micro-prompt words are defined according to the highlight dimension set to obtain a micro-prompt word library, wherein the micro-prompt word library at least includes reference emotion prompt words, problem-solving prompt words, technical method prompt words and achievement contribution prompt words; according to the scenario demand information of the target review document, the corresponding micro-prompt words are called in the micro-prompt word library, and the target prompt word set is output; the text object information and the original reference information are used as input, and semantic parsing is performed in combination with the target prompt word set and a pre-built large language model to obtain the micro-semantic information, wherein the micro-semantic information at least includes emotion information, problem-solving information, technical method information and achievement contribution information.

[0058] Specifically, firstly, according to the common research needs of review literature, the important focus dimensions in the review literature are defined, and a focus dimension set is generated. The focus dimension set corresponds to all the information dimensions that need to be paid attention to in the review literature mind map construction task (that is, the key information that needs to be extracted for building the mind map); then, according to the defined focus dimension set, the corresponding micro-prompt vocabulary is constructed. The micro-prompt vocabulary contains keywords or sentence patterns of specific dimensions, which are used to guide the large language model to extract specific information.

[0059] Specifically, the micro-prompt word library at least includes: citation sentiment prompt words, such as "positive evaluation", "highly innovative", "solved the problem...", "contributed to..." research, etc.; problem-solving prompt words, such as "research problem", "technical bottleneck", "key issue", "the challenge is...", etc.; technical method prompt words, such as "the method includes...", "the technology used is...", "a new...method is proposed", etc.; achievement contribution prompt words, such as "the innovation lies in...", "the contribution lies in...", "proposed for the first time...", "significantly improved...", etc.

[0060] Specifically, combined with the background of the specific target review literature, the scenario requirements of the application scenario corresponding to the mind map construction task of the target review literature are analyzed and determined to help select appropriate prompt words. Among them, the information dimensions involved in the micro-semantic analysis of review literature in different application scenarios are not the same. Therefore, according to the scenario requirements of the target review literature, a suitable prompt word set is extracted from the micro-prompt word library.

[0061] Specifically, the extracted text object information (including paragraph text, chapter information, etc.) and the original reference information (such as the original text, table of contents, chapter information, etc. of the cited document) are input into the large language model together, and a target prompt word set is provided to guide the large language model to identify specific micro-semantic information during the parsing process.

[0062] Exemplarily, micro-semantic information includes: citation sentiment, that is, analyzing the sentiment of citations in the literature, such as whether the literature's evaluation of the predecessors' work is positive or negative, including the overall tendency of the literature, criticism or praise of predecessors' research; problem-solving information, that is, the research objectives solved by the literature and the specific problems actually solved, such as "proposing a new method to solve the problems existing in traditional methods"; technical method information, which refers to the technical methods, models, experimental designs, etc. used in the literature, such as specific algorithms, system architectures, research methods, etc.; achievement contribution information, that is, the main achievements and contributions of the literature, including theoretical innovations, technological breakthroughs, and developments in applications.

[0063] Through the above process, based on the main text object information and reference object information, the micro-semantic information of the document is obtained by parsing through the pre-built large language model, which helps to provide deeper information support for further mind map construction.

[0064] S500: Identify and associate micro semantic information with macro semantic information, and fuse the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information.

[0065] Specifically, the research questions and answers in the macro-semantic information corresponding to the cited references are identified one by one, so as to associate the macro-information with the micro-information, and save the corresponding macro-semantic information into the citation information to obtain the mind map construction information.

[0066] In some embodiments, identifying and associating the micro semantic information with the macro semantic information, and fusing the micro semantic information with the macro semantic information according to the association result to obtain the mind map construction information includes:

[0067] According to the macro-semantic information, the cascade relationship is extracted to construct the main structure of the mind map, wherein the main structure includes a theme corresponding to the review theme and sub-themes corresponding to the research question set and the question answer set; taking the reference ID number of the micro-semantic information as the index constraint and the association relationship between the citation information and the reference object information as the index space, traversing and matching to obtain the association result of the micro-semantic information and the macro-semantic information; according to the association result, extracting the reference object basic information corresponding to the micro-semantic information from the reference object information, and extracting the corresponding emotional information from the micro-semantic information; embedding the reference object basic information and the emotional information into the associated sub-themes to generate multiple reference nodes; according to the association result, embedding the problem solving information, the technical method information and the achievement contribution information into the corresponding reference nodes to form a corresponding leaf node set; outputting the embedding result as the mind map construction information.

[0068] Specifically, first, a trunk structure is defined according to the macro-semantic information, wherein the review topic in the macro-semantic information serves as the highest-level node of the trunk structure, namely the theme layer, and the research question set and the question answer set serve as the sub-node structure to display the key research questions and their corresponding answers. Optionally, the answer to each question forms a sub-node associated with the sub-topic.

[0069] Specifically, first, the corresponding reference ID number is screened in the micro-semantic information, and then, based on the association between the reference ID number in the micro-semantic information and the citation information and the reference object information, a matching index is established; then, the citation relationship between the research questions, answers and references in the matching macro-semantic information is traversed to obtain the association result between the micro-semantic information and the macro-semantic information, which defines the position of the macro-semantic information to which different micro-semantic information belongs.

[0070] Furthermore, based on the association results, the micro-semantic information is embedded into the main structure of the mind map. First, basic information is obtained from the reference object information, such as document title, author, year, etc., and the citation sentiment is extracted from the micro-semantic information, such as positive and negative evaluation; then, it is embedded in the corresponding sub-topic node to generate a reference node.

[0071] Furthermore, the problem-solving information, technical method information and achievement contribution information in the micro-semantic information are further embedded into the corresponding reference nodes, and a corresponding leaf node set is formed under the reference nodes. The embedded trunk structure and leaf node set are then organized into a hierarchical structure that supports mind mapping and output as mind map construction information.

[0072] Through the above-mentioned extraction of the main structure, the association of micro and macro semantic information, the embedding of basic information and emotions, and the generation of leaf nodes, the generation of mind map construction information can be completed. The above-mentioned multi-level nesting method ensures the comprehensiveness, logic and visualization of the information.

[0073] In some implementations, the output mind map construction information is in markdown file format.

[0074] Specifically, the above mind map construction information is output in the markdown file format, which retains the clear multi-level hierarchical relationship and facilitates the subsequent introduction of corresponding mind map tools for visual rendering.

[0075] S600: Importing the mind map construction information into a mind map construction tool to automatically construct a mind map of the target review literature.

[0076] Optionally, the target mind map building tools include mind mapping software such as XMind, MindMeister, and FreeMind. By inputting the mind map building information into the selected mind map building tool and combining the script or tool's API for batch processing and automation, each sub-topic and its related literature information are mapped as a mind map node according to predefined rules to build a mind map; among which, the predefined rules are used to define the node position, color, font style, etc. of the mind map to improve the visualization effect and facilitate users to analyze and understand the core content of the review literature.

[0077] In summary, the method for automatically constructing a review literature mind map provided by the present invention has the following technical effects:

[0078] By parsing the target review literature, the main text object information, reference object information and hierarchical object information are extracted and output as structured text information; the hierarchical object information is analyzed according to the pre-built large language model to extract macro-semantic information; based on the main text object information and the reference object information, the literature database is interacted with to match and extract the corresponding references, and the original reference information is parsed to obtain the original reference information; the main text object information and the original reference information are input into the pre-built large language model to obtain the micro-semantic information; the micro-semantic information and the macro-semantic information are identified and associated, and the micro-semantic information and the macro-semantic information are fused according to the association results to obtain the mind map construction information; the mind map construction information is imported into the target map construction tool to automatically construct the mind map of the target review literature, thereby achieving the technical effect of deep semantic recognition and improving the quality of mind maps.

[0079] Embodiment 2

[0080] Based on the same inventive concept as the method for automatically constructing a review literature mind map in the aforementioned embodiment, Figure 2 As shown, the present application provides a system for automatically constructing a review literature mind map, the system comprising:

[0081] The document information parsing module 11 is used to parse the target review document, extract the main text object information, reference object information and hierarchical object information, and output it as structured text information.

[0082] The macro semantic analysis module 12 is used to analyze the hierarchical object information according to a pre-built large language model and extract macro semantic information.

[0083] The reference matching module 13 is used to interact with the document database based on the text object information and the reference object information, match and extract corresponding references, and parse to obtain original reference information.

[0084] The micro semantic parsing module 14 is used to input the text object information and the original reference information into a pre-built large language model to parse and obtain micro semantic information.

[0085] The semantic information fusion module 15 is used to identify and associate micro semantic information with macro semantic information, and fuse the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information.

[0086] The mind map generation module 16 is used to import the mind map construction information into the target map construction tool to automatically construct a mind map of the target review literature.

[0087] In some embodiments, the document information parsing module 11 includes:

[0088] The original document acquisition unit is used to acquire the original document file of the target review document.

[0089] The format conversion unit is used to perform format conversion of the original document file to obtain a standard document file, wherein the information of the standard document file includes the document title, document author, abstract, text, and references.

[0090] The standard document output unit is used to output the standard document file as a parsing object.

[0091] Furthermore, in the document information parsing module 11, the text object information includes text information and citation information, and the citation information and the reference object information are associated through a reference document ID number.

[0092] Furthermore, in the document information parsing module 11, the text object information and the reference object information are stored in a json file format, and the hierarchical object information is stored in a markdown file format.

[0093] In some embodiments, the macro semantic analysis module 12 includes:

[0094] The identification word definition unit is used to define macro-semantic identification words, including review topic identification prompt words, research question identification prompt words and answer identification prompt words, wherein the review topic identification prompt words, research question identification prompt words and answer identification prompt words have a cascade relationship.

[0095] The information hierarchical identification unit is used to call a large language model to perform hierarchical identification of the hierarchical object information according to the macro-semantic identification words, and obtain the review topic, research question set and corresponding question answer set of the target review document.

[0096] A macro semantic information output unit is used to structurely store the review topic, the research question set and the question answer set based on the cascade relationship, and output the macro semantic information.

[0097] In some embodiments, the micro semantic parsing module 14 includes:

[0098] The micro-prompt word library construction unit is used to define a highlight dimension set, and define micro-prompt words according to the highlight dimension set to obtain a micro-prompt word library, wherein the micro-prompt word library at least includes reference emotion prompt words, problem-solving prompt words, technical method prompt words and achievement contribution prompt words.

[0099] The prompt word set calling unit is used to call the corresponding micro prompt words in the micro prompt word library according to the scene requirement information of the target review document, and output them as the target prompt word set.

[0100] The micro-semantic information parsing unit is used to take the main text object information and the original reference information as input, combine the target prompt word set and the pre-built large language model to perform semantic parsing, and obtain the micro-semantic information, wherein the micro-semantic information at least includes emotional information, problem-solving information, technical method information and achievement contribution information.

[0101] In some embodiments, the semantic information fusion module 15 includes:

[0102] A trunk extraction and construction unit is used to extract the cascade relationship and construct the trunk structure of the mind map according to the macro semantic information, wherein the trunk structure includes the theme corresponding to the review theme and the sub-themes corresponding to the research question set and the question answer set.

[0103] The association result acquisition unit is used to use the reference ID number of the micro semantic information as an index constraint and the association relationship between the citation information and the reference object information as an index space to traverse and match to obtain the association result of the micro semantic information and the macro semantic information.

[0104] The subtopic information extraction unit is used to extract the reference object basic information corresponding to the micro semantic information from the reference object information according to the association result, and to extract the corresponding emotional information from the micro semantic information.

[0105] A reference document node generation unit is used to embed the reference object basic information and the sentiment information into the associated sub-topic to generate multiple reference document nodes.

[0106] The leaf node set forming unit is used to embed the problem solving information, the technical method information and the achievement contribution information into the corresponding reference document nodes according to the association result to form a corresponding leaf node set.

[0107] The construction information output unit is used to output the embedding result as the mind map construction information.

[0108] In some implementations, in the semantic information fusion module 15, the mind map construction information output is in a markdown file format.

[0109] Through the above-mentioned detailed description of a method for automatically constructing a review literature mind map in this specification, those skilled in the art can clearly know a system for automatically constructing a review literature mind map in this embodiment. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0110] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatically constructing a review literature mind map, characterized in that: The method comprises: Parse the target review literature, extract the main text object information, reference object information and hierarchical object information, and output them as structured text information; Analyze the hierarchical object information according to a pre-built large language model to extract macro semantic information; Based on the text object information and the reference object information, the literature database is interacted with, the corresponding references are matched and extracted, and the original reference information is parsed and obtained; Input the text object information and the original reference information into a pre-built large language model, and parse and obtain micro semantic information; Identify and associate micro semantic information with macro semantic information, and fuse the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information; Importing the mind map construction information into the target map construction tool to automatically construct a mind map of the target review literature; Analyze the hierarchical object information according to the pre-built large language model and extract macro semantic information, including: Define macro semantic identification words, including review topic identification prompt words, research question identification prompt words and answer identification prompt words, wherein the review topic identification prompt words, research question identification prompt words and answer identification prompt words have a cascade relationship; According to the macro-semantic recognition words, a large language model is called to perform hierarchical recognition of the hierarchical object information, and the review topic, research question set and corresponding question answer set of the target review literature are obtained; Based on the cascade relationship, the review topic, the research question set and the question answer set are structuredly stored and output as the macro semantic information; Identifying and associating the micro semantic information with the macro semantic information, and fusing the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information, including: According to the macro semantic information, the cascade relationship is extracted to construct a trunk structure of a mind map, wherein the trunk structure includes a theme corresponding to the review theme and sub-themes corresponding to the research question set and the question answer set; Using the reference ID number of the micro-semantic information as an index constraint and the association relationship between the citation information and the reference object information as an index space, traversing and matching to obtain the association result between the micro-semantic information and the macro-semantic information; According to the association result, extracting the reference object basic information corresponding to the micro semantic information from the reference object information, and extracting the corresponding emotion information from the micro semantic information; Embedding the reference object basic information and the sentiment information into the associated subtopics to generate a plurality of reference document nodes; According to the association result, the problem-solving information, technical method information and achievement contribution information are embedded into the corresponding reference nodes to form a corresponding leaf node set; The output embedding result is the mind map construction information.

2. The method for automatically constructing a review literature mind map according to claim 1, characterized in that: Parse the target review literature, extract the main text object information, reference object information and hierarchical object information, and output them as structured text information. Previously, it also included: Obtain the original literature files of the target review literature; Execute format conversion of the original document file to obtain a standard document file, wherein the information of the standard document file includes document title, document author, abstract, text, and references; The standard document file is output as an analysis object.

3. The method for automatically constructing a review literature mind map according to claim 2, characterized in that: The text object information includes text information and citation information, and the citation information is associated with the reference object information through a reference document ID number.

4. The method for automatically constructing a review literature mind map according to claim 2, characterized in that: The body object information and the reference object information are stored in a json file format, and the hierarchy object information is stored in a markdown file format.

5. The method for automatically constructing a review literature mind map according to claim 1, characterized in that: Input the text object information and the original reference information into the pre-built large language model, and parse to obtain micro semantic information, including: Define a highlight dimension set, and define micro prompt words according to the highlight dimension set to obtain a micro prompt word library, wherein the micro prompt word library at least includes reference emotion prompt words, problem-solving prompt words, technical method prompt words and achievement contribution prompt words; According to the scenario requirement information of the target review document, the corresponding micro prompt words are called from the micro prompt word library and output as a target prompt word set; The main text object information and the original reference information are used as input, and semantic analysis is performed in combination with the target prompt word set and the pre-built large language model to obtain the micro semantic information, wherein the micro semantic information at least includes emotional information, problem-solving information, technical method information and achievement contribution information.

6. The method for automatically constructing a review literature mind map according to claim 1, characterized in that: The output mind map construction information is in markdown file format.

7. A system for automatically constructing a review literature mind map, characterized in that: The system is used to execute the method for automatically constructing a review literature mind map according to any one of claims 1 to 6, and the system comprises: The document information parsing module is used to parse the target review document, extract the main text object information, reference object information and hierarchical object information, and output it as structured text information; A macro semantic analysis module, used to analyze the hierarchical object information according to a pre-built large language model and extract macro semantic information; A reference matching module is used to interact with a document database based on the text object information and the reference object information, match and extract corresponding references, and parse to obtain original reference information; A micro-semantic parsing module, used for inputting the text object information and the original reference information into a pre-built large language model, parsing and acquiring micro-semantic information; A semantic information fusion module is used to identify and associate micro semantic information with macro semantic information, and fuse the micro semantic information with the macro semantic information according to the association result to obtain mind map construction information; The mind map generation module is used to import the mind map construction information into the target map construction tool to automatically construct the mind map of the target review literature.

Images