Medical literature intelligent processing and analysis system, method and terminal based on large model

Abstract

The invention provides a medical literature intelligent processing and analysis system and method based on a large model and a terminal.The method comprises the steps that firstly, a query request of a user is obtained through a user search interface, then the large model is called for user intention recognition, related biological entities are extracted, and a corresponding logic chain instruction is generated; matching background information of related entities in a pre-constructed biological entity library according to the extracted biological entities; matching in a literature abstract knowledge graph database according to the logic chain instruction to obtain a correlation sequence of the literatures; and finally, integrating the logic chain instruction, the biological entity background information and the related literature information to generate and display a corresponding query result. According to the method, the problems of tedious document acquisition, time-consuming reading, complicated analysis and the like faced by medicine researchers are solved, intelligent document processing and analysis are performed by adopting a method based on a large model, the document reading process is accelerated, the manual analysis cost is reduced, and the research efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical document processing, and in particular to a medical document intelligent processing and analysis system, method and terminal based on a large model. Background Art

[0002] With the rapid development of scientific research, the annual publication volume of scientific and technological literature has increased exponentially. This growth trend has brought great challenges to researchers, especially in terms of efficiently acquiring, reading and analyzing a large amount of literature. Traditional literature retrieval and reading methods are often inefficient and difficult to cope with the information overload caused by massive literature.

[0003] There are many document management and analysis tools on the market, such as Aminor and Elicit, which use large language models to implement intelligent retrieval, automatic summary generation, and interactive document question-and-answer functions. However, these tools have significant limitations in obtaining medical literature information: their information sources mainly come from the metadata of the document (such as title, abstract) or the PDF full text manually uploaded by the user, and they cannot fully obtain and process the complete content including non-open access documents. This limitation directly affects the mining depth of the document content and the effect of knowledge extraction. In addition, for PDF full-text analysis, existing tools usually adopt the method of directly inputting full-text information into a large language model. However, considering the limitations of the current large language model in terms of context length, this processing method inevitably introduces a large amount of redundant information. This not only reduces the accuracy and efficiency of the analysis, but also significantly increases the consumption of computing resources. Summary of the invention

[0004] In view of the shortcomings of the prior art mentioned above, the purpose of the present invention is to provide a medical literature intelligent processing and analysis system, method and terminal based on a large model, which is used to solve the technical problems of medical literature management and analysis tools in the prior art, such as cumbersome document acquisition, time-consuming reading, and complex analysis.

[0005] To achieve the above-mentioned purpose and other related purposes, the present invention provides a medical literature intelligent processing and analysis system based on a large model, the system comprising: a user interaction module, used to receive a user's query request through a user search interface; wherein the types of the query request include: a document keyword query request or a document-related question query request; a user intention recognition module, connected to the user interaction module, used to perform user intention recognition based on the large model according to the user's query request, and extract biological entities and generate logical chain instructions; a biological entity module, connected to the user intention recognition module, used to retrieve matching corresponding entity background information from a pre-built biological entity library based on the extracted biological entities; a document abstract knowledge graph module, connected to the biological entity module, used to retrieve matching corresponding related document information from a pre-built document abstract knowledge graph database based on the large model according to the extracted biological entities and the generated logical chain instructions; a query result acquisition module, connected to the user interaction module, the user intention recognition module, the biological entity module and the document abstract knowledge graph module, used to integrate the logical chain instructions, the entity background information and the related document information based on the large model to generate a query result corresponding to the user's query request and display it by the user interaction module.

[0006] In one embodiment of the present invention, the user interaction module includes: a user search interface for receiving literature keywords to be queried or literature-related questions input by the user; a user guidance and prompt interface for showing the main functions and usage methods of the system to first-time users; an information display interface, including: a data statistics display area, a graph visualization display area, and an intelligent dialogue area, for displaying literature query results for literature keyword queries or answer results for question feedback.

[0007] In one embodiment of the present invention, the biological entity module includes: a biological entity library construction unit, which is used to integrate multiple data sources to obtain multi-source biomedical entity information to construct a biological entity library; an entity information retrieval unit, which is connected to the biological entity library construction unit and is used to retrieve matching corresponding entity background information from the biological entity library based on the extracted biological entity.

[0008] In one embodiment of the present invention, the document abstract knowledge graph module includes: a document collection unit, which is used to dynamically update documents and extract the core information of documents; a document parsing unit, which is connected to the document collection unit and is used to perform in-depth analysis and systematic summary of the core information of each document based on a large model to obtain multi-dimensional document parsing results; a knowledge graph construction unit, which is connected to the document parsing unit and is used to extract biological entities, relationship information and the level of evidence of each entity relationship in the core information from the document parsing results based on a large model, and construct a document abstract knowledge graph database; a document information retrieval unit, which is connected to the knowledge graph construction unit and is used to retrieve and match the corresponding relevant document information in the document abstract knowledge graph database based on the large model according to the extracted biological entities and the generated logical chain instructions, and can sort them based on relevance, document citations and publication time.

[0009] In one embodiment of the present invention, the document collection unit includes: an incremental crawler update subunit, which is used to crawl documents in an incremental crawler manner based on the set crawling frequency and concurrency; a document grouping and storage subunit, connected to the incremental crawler update subunit, and used to classify and store the crawled documents according to fields; a core information extraction subunit, connected to the document grouping and storage subunit, and used to perform adaptive structural analysis on the classified documents and extract core paragraph information and charts; wherein the core paragraph information includes: Title, Highlight, Abstract, Introduction, Results, legends and titles of Figures and Tables, Discussion, and Conclusion.

[0010] In one embodiment of the present invention, the document parsing unit includes: a translation subunit, which is used to call a large model for performing a translation task and perform a Chinese translation of the core information of the extracted document in context; a structured summary generation subunit, which is connected to the translation subunit and is used to process the core information of each document based on a framework parser to obtain a corresponding structured summary and chart parsing results; wherein the structured summary includes: research highlights, research abstracts, research questions, research methods, research results, research deficiencies and research conclusions.

[0011] In one embodiment of the present invention, the knowledge graph construction unit includes: an entity and relationship extraction subunit, which is used to extract biological entity and relationship information according to the core information of each document based on a large model that performs biological entity and relationship extraction tasks, and determine the evidence level of each entity relationship based on the research method in the structured summary of each document; a knowledge graph generation subunit, which is connected to the entity and relationship extraction subunit and is used to generate a document abstract knowledge graph database based on the extracted biological entities, relationship information and the evidence level of each entity relationship and the structured summary of each document.

[0012] In one embodiment of the present invention, based on the framework parser, the core information of each document is processed to obtain the corresponding structured summary and chart parsing results, including: based on the big model that performs the chart parsing task, the chart in the core information of the document is parsed to obtain the chart parsing results; based on the core paragraph information in the core information of the document and the chart parsing results, the structured summary of the document is obtained, and the method includes: taking the entire Highlight paragraph of the document as the research highlight of the document; based on the big model that performs the document summary task, obtaining the research summary of the document according to the entire Abstract paragraph, the last paragraph of the Introduction, and the first paragraph of the Discussion of the document; based on the big model that performs the document summary task, generating the research problem of the document according to all paragraphs of the Introduction of the document except the last paragraph; based on the big model that performs the document summary task, generating the research method of the document according to the legend and title of the Figure of the document; based on the big model that performs the document summary task, generating the research results of the document according to the title of the Results of the document, the legend and title of the Figure, and the chart parsing results; based on the big model that performs the document summary task, generating the research deficiencies of the document according to all paragraphs of the Discussion of the document except the first paragraph; and taking the entire Conclusion paragraph in the document as the research conclusion of the document.

[0013] To achieve the above-mentioned purpose and other related purposes, the present invention provides a method for intelligent processing and analysis of medical literature based on a big model, the method comprising: receiving a user's query request through a user search interface; wherein the types of the query request include: a document keyword query request or a document-related question query request; based on the big model, performing user intent recognition according to the user's query request, extracting biological entities and generating logical chain instructions; based on the extracted biological entities, retrieving matching corresponding entity background information from a pre-built biological entity library; based on the big model, according to the extracted biological entities and the generated logical chain instructions, retrieving matching corresponding related document information from a pre-built document abstract knowledge graph database; based on the big model, integrating the logical chain instructions, entity background information and related document information to generate a query result corresponding to the user's query request and displaying it.

[0014] To achieve the above-mentioned purpose and other related purposes, the present invention provides an electronic terminal, comprising: one or more memories and one or more processors; the one or more memories are used to store computer programs; the one or more processors are connected to the memories and are used to run the computer program to execute the large model-based intelligent processing and analysis method of medical literature.

[0015] As described above, the present invention is a medical literature intelligent processing and analysis system, method and terminal based on a large model, which has the following beneficial effects: first, the user's query request is obtained through the user search interface, and then the large model is called to identify the user's intention, extract the relevant biological entities, and generate corresponding logical chain instructions; match the background information of the relevant entities in the pre-built biological entity library according to the extracted biological entities; and match the document abstract knowledge graph database according to the logical chain instructions to obtain the relevance ranking of the documents; finally, the logical chain instructions, biological entity background information and related document information are integrated to generate and display the corresponding query results. The present invention solves the problems faced by medical researchers, such as cumbersome document acquisition, time-consuming reading, and complex analysis. It adopts a method based on a large model for intelligent document processing and analysis, accelerates the document reading process, reduces the cost of manual analysis, and improves research efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Shown is a schematic diagram of the structure of a large model-based intelligent processing and analysis system for medical literature in one embodiment of the present invention.

[0017] Figure 2 Shown is a schematic diagram of the structure of a large model-based intelligent processing and analysis system for medical literature in one embodiment of the present invention.

[0018] Figure 3 Shown is a schematic diagram of the structure of the document abstract knowledge graph module in one embodiment of the present invention.

[0019] Figure 4 Shown is a schematic diagram of the core information structure analysis of a document by a framework parser in one embodiment of the present invention.

[0020] Figure 5 Shown is a flowchart of a method for intelligent processing and analysis of medical literature based on a large model in one embodiment of the present invention.

[0021] Figure 6 Shown is a flowchart of a method for intelligent processing and analysis of medical literature based on a large model in one embodiment of the present invention.

[0022] Figure 7Shown is a flowchart of a method for constructing a document abstract knowledge graph database in one embodiment of the present invention.

[0023] Figure 8 Shown is a schematic structural diagram of an electronic terminal in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0025] It should be noted that in the following description, reference is made to the accompanying drawings, which describe several embodiments of the present invention. It should be understood that other embodiments may also be used, and that mechanical composition, structural, electrical and operational changes may be made without departing from the spirit and scope of the present invention. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present invention is limited only by the claims of the published patents. The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. Spatially related terms, such as "upper", "lower", "left", "right", "below", "below", "lower", "above", "upper", etc., may be used in the text to facilitate the description of the relationship between an element or feature shown in the figure and another element or feature.

[0026] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case of "direct connection" but also the case of "indirect connection" by placing other elements therebetween. In addition, when a part is said to "include" a certain constituent element, unless otherwise stated, it does not exclude other constituent elements, but means that other constituent elements may be included.

[0027] The terms first, second and third mentioned herein are used to describe various parts, components, regions, layers and / or segments, but are not limited thereto. These terms are only used to distinguish a certain part, component, region, layer or segment from other parts, components, regions, layers or segments. Therefore, the first part, component, region, layer or segment described below may refer to the second part, component, region, layer or segment within the scope of the present invention.

[0028] Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless there is an indication to the contrary in the context. It should be further understood that the terms "comprise", "include" indicate the presence of the described features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Therefore, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". Exceptions to this definition will only occur when the combination of elements, functions or operations is inherently mutually exclusive in some way.

[0029] The present invention provides a medical literature intelligent processing and analysis system based on a large model. First, the user's query request is obtained through a user search interface, and then the large model is called to identify the user's intention, extract relevant biological entities, and generate corresponding logical chain instructions; match the background information of relevant entities in a pre-constructed biological entity library according to the extracted biological entities; and match the document abstract knowledge graph database according to the logical chain instructions to obtain the relevance ranking of the documents; finally, the logical chain instructions, biological entity background information and relevant document information are integrated to generate and display the corresponding query results. The present invention solves the problems faced by medical researchers, such as cumbersome document acquisition, time-consuming reading, and complex analysis. It uses a method based on a large model to perform intelligent document processing and analysis, accelerate the document reading process, reduce manual analysis costs, and improve research efficiency.

[0030] The following is a detailed description of the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0031] like Figure 1 A structural schematic diagram of a medical literature intelligent processing and analysis system based on a large model in an embodiment of the present invention is shown.

[0032] The system comprises:

[0033] The user interaction module 1 is used to receive the user's query request through the user search interface; wherein the types of the query request include: document keyword query request or document related question query request; specifically, the user inputs specific keywords, and the system retrieves relevant document materials in the database based on these keywords. This type of query is usually used to quickly find documents containing specific terms or topics. Users ask questions related to documents, and the system needs to understand these questions and provide corresponding answers or document recommendations. This type of query focuses more on solving users' questions and may require more complex natural language processing capabilities. While asking questions, users are allowed to upload files, ask targeted questions, and receive and view the system's answers.

[0034] The user intention recognition module 2 is connected to the user interaction module 1 and is used to recognize the user intention based on the large model according to the user's query request, extract biological entities, and generate logical chain instructions, which guide subsequent modules on how to retrieve and process information.

[0035] The biological entity module 3 is connected to the user intention recognition module 2 and is used to retrieve matching entity background information from a pre-built biological entity library based on the extracted biological entity.

[0036] The document abstract knowledge graph module 4 is connected to the biological entity module 3 and is used to retrieve matching corresponding relevant document information from a pre-built document abstract knowledge graph database based on the generated logic chain instructions.

[0037] The query result acquisition module 5 is connected to the user intention recognition module 2, the biological entity module 3 and the document abstract knowledge graph module 4, and is used to integrate the logic chain instructions, entity background information and related document information based on the big model to generate the query result corresponding to the user's query request, and the user interaction module 1 displays it.

[0038] By combining natural language processing, large model generation, retrieval enhancement and deep learning technology, the system provides users with intelligent and systematic literature analysis solutions. The system should be able to automatically identify and extract key information in the literature, such as research abstracts, background, results and conclusions, while filtering out non-critical content, thereby greatly improving the efficiency and accuracy of literature analysis and achieving efficient processing and in-depth mining of medical literature. In addition, the system has cross-literature knowledge association and reasoning capabilities, helping researchers quickly grasp the research frontier and discover potential research opportunities. This intelligent literature analysis system will greatly improve the work efficiency of scientific researchers and accelerate the process of scientific discovery.

[0039] In one embodiment, the system has a dynamic adjustment capability, and can generate prompt words for large models that are adapted to different task types according to different task types, and select large models of corresponding types for parsing.

[0040] Specifically include:

[0041] The large model for performing translation tasks uses an open source large language model deployed on the server, such as the GPT series, to perform Chinese translation in context based on the input prompt words adapted to the model.

[0042] A large model that performs document summarization tasks uses a large open source language model deployed on the server, such as BERT or BART, to generate summaries based on input prompt words adapted to the model.

[0043] The large model that performs chart parsing tasks uses an open source multimodal large language model deployed on the server, such as ChartLlama, to parse charts based on the input prompt words adapted to the model.

[0044] The large model that performs biological entity and relationship extraction tasks is used to extract biological entities and relationships based on the input prompt words that are adapted to the model. For example, the model uses commercial large model interfaces, including large models from OpenAI, Google, Baidu, Microsoft, Alibaba, Tencent and other companies; the interface supports automatic updates to ensure that the large models are the latest versions of each company.

[0045] A large model that performs intent recognition tasks is used to perform graph recognition based on input prompt words adapted to the model, extract biological entities and generate logical chain instructions.

[0046] In one embodiment, if Figure 2 , the user interaction module 1 comprises:

[0047] User guidance and prompt interface, design teaching functions for system use, provide specific examples for users to learn the main functions and usage of the system when using it for the first time;

[0048] The user search interface is used to receive the keywords of the literature to be queried or the related questions of the literature input by the user; the user can enter keywords, questions or specific query content. This can be a keyword of biological entities related to drugs, diseases, proteins, genes, etc., or a specific question related to medical literature. When the user enters the query, the system provides automatic completion and keyword prompt functions to help the user complete the input quickly. User query supports the use of Boolean logic for complex query combinations, such as "A AND B", "A OR B", etc. The configuration can be further set through the search setting module. The search configuration includes journal range, time span, document type, impact factor, and filters the abstracts of literature in specific fields and knowledge graph databases according to the screening conditions.

[0049] The information display interface includes: data statistics display area, graph visualization display area and intelligent dialogue area, which are used to display the literature query results for literature keyword queries or the answer results for question feedback.

[0050] Among them, the data statistics display area is used to display the literature list of the query results, and supports sorting by time, citation volume and relevance; it also supports filtering by time, document type and impact factor; the graph visualization display area is used to support interactive operations of the graph, including zooming, panning, local magnification, node dragging and shortest path analysis; the intelligent dialogue area uses large model retrieval enhancement technology to support multi-round dialogue understanding based on PDF documents.

[0051] In one embodiment, if Figure 2 , the user intention recognition module 2 includes:

[0052] PDF text recognition unit, which can receive local PDF document files uploaded by users, supports batch upload and drag-and-drop upload, and provides users with convenient file management functions, including file preview, management and deduplication. In addition, the unit uses Alibaba OCR interface to efficiently recognize text content in PDF files, supports multi-language recognition and layout analysis. It also has the ability to recognize charts, formula recognition and conversion, and document structure, providing users with one-stop PDF document processing services.

[0053] The user intention parsing unit is used to understand and parse the query requests input by the user, and extract the key entities and query intentions of the query requests; specifically, the system generates prompt words based on the query content input by the user, inputs them into a pre-trained large model based on the execution intention recognition task, performs intent analysis on the query content input by the user, extracts biological entities and identifies the core query targets, and generates logical chain instructions according to a certain method, aiming to guide the subsequent retrieval and information matching process.

[0054] In one embodiment, if Figure 2, the biological entity module comprises:

[0055] The biological entity library construction unit is used to integrate multiple data sources to obtain multi-source biomedical entity information to build a biological entity library; specifically, it integrates biomedical entity information from multiple authoritative biological entity databases, including but not limited to gene databases (such as NCBI Gene, Ensembl), protein databases (such as UniProt), disease databases (such as OMIM) and drug databases (such as DrugBank, ChEMBL). These data sources cover a variety of biological entity types, such as genes, proteins, diseases, drugs, molecular pathways, etc. Through the entity data integration function, the unit obtains, cleans, and standardizes biological entity information from multi-source data, such as NCBI Gene, Ensembl, UniProt, etc., and builds a biological entity library based on the entity background information obtained after processing.

[0056] The entity information retrieval unit is connected to the biological entity library construction module and is used to retrieve the corresponding entity background information from the biological entity library based on the extracted biological entity. Specifically, the unit will match the corresponding entity based on fuzzy search in the pre-built biological entity database, where the biological entity information library covers most diseases, drugs, proteins, and gene biological entity information. After the search is completed, detailed background information related to the entity will be returned, such as: the indications and clinical trial data of a certain drug; the function, mutation data, and correlation with the disease of a certain gene; the pathological characteristics and treatment plans of a certain disease, etc. This unit provides an efficient and accurate biological entity search engine that supports rapid retrieval and access to biological entity information in the database.

[0057] In one embodiment, if Figure 2 , the document abstract knowledge graph module 4 includes:

[0058] The document collection unit is used to dynamically update documents and extract their core information; this unit efficiently and accurately updates academic documents dynamically and extracts the core paragraphs and chart information of the documents.

[0059] A document analysis unit, connected to the document collection unit, is used to perform in-depth analysis and systematic summary of the core information of each document based on a large model to obtain a multi-dimensional document analysis result;

[0060] A knowledge graph construction unit, connected to the document parsing unit, is used to extract biological entities, relationship information and the evidence level of each entity relationship in the core information from the document parsing results based on the big model, and construct a document abstract knowledge graph database;

[0061] The document information retrieval unit is connected to the knowledge graph construction unit and is used to retrieve and match the corresponding relevant document information in the document abstract knowledge graph database through semantic similarity based on the large model, according to the extracted biological entities and the generated logical chain instructions, and can sort based on relevance, the number of citations of the document and the publication time.

[0062] In a specific embodiment, if Figure 3 , the document collection unit comprises:

[0063] The incremental crawler update subunit is used to crawl documents using an incremental crawler method based on the set crawling frequency and concurrency; specifically, the incremental crawler update subunit supports API interfaces that integrate multiple public document sources to obtain document metadata, including but not limited to databases such as PubMed and PubTator3. Among them, document metadata includes title, author, keyword, abstract, DOI number, PMID number, PMCID number, and website. By maintaining document index records and deduplicating the collected documents, it is possible to monitor the updates of the target database regularly or in real time and automatically identify new documents; and according to factors such as server load and network bandwidth, the frequency and concurrency of document crawling are dynamically adjusted to ensure that the incremental crawler method is used to efficiently collect data under legal conditions.

[0064] A document grouping and storage subunit, connected to the incremental crawler update subunit, for classifying and storing the captured documents according to fields;

[0065] A core information extraction subunit, connected to the document grouping and storage subunits, is used to perform adaptive structural analysis on the classified documents, process different journal page structures, and extract core paragraph information and charts;

[0066] The core paragraph information includes: Title, Highlight, Abstract, Introduction, Results, Figure and Table legends and titles, Discussion, and Conclusion. In addition to text information, the core information extraction subunit can also extract chart information, which is a screenshot of Figure and Table.

[0067] In one embodiment, if Figure 3 , the document parsing unit comprises:

[0068] The translation sub-unit is used to call the big model to perform the translation task and translate the core information of the extracted document into Chinese in context; if the document contains English content, the big model will automatically translate it into Chinese so that users can read and understand it.

[0069] The structured summary generation subunit is connected to the translation subunit and is used to process the core information of each document based on the framework parser to obtain the corresponding structured summary and chart analysis results; wherein the structured summary includes: research highlights, research summary, research questions, research methods, research results, research deficiencies and research conclusions. The structured summary is an information organization method that presents the key content of the document in a predefined format so that users can quickly grasp the main content of the document. It helps users quickly identify the value and relevance of the document, especially when conducting a literature review or quickly screening a large number of documents.

[0070] In one embodiment, based on the framework parser, the core information of each document is processed to obtain the corresponding structured abstract and chart parsing results including:

[0071] Based on the big model that performs the chart parsing task, the charts in the core information of the document are parsed to obtain the chart parsing results; specifically, Figure and Table screenshots are used as prompt words, and input into the multimodal big model to generate the explanation of the chart to obtain the chart parsing results;

[0072] The structured summary of the document is obtained based on the core paragraph information in the core information of the document and the results of chart analysis, and the methods include:

[0073] Use the entire Highlight paragraph of the document as the research highlight of the document;

[0074] Based on the large model for performing the literature summary task, the research summary of the literature is obtained according to the entire Abstract paragraph, the last paragraph of the Introduction, and the first paragraph of the Discussion.

[0075] Based on the large model that performs the literature summarization task, the research questions of the literature are generated according to all paragraphs of the literature introduction except the last paragraph;

[0076] Based on a large model that performs document summarization tasks, a research method for generating documents based on the legends and titles of the document's figures;

[0077] Based on the large model that performs the literature summary task, the research results of the literature are generated according to the title of the Results, the legend and title of the Figure, and the results of the chart analysis;

[0078] Based on the large model for performing document summarization tasks, there is insufficient research on generating documents based on all paragraphs of the document’s Discussion except the first paragraph;

[0079] Use the entire Conclusion paragraph in the document as the research conclusion of the document.

[0080] For example, Figure 4 The framework parser is used to perform structural analysis on the input HTML document, identify different parts in the document, extract its content and mark and classify it. The framework parser identifies the specific structure of the document by executing a large model of the document summary task. The processing methods based on the core information of each document include:

[0081] The framework parser directly marks the entire Highlight section of the document as "research highlights", which is very important for quickly understanding the core contributions of the research;

[0082] The framework parser combines the entire Abstract paragraph, the last paragraph of the Introduction, and the first paragraph of the Discussion, generates prompt words, inputs them into the large model that performs the document abstract task, and uses them as the research abstract of the document; the document abstract is a concise summary of the entire research. The framework parser extracts this part from the HTML of the document and marks it as "research abstract". This part usually includes the research background, objectives, methods, results, and conclusions, and is the key for readers to quickly understand the content of the document.

[0083] The framework parser combines all the paragraphs of the introduction of the document except the last paragraph, generates prompt words, inputs them into the large model that performs the document summary task to generate a summary, and uses it as the research question of the document; in the introduction part, the document usually discusses the background of the research, existing problems and the motivation of the research. The framework parser extracts the introduction part and marks it as "research question", which is the starting point of the document and helps understand why the author conducted the research.

[0084] The framework parser combines the legends and titles of the figures in the literature, generates prompt words, inputs them into the large model that performs the literature summary task to generate a summary, and uses it as the research method of the literature; the legends and titles of the figures in the literature describe in detail the methods and experimental steps used in the research. The framework parser extracts relevant content from the legends and titles of the figures and marks them as "research methods". This part is very critical for understanding how the research was carried out.

[0085] The frame parser combines the title of the Results, the legend and title of the Figure, and the parsing results of the charts and graphs of the document to generate prompt words, which are input into the large model that performs the document summary task to generate a summary and use it as the research result of the document; the title of the Results, the legend and title of the Figure, and the parsing results of the charts and graphs are usually used to display the research data and statistical results of the document. The frame parser extracts these contents from the HTML document and marks them as "research results" for subsequent analysis and display.

[0086] The framework parser generates prompt words from all paragraphs of the discussion section of the document except the first paragraph, and inputs them into the large model that performs the document summary task to generate a summary, and uses it as the research deficiencies of the generated document; the discussion section of the document usually mentions the limitations and deficiencies of the research, and the framework parser extracts these contents from the discussion section and marks them as "research deficiencies". This is the researcher's analysis of the limitations of the experiment and future improvement directions.

[0087] The framework parser takes the entire Conclusion section of the document as the research conclusion of the document. The Conclusion section is a summary and final conclusion of the research results. The framework parser will extract key content from the conclusion section of the document and mark it as "research conclusion" to summarize the final findings of the research.

[0088] In one embodiment, if Figure 3 , the knowledge graph construction unit includes:

[0089] The entity and relationship extraction subunit is used to extract biological entity and relationship information based on the core information of each document based on the big model that performs biological entity and relationship extraction tasks; and determine the level of evidence of each entity relationship based on the research methods in the structured abstract of each document. This step requires an evaluation of the reliability and validity of the research methods to determine the strength of evidence of the entity relationship, which is used to consider factors such as the rigor of the research design, sample size, and experimental repeatability; specifically, prompt words are formed based on the core information of each document, and the big model is used to identify biomedical entities, including but not limited to diseases, phenotypes, drugs, proteins, genes, side effects, etc.; the big model that performs biological entity and relationship extraction tasks is used to identify the semantic relationships between extracted entities, such as "treatment", "cause", "inhibition", "interaction", etc.; the research methods in the parsing results are used to provide the level of evidence of each entity relationship.

[0090] The knowledge graph generation subunit is connected to the entity and relationship extraction subunit, and is used to generate a literature abstract knowledge graph database based on the extracted biological entities, relationship information, and the evidence level of each entity relationship and the structured abstract of each document; specifically, the identified entity and relationship information and evidence level are converted into Json format, stored in the Neo4j database, and visualized as a knowledge graph. The knowledge graph will be stored as an independent data set to form a visual knowledge base. Through continuous updating and expansion, the knowledge graph will cover more biomedical entities and the complex relationships between them, providing users with rich reference materials. According to the structured abstracts of each document, they are stored in the literature abstract database of a specific field according to the field.

[0091] In one embodiment, the query result acquisition module is used to integrate the logical chain instructions, entity background information and related document information to generate prompt words, which are input into the large model to generate query results corresponding to the user's query request. The generated query results may be text, charts or other forms of data. These results directly correspond to the user's query request and are finally displayed through the user interaction module.

[0092] Similar to the principle of the above-mentioned embodiment, the present invention provides a medical literature intelligent processing and analysis system based on a large model.

[0093] The following provides specific embodiments in conjunction with the accompanying drawings:

[0094] like Figure 5 A flowchart showing a method for intelligent processing and analysis of medical literature based on a large model in an embodiment of the present invention is shown.

[0095] The method comprises:

[0096] Step S1: receiving a user's query request through a user search interface; wherein the types of the query request include: a document keyword query request or a document related question query request;

[0097] Step S2: Based on the big model, the user intent is identified according to the user's query request, and biological entities are extracted and logic chain instructions are generated;

[0098] Step S3: based on the extracted biological entity, searching for matching corresponding entity background information from a pre-built biological entity library;

[0099] Step S4: Based on the macro model, according to the extracted biological entities and the generated logic chain instructions, the corresponding relevant literature information is retrieved from the pre-built literature abstract knowledge graph database;

[0100] Step S5: Based on the big model, the logic chain instructions, entity background information and related document information are integrated to generate query results corresponding to the user's query request for display.

[0101] Since the implementation principle of the large model-based intelligent processing and analysis method of medical literature has been described in the aforementioned embodiments, it will not be repeated here.

[0102] In order to better describe the intelligent processing and analysis method of medical literature based on a large model, it is now explained in conjunction with the following specific embodiments.

[0103] Example 1: A method for intelligent processing and analysis of medical literature based on a large model. Figure 6 Schematic diagram of the process of intelligent processing and analysis of medical literature based on a large model in this embodiment.

[0104] The method comprises:

[0105] Step 1: First, obtain the query request input by the user through the user search interface. The user can enter keywords, questions or specific query content. This can be keywords related to biological entities such as drugs, diseases, proteins, genes, etc., or specific questions related to medical literature.

[0106] Step 2: Call the pre-trained large language model to perform intent analysis on the query content entered by the user. The model will parse the user's natural language input, identify the core query target, and combine it in a certain way to generate logical chain instructions to guide the subsequent retrieval and information matching process.

[0107] Step 3: According to the biological entities extracted in step 2, the corresponding entities are matched according to fuzzy search in the pre-constructed biological entity database, where the biological entity information database covers most diseases, drugs, proteins, and gene biological entity information. After the search is completed, detailed background information related to the entity will be returned, such as: the indications and clinical trial data of a certain drug; the function, mutation data, and correlation with the disease of a certain gene; the pathological characteristics and treatment plans of a certain disease, etc.

[0108] Step 4: The entities and logic chain instructions extracted in step 3 are matched in the field literature abstract knowledge graph database by semantic similarity matching. The knowledge graph database is a database constructed by semantic analysis and knowledge extraction of a large number of medical documents. Through semantic association, the structural abstracts and corresponding knowledge graph information of the relevant documents are returned. Based on dimensions such as relevance, number of citations of the documents, and publication time, the system will sort the query results and give priority to displaying the top 10 document information with high quality and high relevance.

[0109] Step 5: Integrate logical chain instructions, biological entity background information, literature abstracts and knowledge graphs to generate further prompt words, input them into the big model to output the query results and display the results to the user.

[0110] Embodiment 2: A method for constructing a document abstract knowledge graph database. Figure 7 The figure is a flowchart of the method for constructing the literature abstract knowledge graph database.

[0111] The method comprises:

[0112] Step 1: Download relevant document meta information from different databases or document resource platforms (such as Pubtator3), and use incremental crawler scripts to track public document databases in real time. The document meta information includes title, author, keyword, abstract, DOI number, PMID number, PMCID number, and website.

[0113] Step 2: Use the large language model pipeline to filter literature in specific biomedical fields based on the basic setting prompt words. Based on the needs of the user or system, the model automatically determines the specific fields described in the literature through semantic understanding analysis, such as tumors, autoimmune diseases, etc.

[0114] Step 3: For each document in the professional field screened, the system will access its corresponding website and obtain the HTML source code of the document page. The acquisition strategy is to establish API cooperation with major publishing houses and use a headless browser to automatically simulate the document page click operation to obtain the HTML source code of the page.

[0115] The obtained HTML source code is used to extract key paragraphs and chart information according to the Xpath parsing template based on different publishers, including: Title, Highlight, Abstract, Introduction, Results title, Figure, Table screenshots, legends and titles, Discussion, and Conclusion.

[0116] Step 4: Input the extracted information into the big model pipeline for in-depth analysis and generate a series of output content: If the document contains English content, the big model will automatically translate it into Chinese so that users can read and understand it. The big model will automatically generate concise document summaries based on different paragraphs, condense the research content, and facilitate users to quickly understand the main findings and contributions of the document. For charts in the document, the big model can generate explanations for the charts by calling the multimodal analysis in the big model pipeline to help users understand the data and research methods in the charts.

[0117] Step 5: The system identifies biomedical entities in the literature through the big model interface. These entities may include diseases, drugs, genes, proteins, molecular pathways, etc. The big model automatically extracts these key terms through semantic understanding. The system not only extracts individual entities, but also identifies the relationships between entities. For example, how a certain drug treats a certain disease, and how a certain gene mutation affects the occurrence of a certain disease. These relationships are the core of knowledge graph construction. Based on the extracted entities and relationships, the system begins to build a knowledge graph. The knowledge graph displays the connections between entities in the form of nodes and edges and can be visualized. Through the knowledge graph, users can intuitively understand the research content and its relevance in the literature.

[0118] Step 6: After the system generates the document translation, document abstract, and chart analysis, this information will be stored in a document abstract database for a specific field. The original text data of each document will be deleted from the database, and only the secondary abstract version of each document will be included to avoid leaking non-public document data. The knowledge graph will be stored as an independent data set to form a visual knowledge base. Through continuous updating and expansion, the knowledge graph will cover more biomedical entities and the complex relationships between them, providing users with rich reference materials. Document abstracts and knowledge graphs will be refined and classified according to fields to ensure that users can quickly find relevant research results in specific fields.

[0119] The medical literature intelligent processing and analysis system based on a large model provided in the embodiment of the present invention can be implemented on the terminal side or the server side. As for the hardware structure of the electronic terminal, please refer to Figure 8 , is an optional hardware structure diagram of an electronic terminal 1000 provided in an embodiment of the present invention. The terminal 1000 may be a mobile phone, a computer device, a tablet device, etc. The terminal 1000 includes: at least one processor 1001, a memory 1002, at least one network interface 10010 and a user interface 1009. The various components in the device are coupled together through a bus system 1005. It can be understood that the bus system 1005 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 1005 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 8 In the specification, various buses are labeled as bus systems.

[0120] The user interface 1009 may include a display, a keyboard, a mouse, keys, buttons, a touch panel or a touch screen.

[0121] It is understood that the memory 1002 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of exemplary but not limiting explanation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but is not limited to these and any other suitable categories of memory.

[0122] The memory 1002 in the embodiment of the present invention is used to store various categories of data to support the operation of the terminal 1000. Examples of these data include: any executable program for operating on the terminal 1000, such as an operating system 10021 and an application 10022; the operating system 10021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 10022 may include various applications, such as a media player (MediaPlayer), a browser (Browser), etc., for implementing various application services. The intelligent processing and analysis system for medical literature based on a large model provided in the embodiment of the present invention may be included in the application 10022.

[0123] The method disclosed in the above embodiment of the present invention can be applied to the processor 1001, or implemented by the processor 1001. The processor 1001 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 1001 or the instruction in the form of software. The above processor 1001 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 1001 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor 1001 can be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0124] In an exemplary embodiment, the terminal 1000 may be implemented by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD) to execute the aforementioned method.

[0125] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0126] In the embodiments provided in the present application, the computer readable and writable storage medium may include a read-only memory, a random access memory, an EEPROM, a CD-ROM or other optical disk storage device, a disk storage device or other magnetic storage device, a flash memory, a USB flash drive, a mobile hard disk, or any other medium that can be used to store a desired program code in the form of an instruction or data structure and can be accessed by a computer. In addition, any connection can be appropriately referred to as a computer-readable medium. For example, if the instruction is sent from a website, a server or other remote source using a coaxial cable, an optical fiber cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, optical fiber cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. However, it should be understood that computer readable and writable storage media and data storage media do not include connections, carriers, signals, or other temporary media, but are intended to be non-temporary, tangible storage media. Disk and disc, as used in this application, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.

[0127] In summary, the medical literature intelligent processing and analysis system, method and terminal based on the big model of the present invention first obtains the user's query request through the user search interface, then calls the big model to identify the user's intention, extracts the relevant biological entities, and generates the corresponding logical chain instructions; matches the background information of the relevant entities in the pre-built biological entity library according to the extracted biological entities; and matches the document abstract knowledge graph database according to the logical chain instructions to obtain the relevance ranking of the documents; finally integrates the logical chain instructions, biological entity background information and related document information to generate and display the corresponding query results. The present invention solves the problems faced by medical researchers, such as cumbersome document acquisition, time-consuming reading, and complex analysis. It uses a method based on a big model to perform intelligent document processing and analysis, accelerate the document reading process, reduce the cost of manual analysis, and improve research efficiency. Therefore, the present invention effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0128] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A medical literature intelligent processing and analysis system based on a large model, characterized in that: The system comprises: A user interaction module is used to receive a user's query request through a user search interface; wherein the types of the query request include: a document keyword query request or a document related question query request; A user intention recognition module, connected to the user interaction module, for performing user intention recognition based on the large model according to the user's query request, extracting biological entities and generating logic chain instructions; A biological entity module, connected to the user intention recognition module, is used to retrieve matching corresponding entity background information from a pre-built biological entity library based on the extracted biological entity; A document abstract knowledge graph module is connected to the biological entity module and is used to retrieve corresponding relevant document information from a pre-built document abstract knowledge graph database based on the macro model and the extracted biological entities and the generated logic chain instructions; The query result acquisition module connects the user interaction module, the user intention recognition module, the biological entity module and the literature abstract knowledge graph module, and is used to integrate the logic chain instructions, entity background information and related literature information based on the big model to generate the query results corresponding to the user's query request and display them by the user interaction module.

2. The medical literature intelligent processing and analysis system based on a large model according to claim 1, characterized in that: The user interaction module comprises: A user search interface is used to receive keywords of documents to be searched or questions related to documents input by users; User guidance and prompt interface, used to show the main functions and usage of the system to first-time users; The information display interface includes: data statistics display area, graph visualization display area and intelligent dialogue area, which are used to display the literature query results for literature keyword queries or the answer results for question feedback.

3. The medical literature intelligent processing and analysis system based on a large model according to claim 1, characterized in that: The biological entity module comprises: A biological entity library construction unit is used to integrate multiple data sources to obtain multi-source biomedical entity information to construct a biological entity library; The entity information retrieval unit is connected to the biological entity library construction unit and is used to retrieve matching corresponding entity background information from the biological entity library based on the extracted biological entity.

4. The medical literature intelligent processing and analysis system based on a large model according to claim 1, characterized in that: The document abstract knowledge graph module includes: Document collection unit, used to dynamically update documents and extract core information of documents; A document analysis unit, connected to the document collection unit, is used to perform in-depth analysis and systematic summary of the core information of each document based on a large model to obtain multi-dimensional document analysis results; A knowledge graph construction unit, connected to the document parsing unit, is used to extract biological entities, relationship information and the evidence level of each entity relationship in the core information from the document parsing results based on the big model, and construct a document abstract knowledge graph database; The document information retrieval unit is connected to the knowledge graph construction unit and is used to retrieve and match the corresponding relevant document information in the document abstract knowledge graph database through semantic similarity based on the large model, according to the extracted biological entities and the generated logical chain instructions, and can sort based on relevance, the number of citations of the document and the publication time.

5. The medical literature intelligent processing and analysis system based on a large model according to claim 4 is characterized in that: The document collection unit comprises: The incremental crawler update subunit is used to crawl documents using an incremental crawler method based on the set crawling frequency and concurrency; A document grouping and storage subunit, connected to the incremental crawler update subunit, for classifying and storing the captured documents according to fields; The core information extraction subunit is connected to the document grouping and storage subunits, and is used to perform adaptive structural analysis on the classified documents and extract core paragraph information and charts; wherein the core paragraph information includes: Title, Highlight, Abstract, Introduction, Results, legends and titles of Figures and Tables, Discussion, and Conclusion.

6. The medical literature intelligent processing and analysis system based on a large model according to claim 5, characterized in that: The document parsing unit comprises: The translation subunit is used to call the large model that performs the translation task and translate the core information of the extracted document into Chinese in context; The structured summary generation subunit is connected to the translation subunit and is used to process the core information of each document based on the framework parser to obtain the corresponding structured summary and chart analysis results; wherein the structured summary includes: research highlights, research summary, research questions, research methods, research results, research deficiencies and research conclusions.

7. The medical literature intelligent processing and analysis system based on a large model according to claim 6, characterized in that: The knowledge graph construction unit includes: The entity and relationship extraction subunit is used to extract biological entity and relationship information based on the core information of each document based on the large model that performs biological entity and relationship extraction tasks, and determine the evidence level of each entity relationship based on the research method in the structured summary of each document; The knowledge graph generation subunit is connected to the entity and relationship extraction subunit, and is used to generate a document abstract knowledge graph database based on the extracted biological entities, relationship information and the evidence level of each entity relationship and the structured abstract of each document.

8. The medical literature intelligent processing and analysis system based on a large model according to claim 6, characterized in that: Based on the framework parser, the core information of each document is processed to obtain the corresponding structured abstract and chart analysis results, including: Based on the big model that performs the chart parsing task, the charts in the core information of the document are parsed to obtain the chart parsing results; The structured summary of the document is obtained based on the core paragraph information in the core information of the document and the results of chart analysis, and the methods include: Use the entire Highlight paragraph of the document as the research highlight of the document; Based on the large model for performing the literature summary task, the research summary of the literature is obtained according to the entire Abstract paragraph, the last paragraph of the Introduction, and the first paragraph of the Discussion. Based on the large model that performs the literature summarization task, the research questions of the literature are generated according to all paragraphs of the literature introduction except the last paragraph; Based on a large model that performs document summarization tasks, a research method for generating documents based on the legends and titles of the document's figures; Based on the large model that performs the literature summary task, the research results of the literature are generated according to the title of the Results, the legend and title of the Figure, and the results of the chart analysis; Based on the large model for performing document summarization tasks, there is insufficient research on generating documents based on all paragraphs of the document’s Discussion except the first paragraph; Use the entire Conclusion paragraph in the document as the research conclusion of the document.

9. A method for intelligent processing and analysis of medical literature based on a large model, characterized in that: The method comprises: Receiving a user's query request through a user search interface; wherein the types of the query request include: a document keyword query request or a document related question query request; Based on the big model, the user intent is identified according to the user's query request, and biological entities are extracted and logical chain instructions are generated; Based on the extracted biological entities, matching corresponding entity background information is retrieved from a pre-built biological entity library; Based on the big model, according to the extracted biological entities and the generated logical chain instructions, the corresponding relevant literature information is retrieved from the pre-built literature abstract knowledge graph database; Based on the big model, the logic chain instructions, entity background information and related document information are integrated to generate query results corresponding to the user's query request and display them.

10. An electronic terminal, characterized in that: include: one or more memories and one or more processors; The one or more memories are used to store computer programs; The one or more processors, connected to the memory, are configured to run the computer program to perform the method according to claim 9.

Images