Search

CN-121980026-A - All-weather system overview automatic generation and updating method and system

CN121980026ACN 121980026 ACN121980026 ACN 121980026ACN-121980026-A

Abstract

The invention relates to the technical field of computer-aided scientific research and discloses an all-weather automatic system review generation and updating method and system, wherein the method comprises the steps of automatically scanning all-weather data of pre-recorded academic documents, identifying system reviews which are lack or need to be updated, generating corresponding research topic descriptions, expanding and generating a multi-database retrieval strategy set; searching a plurality of academic databases to obtain multi-source data of candidate documents, carrying out standardized cleaning and de-duplication treatment on the obtained multi-source data to form a unified candidate document set, inputting document titles and summaries in the candidate document set into a trained screening model, automatically outputting judgment results and reasons of the documents, inputting the set formed by the documents with the judgment results being the incorporated into an artificial intelligent writing model, and structurally generating a system review manuscript, wherein statements in the manuscript are related to corresponding document sources, so that the full-flow automation from intelligent searching and automatic screening to structural manuscript forming of the system review is realized.

Inventors

  • LI JIAJUN
  • LI JIAZHENG
  • LI JIAHUI

Assignees

  • 佳美惠通科技(深圳)有限公司

Dates

Publication Date
20260505
Application Date
20260205

Claims (14)

  1. 1. An all-weather system overview automatic generation and updating method is characterized by comprising the following steps: S1, pre-entering locally stored academic literature data through all-weather automatic scanning and analysis of a large language model, identifying a lack or system review which needs to be updated, automatically generating corresponding research topic description, and expanding a multi-database retrieval strategy set which covers synonyms and related words; S2, searching a plurality of academic databases by using the search strategy set to obtain metadata and abstract texts of candidate documents, and carrying out standardized cleaning and duplicate removal processing on the obtained multi-source data to form a unified candidate document set; S3, inputting the document titles and abstracts in the candidate document set into a trained artificial intelligent screening model, carrying out automatic document screening, and automatically outputting the inclusion, exclusion or uncertain judgment result and reason of each document; S4, inputting the set formed by the documents judged to be included into an artificial intelligent writing model, and structurally generating a system review manuscript comprising methodologies, results and discussion parts, wherein the statements in the manuscript are all related to corresponding document sources.
  2. 2. The method according to claim 1, wherein in step S1, the study topic description further includes manually entered study questions, and the expanding the set of multi-database search strategies to generate coverage synonyms and related words includes: s11, receiving research topic description input in a natural language or PICOS structural framework, inputting the research topic description into a large language model, and generating candidate synonym and paraphrasing sets related to core concepts; S12, carrying out semantic relevance evaluation on the candidate synonyms and the paraphrasing, screening out words with relevance scores higher than a preset threshold value, and forming an expanded word set, wherein the relevance scores are obtained by calculating cosine similarity of the candidate words and the core concept words in a model semantic space; And S13, combining the core concept words with the extended word sets according to the query grammar rules of the target database, constructing Boolean search formulas applicable to different databases, and forming the search strategy set.
  3. 3. The method for automatically generating and updating an all-weather system overview as claimed in claim 2, wherein step S1 further comprises performing adaptive search optimization: performing a preliminary search and obtaining the number of returned documents According to Automatically adjusting a search strategy in a falling interval: If it is The search conditions are relaxed, if The search conditions are tightened, wherein, And The adjustment includes modification to logical operators, field definitions, or time ranges for a preset document quantity threshold.
  4. 4. The method for automatically generating and updating an all-weather system overview according to claim 1, wherein in step S2, the performing standardized cleaning and deduplication on the acquired multi-source data specifically includes: Normalization is achieved by field mapping: Uniformly mapping the document identifiers, authors and journal name fields from different databases to an internal general format, and preferentially adopting the global general identifier as a document main key; Performing exact deduplication based on the global universal identifier: And calculating the similarity of the title text of the record without the universal identifier, and judging the record as a repeated document when the similarity value is higher than a preset threshold value, wherein the similarity is calculated through a Jaccard similarity coefficient or SimHash algorithm.
  5. 5. The method for automatically generating and updating an all-weather system overview as claimed in claim 1, wherein in step S3, said training of the artificial intelligence screening model comprises: the model performs end-to-end training through massive published system reviews and corresponding inclusion/exclusion document sets, so that complex mapping relations between research problems, document texts and screening decisions are learned, and implicit screening standards are formed by internalization instead of rules relying on artificial plaintext coding.
  6. 6. The method for automatically generating and updating an all-weather system overview according to claim 1, wherein in step S3, said automated document screening employs a multimodal consensus mechanism, comprising: And respectively inputting titles and abstracts of the same document into a plurality of different screening models or different examples of the same model, synthesizing judgment results of the models, adopting the consensus judgment when the judgment results are that the number of the included or excluded tickets exceeds a preset proportion, and otherwise marking the judgment results as uncertain.
  7. 7. The method for automatically generating and updating all-weather system overview according to claim 1, wherein in step S4, the generation of the system overview manuscript specifically includes: s41, automatically extracting key elements from the literature with the judging result being inclusion, including research design, crowd, contrast, intervention and ending, and generating a structured data table; S42, generating a manuscript text based on the research feature data by the artificial intelligence authoring model, and ensuring that all statements requiring evidence support are associated with corresponding literature sources, wherein the process is realized by any one or combination of the following modes: for specific claims in the text, searching from a document set which is judged to be included or a preset extended document library or an authoritative information source which is accessed through an Internet public interface automatically, matching supporting documents for the claims, and if the supporting documents cannot be matched, automatically adjusting or deleting the claims and regenerating the text; directly generating corresponding summarized, analytical or comparative paragraphs based on the determination result as specific evidence in the incorporated document collection; And S43, after a draft text is generated by the main writing model, at least one verification model is used for verifying the accuracy and logic consistency of key data and conclusions in the text and the references of corresponding documents, so that each data statement or medical definition in the draft is ensured to have document support.
  8. 8. The all-weather system overview automatic generation and update method of claim 7, wherein said system overview manuscript generation further comprises: Automatically generating a structured study feature data table, and a system overview flowchart conforming to PRISMA report specifications, based on the determination as to the set of incorporated documents; Selecting whether related scientific research schemes need to be generated according to the generated results in the system overview, and automatically generating the scientific research schemes if the related scientific research schemes need to be generated; The system overview manuscript, data sheet, and flowchart are output in an editable document format.
  9. 9. The method for automatically generating and updating an all-weather system overview as claimed in claim 8, wherein said generating of the system overview manuscript further comprises automatically adjusting and formatting the structure of the generated system overview manuscript, specifically comprising: Reading a target format specification, analyzing requirements on a document structure, a quotation pattern, a chart title and typesetting details through natural language processing, and generating a structured format instruction; analyzing the system overview manuscript, identifying functional paragraphs, quotation marks and chart elements therein, and mapping the content elements into a target chapter frame specified by the structured format instruction; Automatically reorganizing the manuscript content according to the mapping relation, adjusting the chapter sequence, and uniformly converting the quotation patterns, the chart titles and the text formats into patterns required by target specifications; a system overview document is generated and output that fully complies with the target format specification.
  10. 10. The all-weather system overview automatic generation and update method of claim 1, wherein the method operates in an all-weather mode and supports continuous update of dynamic system overview, further comprising: Steps S1 to S4 are automatically and repeatedly executed according to a preset period, but only the document newly published or updated since the last retrieval is retrieved in step S2; the complete log of each execution is recorded, including the search strategy used, the document ID retrieved, the screening decision flow, the manuscript version differences, and an auditable report is generated.
  11. 11. The method for automatically generating and updating all-weather system overview according to claim 1, wherein in step S2, the full text of the candidate document is further acquired and stored through an automation interface on the basis of acquiring document metadata and abstract text; When the models in the steps S3 and S4 are processed, the available full text information is preferentially used so as to improve the accuracy of decision making and generation.
  12. 12. An all-weather system overview automatic generation and update system, comprising: The intelligent search strategy generation module is used for automatically identifying core research concepts through an artificial intelligent model based on the input research topic description and expanding and generating a multi-database search strategy set covering synonyms and related words; The multi-source document acquisition and processing module is used for searching a plurality of academic databases by utilizing the search strategy set to acquire metadata and abstract text of candidate documents, and carrying out standardized cleaning and duplicate removal processing on the acquired multi-source data to form a unified candidate document set; The automatic document screening module is used for inputting the document titles and abstracts in the candidate document set into the trained artificial intelligent screening model, carrying out automatic document screening and automatically outputting the inclusion, exclusion or uncertain judgment result and reason of each document; And the system review content generation module is used for inputting the set formed by the documents with the judging results into the artificial intelligent writing model and structurally generating a system review manuscript comprising methodologies, results and discussion parts, wherein the statement in the manuscript is related to the corresponding document source.
  13. 13. A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the all-weather system overview automatic generation and update method of any of claims 1-11.
  14. 14. An electronic device comprising one or more processors and storage means for storing one or more programs that when executed by the one or more processors cause the one or more processors to implement the method for automatically generating and updating an all-weather system overview as claimed in any one of claims 1 to 11.

Description

All-weather system overview automatic generation and updating method and system Technical Field The invention relates to the technical field of computer-aided scientific research, in particular to an all-weather system review automatic generation and updating method and system. Background The system overview is taken as a highest-level evidence-based research method and is important for integrating evidence and guiding practice in the fields of medicine, public health, social science and the like. The core value is that all existing researches under a certain topic are comprehensively searched, strictly screened and scientifically integrated through a systematic, transparent and repeatable method. However, traditional system overview fabrication relies entirely on manual labor, an extremely complex, time-consuming and resource-intensive process, and there are primarily significant bottlenecks in the efficiency and consistency of document processing. With the exponential increase of the amount of academic database documents, manual retrieval and screening of massive documents are huge in consumption, and errors are easily generated due to subjective fatigue, so that the objectivity and reproducibility of results are affected. Second, policy complexity issues across database retrieval. Query grammar and term hierarchy for different databases (e.g., pubMed, scopus, woS) differ significantly. Researchers need to manually construct a search type for each library, the process is complicated, policy equivalence is difficult to ensure, and important documents are easy to miss. Again, there is a persistent dilemma of evidence updating. Traditional reviews have long production cycles and conclusions may be outdated at the time of publication. While "dynamic system reviews" are presented to continually update evidence, they require periodic literature monitoring, a burden that is difficult to maintain for personnel. In recent years, the development of natural language processing and artificial intelligence technology has provided possibilities for automation. Existing exploration focuses on a single link, e.g., using machine learning to assist in document classification or search construction. However, these techniques are mostly stand alone, fragmented solutions, and end-to-end intelligent workflows have not been formed yet. They generally lack the ability to adapt to multi-database heterogeneity, and cannot go through the whole process from understanding problems, generating policies, to synthesizing reports of composition, making it more difficult to support the automated, sustainable update systems required for "dynamic systems reviews". Therefore, a highly integrated artificial intelligence system is needed in the art, which can understand complex research problems, intelligently adapt to a multi-source database, automatically execute full-chain tasks from retrieval and screening to evidence synthesis, and realize continuous updating, so that the efficiency, timeliness and reliability of system review are remarkably improved. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a computer-implemented method and a system for generating and updating a system overview (SYSTEMATIC REVIEW) by utilizing an artificial intelligence model for assistance or automation, and a full-automatic end-to-end intelligent platform is constructed, and the core of the system is that the full-chain work from research problem analysis to overview text generation in the system overview is automatically and structurally processed and integrated through the artificial intelligence model. In one aspect, an all-weather system overview automatic generation and update method is provided, including the steps of: s1, pre-entering locally stored academic literature data through all-weather automatic scanning and analysis of a large language model, identifying a lack or system review which needs to be updated, automatically generating corresponding research topic description, and expanding and generating a multi-database retrieval strategy set covering synonyms and related words; S2, searching a plurality of academic databases by using the search strategy set to obtain metadata and abstract texts of candidate documents, and carrying out standardized cleaning and duplicate removal processing on the obtained multi-source data to form a unified candidate document set; S3, inputting the document titles and abstracts in the candidate document set into a trained artificial intelligent screening model, carrying out automatic document screening, and automatically outputting the inclusion, exclusion or uncertain judgment result and reason of each document; S4, inputting the set formed by the documents judged to be included into an artificial intelligent writing model, and structurally generating a system review manuscript comprising methodologies, results and discussion parts, wherein the statements in the manuscri