Literature screening method, device, electronic device and storage medium

Abstract

The present invention provides a document screening method, device, electronic device and storage medium. The method comprises the following steps: extracting all document data related to a screening topic from a document library and downloading the full text of the document data and its corresponding abstract; generating prompt words according to the screening topic, inputting the abstract and the prompt words corresponding to the document data into an intelligent screening model, judging whether the document data meets PICO requirements through the intelligent screening model, and outputting a first screening result; inputting the prompt words and the full text of the document data into the intelligent screening model, judging whether the document data meets PICO requirements through the intelligent screening model, and outputting a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities; performing cross-validation based on the first and second screening results, and using the large language model to screen documents that meet PICO requirements or are related to the current topic, thereby improving the accuracy and efficiency of document screening.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a document screening method, device, electronic equipment and storage medium. Background Art

[0002] Medical literature is an important reference for basic medical theory and clinical research. Selecting the required literature from the vast literature is a crucial task. Literature screening methods based on intelligent literature screening models can improve literature screening efficiency. Traditional intelligent literature screening models focus on removing duplicate articles and then screening using basic keywords. These models lack the precise screening required by specific research questions, such as those based on the PICO (P-subject, I-intervention, C-control, O-outcome) criteria, and are prone to missing important articles or including irrelevant ones. Summary of the Invention

[0003] The present invention provides a document screening method, device, electronic device and storage medium, which are used to solve the defects of traditional document screening methods that are prone to omitting important documents or including irrelevant documents.

[0004] The present invention provides a document screening method, comprising:

[0005] Extract all literature data related to the screening topic from the literature library, and download the full text of the literature data and its corresponding abstract;

[0006] Generate prompt words according to the screening theme, input the abstract part and the prompt words corresponding to the document data into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a first screening result;

[0007] Inputting the prompt word and the full text of the document data into the intelligent screening model, determining whether the document data meets the PICO requirements through the intelligent screening model, and outputting a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities;

[0008] A cross-validation is performed based on the first screening result and the second screening result to obtain a final screening result.

[0009] According to the literature screening method provided by the present invention, the screening subject is vaccine, and determining whether the literature data meets the PICO requirements by the intelligent screening model includes:

[0010] Outputting evaluation details of each document through the intelligent screening model;

[0011] Determine, based on the evaluation details, whether the literature research is related to the vaccine, whether it includes at least one of safety research, vaccine application targets, vaccine use, efficacy research, immunogenicity research, vaccination rate research, immunization program policy research, vaccine development, economic evaluation, literature type, clinical research type, and secondary research type related to the vaccine;

[0012] If at least one of the above conditions is met, the literature data is determined to meet the PICO requirements.

[0013] According to the document screening method provided by the present invention, generating prompt words according to the screening topics includes:

[0014] Preset multiple vaccine-related prompt words, input the prompt words into a first large language model, and fine-tune the multiple prompt words according to output results of the first large language model to generate new prompt words;

[0015] Input the new prompt word into the second largest language model, and adjust the temperature of the output result of the second largest language model according to the output result of the first largest language model;

[0016] Use a script to compare the output of the first language model and the output of the second language model, and obtain a final prompt word when the output of the first language model and the output of the second language model are consistent;

[0017] The first language model and the second language model are the same or different models.

[0018] According to the document screening method provided by the present invention, after outputting the first screening result, the method further includes:

[0019] Extracting some documents from the first screening results, verifying the some documents, and if the verification pass rate of the some documents does not reach a preset threshold, screening multiple documents from all document data related to the screening topic;

[0020] The prompt word, the abstract part corresponding to the document included in the first screening result, and the abstract parts corresponding to multiple documents screened from all document data related to the screening topic are re-input into the intelligent screening model to update the first screening result.

[0021] According to the document screening method provided by the present invention, the step of inputting the prompt word and the full text of the document data into the intelligent screening model comprises:

[0022] Obtaining text data of the full text of the document data;

[0023] Converting the text data into a text vector;

[0024] Obtaining relevant sentences and paragraphs based on the distance between the text vectors;

[0025] The relevant sentences and paragraphs are input into the intelligent screening model.

[0026] According to the document screening method provided by the present invention, extracting all document data related to the screening topic from the document library includes:

[0027] Acquire multiple document data from multiple document libraries; perform deduplication preprocessing on the document data;

[0028] All literature data related to the screening topic are extracted from the deduplicated literature through keyword technology.

[0029] The document screening method provided by the present invention further comprises:

[0030] The final screening results are classified and sorted by the intelligent screening model to obtain multiple types of literature data.

[0031] The present invention also provides a document screening device, comprising:

[0032] The extraction module is used to extract all document data related to the screening topic from the document library and download the full text of the document data and its corresponding abstract;

[0033] a first output module, configured to generate prompt words according to the screening theme, input the abstract portion corresponding to the document data and the prompt words into an intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a first screening result;

[0034] a second output module, configured to input the prompt word and the full text of the document data into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities;

[0035] The verification module is used to perform cross-validation based on the first screening result and the second screening result to obtain a final screening result.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the document screening method as described in any one of the above items when executing the program.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the above-mentioned document screening methods.

[0038] The document screening method, device, electronic device, and storage medium provided by the present invention extract all document data related to a screening topic from a document library and download the full text of the document data and its corresponding abstract; generate prompt words based on the screening topic, input the abstract corresponding to the document data and the prompt words into an intelligent screening model, use the intelligent screening model to determine whether the document data meets PICO requirements, and output a first screening result; input the prompt words and the full text of the document data into the intelligent screening model, use the intelligent screening model to determine whether the document data meets PICO requirements, and output a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities; cross-validate the first and second screening results to obtain a final screening result, and use the large language model to screen documents that meet PICO requirements or are related to the current topic, thereby improving the accuracy and efficiency of document screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 Schematic diagram of the process of the literature screening method provided by the embodiment of the present invention;

[0041] Figure 2 Schematic diagram of the functional structure of a document screening device provided by an embodiment of the present invention;

[0042] Figure 3 It is a functional structure diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0044] Figure 1A flowchart of the literature screening method provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the literature screening method provided by the embodiment of the present invention includes:

[0045] Step 101: extract all document data related to the screening topic from the document library, and download the full text of the document data and its corresponding abstract;

[0046] In an embodiment of the present invention, extracting all document data related to the screening topic from the document library includes:

[0047] Acquire multiple document data from multiple document libraries; perform deduplication preprocessing on the document data;

[0048] All literature data related to the screening topic are extracted from the deduplicated literature through keyword technology.

[0049] For example, using keyword search technology, all vaccine-related literature can be automatically extracted from online evidence-based medicine databases such as PubMed and Embase, and the abstract part can be automatically downloaded.

[0050] Step 102: Generate prompt words based on the screening theme, input the abstract part of the document data and the prompt words into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a first screening result;

[0051] Step 103: Input the prompt word and the full text of the document data into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities;

[0052] Step 104: perform cross-validation based on the first screening result and the second screening result to obtain a final screening result.

[0053] The literature screening model used in traditional literature screening methods focuses on removing duplicate literature and then using basic keywords for screening. It lacks the precise screening required by specific research questions such as the PICO (P-subject, I-intervention, C-control, O-outcome indicator) criteria, and is prone to missing important literature or including irrelevant literature.

[0054] The document screening method provided by an embodiment of the present invention extracts all document data related to a screening topic from a document library and downloads the full text of the document data and its corresponding abstract. A prompt word is generated based on the screening topic, and the abstract and the prompt word corresponding to the document data are input into an intelligent screening model. The intelligent screening model determines whether the document data meets PICO requirements and outputs a first screening result. The prompt word and the full text of the document data are input into the intelligent screening model, and the intelligent screening model determines whether the document data meets PICO requirements and outputs a second screening result. The intelligent screening model is one or more large language models with natural language understanding capabilities. Cross-validation is performed based on the first and second screening results to obtain a final screening result. The large language model is used to screen documents that meet PICO requirements or are related to the current topic, thereby improving the accuracy and efficiency of document screening.

[0055] Based on any of the above embodiments, the screening topic is vaccine, and determining whether the literature data meets the PICO requirements by the intelligent screening model includes:

[0056] Step 201: Output evaluation details of each document through the intelligent screening model;

[0057] Step 202: Determine, based on the evaluation details, whether the literature research is vaccine-related, whether it includes at least one of safety research, vaccine application targets, vaccine use, efficacy research, immunogenicity research, vaccination coverage research, immunization program policy research, vaccine development, economic evaluation, literature type, clinical research type, and secondary research type related to vaccines;

[0058] Step 203: If at least one of the above conditions is met, it is determined that the literature data meets the PICO requirements.

[0059] In this embodiment of the present invention, the Large Language Model (LLM) possesses powerful natural language understanding capabilities, enabling it to accurately understand the semantics and contextual relationships within a document. By performing semantic analysis on the document's content, the LLM can determine whether the document meets PICO requirements, such as whether the research is vaccine-related, the intended recipients of the vaccine, and the intended use of the vaccine.

[0060] Based on any of the above embodiments, generating prompt words according to the screening topic includes:

[0061] Step 301: Preset multiple vaccine-related prompt words, input the prompt words into a first large language model, and fine-tune the multiple prompt words according to the output of the first large language model to generate new prompt words;

[0062] Step 302: Input the new prompt word into the second largest language model, and adjust the temperature of the output result of the second largest language model according to the output result of the first largest language model;

[0063] Step 303: Use a script to compare the output result of the first language model and the output result of the second language model. When the output result of the first language model and the output result of the second language model are consistent, obtain a final prompt word;

[0064] The first language model and the second language model are the same or different models.

[0065] In the present embodiment, prompt words include, for example, the age of the subject of interest and the corresponding annotated data, the vaccine recipient and the corresponding annotated data, etc. By using LLM to build an automated screening process, large amounts of literature can be processed quickly, avoiding the inefficiency and error-proneness of manual screening and improving overall work efficiency.

[0066] Based on any of the above embodiments, after outputting the first screening result, the method further includes:

[0067] Step 401: extract some documents from the first screening results, verify the documents, and if the verification pass rate of the documents does not reach a preset threshold, screen multiple documents from all document data related to the screening topic;

[0068] Step 402: re-input the prompt word, the abstract parts corresponding to the documents included in the first screening result, and the abstract parts corresponding to multiple documents selected from all document data related to the screening topic into the intelligent screening model to update the first screening result.

[0069] The existing technology lacks targeted screening and is prone to missing important documents or including irrelevant documents. When the first screening results do not meet the requirements, the present invention re-screens from all document data related to the screening topic, which can introduce more documents, avoid omissions, and further improve screening accuracy.

[0070] Based on any of the above embodiments, inputting the prompt word and the full text of the document data into the intelligent screening model includes:

[0071] Step 501: Obtain the text data of the full text of the document data;

[0072] Step 502: convert the text data into a text vector;

[0073] Step 503: Obtain relevant sentences and paragraphs based on the distances between the text vectors;

[0074] Step 504: Input the relevant sentences and paragraphs into the intelligent screening model.

[0075] In this embodiment of the present invention, the Retrieval Augmented Generation (RAG) architecture is applied. The RAG architecture operates through the following steps: indexing, retrieval, and generation. During the indexing phase, documents are segmented into small chunks, encoded into vector form, and then stored in a vector database. This process involves segmenting document content into small chunks and mapping these document chunks into a high-dimensional vector space using an encoder (such as a pre-trained language model like BERT or RoBERTa). This allows each document to be quickly retrieved based on its vector representation. During the retrieval phase, a query is input into a query encoder, which also maps the query into a vector space. The most relevant documents are then found by calculating the similarity between the query vector and all stored document vectors. This process utilizes the Dual-Encoder architecture, which consists of two independent encoders: one for encoding the query and the other for encoding the document. By calculating the similarity between them, efficient retrieval can be achieved. The generation phase inputs the original question and the retrieved relevant documents into a large language model to generate the final answer. This step utilizes the power of the generator, which integrates and transforms the retrieved information into coherent and fluent text. The generator is guided by the retrieved text to ensure consistency between the generated text and the retrieved information. This comprehensive input enables the generator to deeply understand the context of the question, thereby providing more informative and contextual responses.

[0076] The RAG architecture combines retrieval and generation processes, using a retrieval model to find the most relevant documents for a query and a generative model to perform in-depth analysis of these documents. This allows for more accurate and relevant responses when processing natural language tasks. The RAG architecture efficiently processes large amounts of document data and dynamically generates screening criteria and results during the screening process. Therefore, the intelligent screening model, implemented using the RAG architecture, uses the full text of the document data as input, enabling rapid extraction of useful information from this vast amount of information, further improving screening efficiency.

[0077] Traditional intelligent screening models, such as the XGBoost algorithm, are complex to build and optimize. The RAG architecture, by combining retrieval and generation, simplifies the model building process while improving screening efficiency. The prompt word engineering in this embodiment of the present invention can guide LLM to better understand the content of document abstracts, while RAG technology can incorporate external full-text document search results. By cross-validating screening results based on abstract content with those based on the full text of the document, the accuracy of the screening results can be further verified and improved.

[0078] Based on any of the above embodiments, the document screening method further includes:

[0079] The final screening results are classified and sorted by the intelligent screening model to obtain multiple types of literature data.

[0080] In some embodiments of the present invention, manual sampling verification is also included to ensure the accuracy of the screening model.

[0081] In an embodiment of the present invention, the screened literature data includes evaluations based on different screening criteria, and the literature that meets the requirements is classified and organized to generate a detailed report for use in subsequent evidence-based medicine research and clinical practice.

[0082] This embodiment of the present invention is to screen a batch of literature on the new coronavirus vaccine (COVID-19 vaccine), with the goal of finding all research literature related to the effectiveness, safety, immunogenicity, vaccination rate, etc. of the new coronavirus vaccine.

[0083] (1) Data preprocessing: 5,000 relevant articles were retrieved from the PubMed database using the keyword “COVID-19 vaccine” and the abstracts were downloaded. These articles were preprocessed to remove duplicates, resulting in 4,500 independent articles.

[0084] (2) Initial keyword screening: A keyword screening model was used to conduct a preliminary screening of the literature, filtering out literature not related to vaccines and retaining 2,500 potentially relevant literature.

[0085] (3) Abstract screening using a large language model: 2,500 literature abstracts were input into the large language model, and the input was optimized through prompt word engineering to guide the model to perform deep semantic analysis to determine whether each literature meets the following criteria: whether the research is related to vaccines, the target population of vaccines (such as adults, children, and the elderly), the purpose of vaccines (prevention, treatment, etc.), whether it includes safety studies, whether it includes efficacy studies, whether it includes immunogenicity studies, whether it includes vaccination rate studies, whether it includes immunization program policy studies, whether it includes vaccine development, whether it includes economic evaluation, literature type (such as RCT, Meta-analysis, etc.), clinical research type, secondary research type, and 1,500 literature that meets the criteria was screened out.

[0086] (4) Full-text screening using a large language model: The full texts of 2,500 documents were input into the large language model. The input was optimized through prompt word engineering, guiding the model to perform deep semantic analysis to determine whether each document met the following criteria: whether the research is related to vaccines, the target population of vaccines (such as adults, children, and the elderly), the purpose of vaccines (prevention, treatment, etc.), whether safety studies are included, whether efficacy studies are included, whether immunogenicity studies are included, whether vaccination rate studies are included, whether immunization program policy studies are included, whether vaccine development is included, whether economic evaluation is included, document type (such as RCT, Meta-analysis, etc.), clinical research type, and secondary research type. 1,600 documents that met the criteria were screened out.

[0087] (5) Cross-validation: compare the 1,500 documents that meet the criteria screened out through abstracts with the 1,600 documents that meet the criteria screened out through full-text screening, output the screening results, and obtain 1,000 documents;

[0088] (6) Classify and organize the evaluation of the 1,000 documents based on different screening criteria.

[0089] (7) Manual sampling verification was conducted, and 100 articles were randomly selected from 1,000 articles for manual review to ensure the accuracy of the screening model. The verification results showed that the model accuracy reached 95%.

[0090] (8) Data output and application: Generate a detailed report based on the screening results, which includes the evaluation details and classification results of each document. The screening results will be applied to subsequent COVID-19 vaccine research and clinical practice to provide researchers with accurate literature support.

[0091] The present invention has been independently verified by over 500 papers and has achieved an accuracy rate of over 99%, surpassing the 85% accuracy of manual extraction and completely replacing manual labor.

[0092] The literature screening method provided by the embodiments of the present invention utilizes a multi-level screening mechanism. In the initial screening phase, a keyword screening model is used to quickly remove obviously irrelevant documents. In the in-depth screening phase, a large language model is used to perform detailed semantic analysis of the documents to ensure that their content meets PICO requirements. This combination of initial and in-depth screening ensures the accuracy of the screening results. This creates a dynamic vaccine monitoring mechanism that is updated in real time. High-quality vaccine-related literature collections can be updated in real time, enabling real-time risk warnings and safety assessments.

[0093] The document screening device provided by the present invention is described below. The document screening device described below and the document screening method described above can be referenced to each other.

[0094] Figure 2 A schematic diagram of a document screening device provided by an embodiment of the present invention, such as Figure 2 As shown, the document screening device provided by the embodiment of the present invention includes:

[0095] Extraction module 201, for extracting all document data related to the screening topic from the document library, and downloading the full text of the document data and its corresponding abstract;

[0096] A first output module 202 is configured to generate prompt words according to the screening theme, input the abstract portion of the document data and the prompt words into an intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a first screening result;

[0097] A second output module 203 is configured to input the prompt word and the full text of the document data into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities;

[0098] The verification module 204 is used to perform cross-validation based on the first screening result and the second screening result to obtain a final screening result.

[0099] A document screening device provided by an embodiment of the present invention extracts all document data related to a screening topic from a document library and downloads the full text of the document data and its corresponding abstract. A prompt word is generated based on the screening topic, and the abstract and the prompt word corresponding to the document data are input into an intelligent screening model. The intelligent screening model determines whether the document data meets PICO requirements and outputs a first screening result. The prompt word and the full text of the document data are input into the intelligent screening model, and the intelligent screening model determines whether the document data meets PICO requirements and outputs a second screening result. The intelligent screening model is one or more large language models with natural language understanding capabilities. A final screening result is obtained by cross-validation based on the first and second screening results. The large language model is used to screen documents that meet PICO requirements or are related to the current topic, thereby improving the accuracy and efficiency of document screening.

[0100] Figure 3 An example of a physical structure diagram of an electronic device is shown below. Figure 3 As shown, the server may include: a processor 310 , a communications interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communications interface 320 and the memory 330 communicate with each other via the communication bus 340 . The memory 330 includes a computer program, an operating system, and acquired data. The processor 310 can call logic instructions in the memory 330 to execute a document screening method, which includes: extracting all document data related to the screening topic from a document library and downloading the full text of the document data and its corresponding abstract; generating prompt words based on the screening topic, inputting the abstract corresponding to the document data and the prompt words into an intelligent screening model, determining whether the document data meets the PICO requirements through the intelligent screening model, and outputting a first screening result; inputting the prompt words and the full text of the document data into the intelligent screening model, determining whether the document data meets the PICO requirements through the intelligent screening model, and outputting a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities; and performing cross-validation based on the first and second screening results to obtain a final screening result.

[0101] Furthermore, the logic instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the relevant art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0102] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the document screening method provided by the above-mentioned methods, the method comprising: extracting all document data related to the screening topic from a document library, and downloading the full text of the document data and its corresponding abstract; generating prompt words according to the screening topic, inputting the abstract corresponding to the document data and the prompt words into an intelligent screening model, determining whether the document data meets the PICO requirements through the intelligent screening model, and outputting a first screening result; inputting the prompt words and the full text of the document data into the intelligent screening model, determining whether the document data meets the PICO requirements through the intelligent screening model, and outputting a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities; and performing cross-validation based on the first screening result and the second screening result to obtain a final screening result.

[0103] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0104] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the relevant technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for document screening, characterized in that: include: Extract all literature data related to the screening topic from the literature library, and download the full text of the literature data and its corresponding abstract; Generate prompt words according to the screening theme, input the abstract part and the prompt words corresponding to the document data into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a first screening result; Inputting the prompt word and the full text of the document data into the intelligent screening model, determining whether the document data meets the PICO requirements through the intelligent screening model, and outputting a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities; Perform cross-validation based on the first screening result and the second screening result to obtain a final screening result; Generating prompt words according to the screening topic includes: Preset multiple vaccine-related prompt words, input the prompt words into a first large language model, and fine-tune the multiple prompt words according to output results of the first large language model to generate new prompt words; Input the new prompt word into the second largest language model, and adjust the temperature of the output result of the second largest language model according to the output result of the first largest language model; Use a script to compare the output of the first language model and the output of the second language model, and obtain a final prompt word when the output of the first language model and the output of the second language model are consistent; The first language model and the second language model are the same or different models.

2. The method for document screening according to claim 1, wherein The screening subject is vaccine, and determining whether the literature data meets the PICO requirements by the intelligent screening model includes: Outputting evaluation details of each document through the intelligent screening model; Determine, based on the evaluation details, whether the literature research is related to the vaccine, whether it includes at least one of safety research, vaccine application targets, vaccine use, efficacy research, immunogenicity research, vaccination rate research, immunization program policy research, vaccine development, economic evaluation, literature type, clinical research type, and secondary research type related to the vaccine; If at least one of the above conditions is met, the literature data is determined to meet the PICO requirements.

3. The method for document screening according to claim 1, wherein After outputting the first screening result, the method further includes: Extracting some documents from the first screening results, verifying the some documents, and if the verification pass rate of the some documents does not reach a preset threshold, screening multiple documents from all document data related to the screening topic; The prompt word, the abstract part corresponding to the document included in the first screening result, and the abstract parts corresponding to multiple documents screened from all document data related to the screening topic are re-input into the intelligent screening model to update the first screening result.

4. The method for document screening according to claim 1, wherein The step of inputting the prompt word and the full text of the literature data into the intelligent screening model comprises: Obtaining text data of the full text of the document data; Converting the text data into a text vector; Obtaining relevant sentences and paragraphs based on the distance between the text vectors; The relevant sentences and paragraphs are input into the intelligent screening model.

5. The method for document screening according to claim 1, wherein The method extracts all literature data related to the screening topic from the literature library, including: Acquire multiple document data from multiple document libraries; perform deduplication preprocessing on the document data; All literature data related to the screening topic are extracted from the deduplicated literature through keyword technology.

6. The method for document screening according to claim 1, wherein Also includes: The final screening results are classified and sorted by the intelligent screening model to obtain multiple types of literature data.

7. A document screening device, characterized in that: include: The extraction module is used to extract all literature data related to the screening topic from the literature library and download the full text of the literature data and its corresponding abstract; a first output module, configured to generate prompt words according to the screening theme, input the abstract portion corresponding to the document data and the prompt words into an intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a first screening result; a second output module, configured to input the prompt word and the full text of the document data into the intelligent screening model, determine whether the document data meets the PICO requirements through the intelligent screening model, and output a second screening result, wherein the intelligent screening model is one or more large language models with natural language understanding capabilities; A verification module, configured to perform cross-validation based on the first screening result and the second screening result to obtain a final screening result; Generating prompt words according to the screening topic includes: Preset multiple vaccine-related prompt words, input the prompt words into a first large language model, and fine-tune the multiple prompt words according to output results of the first large language model to generate new prompt words; Input the new prompt word into the second largest language model, and adjust the temperature of the output result of the second largest language model according to the output result of the first largest language model; Use a script to compare the output of the first language model and the output of the second language model, and obtain a final prompt word when the output of the first language model and the output of the second language model are consistent; The first language model and the second language model are the same or different models.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the document screening method according to any one of claims 1 to 6 is implemented.

9. A non-transitory readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the document screening method according to any one of claims 1 to 6 is implemented.

Images