Material literature element automatic extraction method and system based on large language model

Abstract

This invention provides a method and system for automatically extracting material elements from scientific and technological literature based on a large language model, belonging to the field of materials science and technology literature data extraction. First, this invention automatically parses documents of different formats to obtain hierarchical documents. Then, based on the large model, it locates all performance names and values ​​related to the target in the entire text using prompt words. Next, based on all performance values, it extracts relevant material composition and process information using prompt words, simplifying the complex extraction task and improving the extraction scope and accuracy, forming structured data containing all elements. Finally, through visualization processing of the complete dataset, it generates an element data distribution map. This invention achieves automatic extraction of target-related material elements from scientific and technological literature with high extraction accuracy and recall, requiring no manual labeling. This method is applicable to data extraction from scientific and technological literature in the field of materials science.

Description

Technical Field

[0001] This invention relates to the field of materials science and technology literature data extraction, and in particular to a method and system for automatic extraction of materials literature elements based on a large language model. Background Technology

[0002] Artificial intelligence and machine learning are transforming the research model in the field of materials science. Data-driven methods are gradually replacing the traditional "trial and error" model, significantly accelerating the research and development process of new materials. In the data-driven research model, high-quality and abundant data is the key to ensuring the successful application of artificial intelligence and machine learning. Currently, the acquisition of scientific data mainly relies on two channels: (1) data generated by researchers in the laboratory through experiments and calculations; (2) data manually collected from academic papers published by peers. Scientific literature contains a large amount of high-quality data resources that have been peer-reviewed. However, in the face of tens of thousands of academic articles written in natural language, relying on experts to manually extract data is not only time-consuming and laborious, but also increasingly unable to keep up with the speed of article publication. With the continuous surge in the number of new publications, the task of manually maintaining and continuously updating the database has become increasingly difficult for individual researchers. Summary of the Invention

[0003] To address the problems in the existing technologies, this invention provides a method and system for automatically extracting material document elements based on a large language model. The invention first automatically parses documents of different formats to obtain hierarchical documents. Then, based on the large model, it locates all performance names and values ​​related to the target in the entire text using prompt words. Next, based on all performance values, it extracts relevant material composition and process information using prompt words, simplifying the complex extraction task and improving the extraction scope and accuracy, forming structured data containing all elements. Finally, through visualization processing of the complete data set, it generates an element data distribution map. To achieve the above objectives, the technical solution is as follows: On the one hand, this invention provides a method for automatically extracting material document elements based on a large language model, the method comprising: S1. Based on multi-source heterogeneous literature data, a hierarchical document dataset with semantic structure is obtained through processing and parsing. S2. Based on this hierarchical document dataset, obtain the table information dataset through rule-based keyword matching and filtering; S3. Based on the table information dataset and the hierarchical document dataset, obtain the feature context dataset through textification and splicing. S4. Based on the context dataset of this element, the target dataset is obtained by guiding the processing with prompt words; S5. Based on the target dataset and the hierarchical document dataset, the performance dataset is obtained by calling the prompt word model; S6. Based on the performance dataset and the context dataset of the feature, obtain the structured extraction results through standard output prompts; S7. Based on the structured extraction result, the validated structured extraction result is obtained through judgment. The judgment rules are as follows: Judgment Rule 1: If the structured extraction result does not exceed the limit, it is deemed valid, and the verified structured extraction result is obtained; Judgment Rule 2: If the structured extraction result exceeds the limit, after supplementary processing, a verified structured extraction result is obtained; Judgment Rule 3: If there are multiple data entries with the same target performance in the structured extraction result, retain the first data entry with the same target performance and delete the other data entries with the same target performance. S8. Based on the verified structured extraction results, the feature data file is obtained through post-processing. S9. Based on the feature data file, obtain the feature data distribution map through visualization processing.

[0004] Optionally, in S1, based on multi-source heterogeneous document data, a hierarchical document dataset with semantic structure is obtained through processing and parsing, including: S11. Based on multi-source heterogeneous literature data, obtain the document format by identifying the data format; S12. Based on the document format, a hierarchical document dataset with semantic structure is obtained through judgment. The judgment rules are as follows: Judgment Rule 1: If the document is in HTML format, process and parse it using HTML format; Judgment Rule 2: If the document is in XML format, process and parse it using XML format; Judgment Rule 3: If the document is in PDF format, process and parse it using PDF format.

[0005] Optionally, in S12, by judging rule 1: if the document format is HTML, it is processed and parsed using HTML format to obtain a hierarchical document dataset with semantic structure, including: S1211. If the document format is HTML, by constructing a feature keyword dictionary, identify and remove the tags and content of the feature keyword dictionary to obtain the first-stage HTML hierarchical document dataset. S1212. Based on the hierarchical document dataset of the first-stage HTML, the publication metadata is matched and filtered using regular expressions to obtain a hierarchical document dataset with semantic structure.

[0006] Optionally, in S12, by judging rule 2: if the document format is XML, XML format processing and parsing are used to obtain a hierarchical document dataset with semantic structure, including: S1221. If the document format is XML, the hierarchical document dataset of the first stage of XML is obtained by constructing an XML document tree structure. S1222. Based on the hierarchical document dataset of the first-stage XML, the hierarchical document dataset of the second-stage XML is obtained by recursively traversing the text extraction of all XML element nodes. S1223. Based on the hierarchical document dataset of the second-stage XML, the text content is concatenated to obtain a hierarchical document dataset with semantic structure.

[0007] Optionally, in S12, by judging rule 3: if the document format is PDF, it is processed and parsed using PDF format to obtain a hierarchical document dataset with semantic structure, including: S1231. If the document format is PDF, the text blocks and corresponding location datasets are obtained through a dual parsing strategy based on physical layout and semantic rules. S1232. Based on the text block and the corresponding position dataset, the document dataset with restored paragraph structure is obtained through intelligent splicing and line break processing. S1233. Based on the document dataset that restores the paragraph structure, a hierarchical document dataset with semantic structure is obtained by identifying and removing page metadata information.

[0008] Optionally, the feature is characterized in that, in step S3, based on the table information dataset and the hierarchical document dataset, a feature context dataset is obtained through textification and concatenation processing, including: S31. Based on the table information dataset, obtain a text-based table dataset through textification; S32. Based on the textualized tabular dataset, the feature context dataset is obtained by concatenating it with the hierarchical document dataset.

[0009] Optionally, in S7, by judging rule 2: if the structured extraction result exceeds the limit, after supplementary processing, a verified structured extraction result is obtained, including: S721. If the structured extraction results exceed the limit, the remaining extraction results are obtained by constructing supplementary prompt words. S722. Based on the remaining extraction result and the extraction result that exceeds the limit, the extraction result is obtained by splicing. S723. Based on the spliced ​​extraction result, the structured extraction result is obtained after integrity check and blanking.

[0010] Optionally, the feature is characterized in that, in step S9, based on the feature data file, a feature data distribution map is obtained through visualization processing, including: S91. Based on the feature data file, the first-stage visualization analysis dataset is obtained by merging the data from multiple files into a unified data frame structure. S92. Based on the first-stage visualization analysis dataset, identify the element columns by column name pattern matching to obtain a standardized component data matrix; S93. Based on the standardized component data matrix, a performance parameter-unit mapping dataset is obtained through column name analysis and data type identification. S94. Based on the mapping dataset of this performance parameter-unit, obtain the feature data distribution map through visualization processing.

[0011] On the other hand, the present invention provides an automatic extraction system for material document elements based on a large language model. This system is applied to the automatic extraction method for material document elements based on a large language model. The system includes: The hierarchical document acquisition module is used to obtain a hierarchical document dataset with semantic structure by processing and parsing multi-source heterogeneous literature data. The table information acquisition module is used to obtain the table information dataset by filtering it through rule-based keyword matching based on the hierarchical document dataset. The feature context acquisition module is used to obtain the feature context dataset by textualization and concatenation processing based on the table information dataset and the hierarchical document dataset. The target extraction module is used to obtain the target dataset by guiding the processing with prompt words based on the context dataset of the feature. The performance dataset acquisition module is used to obtain the performance dataset by calling the prompt word model based on the target extraction dataset and the hierarchical document dataset. The structured extraction result acquisition module is used to obtain structured extraction results based on the performance dataset and the context dataset of the feature, through standard output prompts. The structured extraction result verification module obtains the verified structured extraction result by judging the structured extraction result. The feature data file acquisition module is used to obtain the feature data file through post-processing based on the verified structured extraction results. The feature visualization module is used to generate a feature data distribution map based on the feature data file through visualization processing.

[0012] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects: The above-mentioned solution has several advantages. First, based on the deep semantic understanding capabilities of large language models, the system can accurately grasp the professional terminology and complex contextual relationships in the field of materials science, significantly improving the accuracy of information extraction. Second, through full-process automation, it achieves seamless integration from literature analysis to visualization analysis, greatly reducing manual intervention and subjective errors. Third, through prompt word templates and segmented, hierarchical extraction strategies, the system can quickly adapt to different material systems and research focuses, demonstrating strong scalability. Fourth, the generated structured data and visualization results can directly serve deep data analysis, model building, and knowledge discovery, providing efficient and reliable data support for materials science research and improving overall research efficiency. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of an embodiment of the automatic extraction method for material document elements based on a large language model according to the present invention; Figure 2 This is a flowchart illustrating the process of obtaining a hierarchical document dataset in an embodiment of the automatic extraction method for material document elements based on a large language model according to the present invention. Figure 3 This is a flowchart illustrating the method for automatically extracting material document elements based on a large language model according to the present invention, which obtains a hierarchical document dataset in HTML format. Figure 4 This is a flowchart illustrating the automatic extraction method for material document elements based on a large language model according to the present invention, which obtains a hierarchical document dataset in XML format. Figure 5 This is a flowchart illustrating the automatic extraction method for material document elements based on a large language model according to the present invention for obtaining a hierarchical document dataset in PDF format; Figure 6 This is a flowchart illustrating the acquisition of the element context dataset in an embodiment of the automatic extraction method for material document elements based on a large language model according to the present invention. Figure 7 This is a schematic diagram illustrating the performance extraction prompt word template of an embodiment of the automatic extraction method for material document elements based on a large language model of the present invention; Figure 8 This is a flowchart illustrating the structured extraction results obtained after verification in an embodiment of the automatic extraction method for material and document elements based on a large language model according to the present invention. Figure 9 This is a flowchart of the element data visualization of an embodiment of the automatic extraction method for material document elements based on a large language model of the present invention; Figure 10 This is a system block diagram of an embodiment of the automatic extraction system for material and document elements based on a large language model according to the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] like Figure 1 The flowchart shown is an embodiment of the automatic extraction method for material document elements based on a large language model according to the present invention. The present invention provides an automatic extraction method for material document elements based on a large language model, which is implemented by a material document element automatic extraction system based on a large language model. The method includes: S1. Based on multi-source heterogeneous literature data, a hierarchical document dataset with semantic structure is obtained through processing and parsing. Specifically, such as Figure 2 The flowchart shown is from an embodiment of the automatic extraction method for document elements based on a large language model of the present invention for obtaining a hierarchical document dataset. In step S1, based on multi-source heterogeneous document data, a hierarchical document dataset with semantic structure is obtained through processing and parsing, including: S11. Based on multi-source heterogeneous literature data, obtain the document format by identifying the data format; S12. Based on the document format, a hierarchical document dataset with semantic structure is obtained through judgment. The judgment rules are as follows: Judgment Rule 1: If the document is in HTML format, process and parse it using HTML format; Furthermore, such as Figure 3The flowchart shown is from an embodiment of the automatic extraction method for document elements based on a large language model of the present invention, illustrating the acquisition of a hierarchical document dataset in HTML format. In S12, rule 1 is used: if the document format is HTML, HTML format processing and parsing are employed to obtain a hierarchical document dataset with semantic structure, including: S1211. If the document format is HTML, by constructing a feature keyword dictionary, identify and remove the tags and content of the feature keyword dictionary to obtain the first-stage HTML hierarchical document dataset. S1212. Based on the hierarchical document dataset of the first-stage HTML, the publication metadata is matched and filtered by regular expressions to obtain a hierarchical document dataset with semantic structure. Judgment Rule 2: If the document is in XML format, process and parse it using XML format; Furthermore, such as Figure 4 The flowchart shown is from an embodiment of the automatic extraction method for document elements based on a large language model of the present invention, illustrating the acquisition of a hierarchical document dataset in XML format. In step S12, rule 2 is used: if the document format is XML, XML format processing and parsing are employed to obtain a hierarchical document dataset with semantic structure, including: S1221. If the document format is XML, the hierarchical document dataset of the first stage of XML is obtained by constructing an XML document tree structure. S1222. Based on the hierarchical document dataset of the first-stage XML, the hierarchical document dataset of the second-stage XML is obtained by recursively traversing the text extraction of all XML element nodes. S1223. Based on the hierarchical document dataset of the second-stage XML, the text content is concatenated to obtain a hierarchical document dataset with semantic structure. Judgment Rule 3: If the document is in PDF format, process and parse it using PDF format; Furthermore, such as Figure 5 The flowchart shown is from an embodiment of the automatic extraction method for document elements based on a large language model of the present invention, illustrating the acquisition of a hierarchical document dataset in PDF format. In S12, according to judgment rule 3: if the document format is PDF, PDF format processing and parsing are used to obtain a hierarchical document dataset with semantic structure, including: S1231. If the document format is PDF, the text blocks and corresponding location datasets are obtained through a dual parsing strategy based on physical layout and semantic rules. S1232. Based on the text block and the corresponding position dataset, the document dataset with restored paragraph structure is obtained through intelligent splicing and line break processing. S1233. Based on the document dataset that restores the paragraph structure, a hierarchical document dataset with semantic structure is obtained by identifying and removing page metadata information.

[0019] S2. Based on this hierarchical document dataset, obtain the table information dataset through rule-based keyword matching and filtering.

[0020] S3. Based on the table information dataset and the hierarchical document dataset, obtain the feature context dataset through textification and splicing. Specifically, such as Figure 6 The flowchart shown is from an embodiment of the automatic extraction method for material document elements based on a large language model of the present invention, illustrating the process of obtaining the element context dataset. In step S3, based on the tabular information dataset and the hierarchical document dataset, the element context dataset is obtained through textualization and concatenation processing, including: S31. Based on the table information dataset, obtain a text-based table dataset through textification; S32. Based on the textualized tabular dataset, the feature context dataset is obtained by concatenating it with the hierarchical document dataset.

[0021] S4. Based on the context dataset of this element, the target dataset is obtained by guiding the processing with prompt words; Guided by predefined prompt word templates, a large language model is used to extract research keywords from literature across three dimensions: material category, processing technology, and material properties. The confidence score of each candidate term is calculated, generating a candidate list containing extracted names and popularity for various material elements. This generates the target dataset for extraction. For example, guided by a single prompt word: "Answer the question as accurately as possible using the provided context. Extract and summarize all information from the provided context with the following details: {material_type} composition or name (if the name is not provided), sample name, distinguishing factor, synthesis and processing routes, properties, and sourced figure. Key requirements: 1. {material_type} composition: The elemental content of each element with its corresponding unit (at.% or wt.%). Composition unit: The unit of elemental content (at.% or wt%). 2. {material_type} name: The specific names or designation of {material_type}s if available (e.g., Co-9Al-8W, CMSX-4). 3. Sample name: The specific names or designation." of samples. 4. distinguishing factor: Anycharacteristic or parameter that differentiates {material_type}s.Treat{material_type}s with different factors as separate entries even with similarbase composition. 5. Synthesis and processing routes: Thoroughly combineinformation from all sections (text, tables, figures). Include ALL concreteprocessing parameters (temperature, time, thickness, passes, etc.). Whenparameters (conditions) are given as ranges in text but exact values exist intables, use the exact values. Maintain chronological order of processingsteps. For multi-pass operations (e.g., rolling), include all pass-specificparameters when available. The synthesis and processing operations of theextraction should include but are not limited to the following operations:{actions}. test route / condition: the specific and distinct test conditionsapplied to each sample during the test stage. This includes only the uniqueaspects of the test process, such as test temperature, atmosphere (e.g.,hydrogen / helium), gas pressure, strain rate, and loading mode. 6.Propertiesof {material_type}: {str(targets)} with units. 8. Sourced figure: The figurelabel containing property data. Processing rules: Create separate entries for{material_type} with different distinguishing factors. Never combineproperties from different {material_type}s. For multiple units / componentswithin same {material_type}, create separate rows. If there are varioussynthesis and processing routes mentioned for the same {material_type}(different samples), represent them in multiple rows. When multipleprocessing routes exist for same {material_type}, create separate rows.Preserve all numerical values exactly as presented (no averaging or merging).Output should contains all entries with differing compositions, synthesis andprocessing routes, and properties present in the context.The output shouldmaintain the same dictionary structure as {output_pattern}, must have these 9keys (all lowercase): '{material_type} composition', '{material_type} name','sample name, 'distinguishing factor', 'synthesis and processing routes',{str(targets)}. The output should include all relevant information from thearticle and should not use ellipses to omit any content. Return ONLY thecomplete output with ALL {material_type}s and samples, NO explanatory text.For all 'value' fields in the output, a single numerical value must beoutput. If the original data contains multiple values (e.g., [535, 551, 567,585]), please split them into multiple separate records. Each record shouldinclude: one individual value and its corresponding specific test condition(extracted from the compound condition). If the value of a field is empty(e.g., 'Al': ''), simply ignore that field and do not include it in theoutput. Only retain fields with specific numerical values or non-emptystrings.Ensure every data point from all tables / figures / text is captured without exception. If the term is ambiguous or not explicitly named, do not assume or infer it. Extraction must be label-based, not semantic-based. The provided context is {all_infos}, where {material_type} represents the material type (e.g., steel, alloy), {output_pattern} represents the defined output structure, and {all_infos} represents the full text after concatenating the text and tables.

[0022] S5. Based on the target dataset and the hierarchical document dataset, the performance dataset is obtained by calling the prompt word model; Since performance values ​​can serve as material identifiers, the performance name and value are extracted from each paragraph first to simplify the extraction task and improve extraction accuracy, thereby ensuring the coverage of subsequent full-element extraction. The prompt word model can dynamically generate a highly customized performance extraction prompt word template, such as... Figure 7 As shown, the template clearly specifies the types of materials to be extracted, the specific list of properties, and the strict output data structure, ultimately resulting in a performance dataset.

[0023] S6. Based on the performance dataset and the context dataset of the feature, obtain the structured extraction results through standard output prompts.

[0024] By using performance datasets with various performance names and values, relevant components and processes can be identified in the full text in a targeted manner, greatly reducing the problems of model illusion and extraction omissions.

[0025] S7. Based on the structured extraction result, the validated structured extraction result is obtained through judgment. The judgment rules are as follows: Judgment Rule 1: If the structured extraction result does not exceed the limit, it is deemed valid, and the verified structured extraction result is obtained; Judgment Rule 2: If the structured extraction result exceeds the limit, after supplementary processing, a verified structured extraction result is obtained; Specifically, such as Figure 8The flowchart shown in this embodiment of the automatic extraction method for material document elements based on a large language model of the present invention obtains the verified structured extraction result. In S7, according to judgment rule 2: if the structured extraction result exceeds the limit, after supplementary processing, the verified structured extraction result is obtained, including: S721. If the structured extraction results exceed the limit, the remaining extraction results are obtained by constructing supplementary prompt words. S722. Based on the remaining extraction result and the extraction result that exceeds the limit, the extraction result is obtained by splicing. S723. Based on the spliced ​​extraction result, the structured extraction result is obtained after integrity check and blank filling. Judgment Rule 3: If there are multiple data entries with the same target performance in the structured extraction result, retain the first one of the multiple data entries with the same target performance and delete the other data entries with the same target performance.

[0026] S8. Based on the structured extraction results, the feature data file is obtained through post-processing.

[0027] S9. Based on the feature data file, obtain the feature data distribution map through visualization processing; Optionally, the feature is characterized in that, in step S9, based on the feature data file, a feature data distribution map is obtained through visualization processing, including: Specifically, such as Figure 9 The flowchart shown is a feature data visualization process for an embodiment of the automatic feature extraction method for materials and documents based on a large language model of the present invention. In step S9, based on the feature data file, a feature data distribution map is obtained through visualization processing, including: S91. Based on the feature data file, the first-stage visualization analysis dataset is obtained by merging the data from multiple files into a unified data frame structure. S92. Based on the first-stage visualization analysis dataset, identify the element columns by column name pattern matching to obtain a standardized component data matrix; S93. Based on the standardized component data matrix, a performance parameter-unit mapping dataset is obtained through column name analysis and data type identification. S94. Based on the mapping dataset of this performance parameter-unit, obtain the feature data distribution map through visualization processing.

[0028] like Figure 10The diagram shown is a system block diagram of an embodiment of the material document element automatic extraction system based on a large language model of the present invention. The present invention provides a material document element automatic extraction system based on a large language model. The system is applied to a material document element automatic extraction method based on a large language model. The system includes: a hierarchical document acquisition module, used to obtain a hierarchical document dataset with semantic structure by processing and parsing multi-source heterogeneous document data. The module includes modules for obtaining table information, obtaining feature context, obtaining extraction targets, obtaining performance datasets, obtaining structured extraction results, validating structured extraction results, obtaining feature data files, and visualizing features. Specifically: The hierarchical document acquisition module is used to obtain a hierarchical document dataset with semantic structure by processing and parsing multi-source heterogeneous literature data. The table information acquisition module is used to obtain the table information dataset by filtering it through rule-based keyword matching based on the hierarchical document dataset. The feature context acquisition module is used to obtain the feature context dataset by textualization and concatenation processing based on the table information dataset and the hierarchical document dataset. The target extraction module is used to obtain the target dataset by guiding the processing with prompt words based on the context dataset of the feature. The performance dataset acquisition module is used to obtain the performance dataset by calling the prompt word model based on the target extraction dataset and the hierarchical document dataset. The structured extraction result acquisition module is used to obtain structured extraction results based on the performance dataset and the context dataset of the feature, through standard output prompts. The structured extraction result verification module obtains the verified structured extraction result by judging the structured extraction result. The feature data file acquisition module is used to obtain the feature data file through post-processing based on the verified structured extraction results. The feature visualization module is used to generate a feature data distribution map based on the feature data file through visualization processing.

[0029] This invention provides a method and system for automatically extracting material document elements based on a large language model. First, the invention automatically parses documents of different formats to obtain hierarchical documents. Then, based on the large model, it locates all performance names and values ​​related to the target in the entire text using prompt words. Second, based on all performance values, it extracts relevant material composition and process information using prompt words, simplifying the complex extraction task and improving the extraction scope and accuracy, forming structured data containing all elements. Finally, through visualization processing of the complete data dataset, it generates an element data distribution map.

[0030] It is understood that the present invention has been described through the above embodiments and should not be construed as limiting the implementation and scope of the present invention. Those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the present invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for automatically extracting material document elements based on a large language model, characterized in that, The method includes: S1. Based on multi-source heterogeneous literature data, a hierarchical document dataset with semantic structure is obtained through processing and parsing. S2. Based on the hierarchical document dataset, obtain the table information dataset by filtering through rule-based keyword matching; S3. Based on the table information dataset and the hierarchical document dataset, obtain the element context dataset through textification and splicing processing; S4. Based on the element context dataset, the target dataset is obtained by guiding the processing with prompt words; S5. Based on the extracted target dataset and the hierarchical document dataset, the performance dataset is obtained by calling the prompt word model; S6. Based on the performance dataset and the feature context dataset, obtain structured extraction results through standard output prompts; S7. Based on the structured extraction results, a verified structured extraction result is obtained through judgment. The judgment rules are as follows: Judgment Rule 1: If the structured extraction result does not exceed the limit, it is determined to be valid, and the verified structured extraction result is obtained; Judgment Rule 2: If the structured extraction result exceeds the limit, after supplementary processing, a verified structured extraction result is obtained; Judgment Rule 3: If there are multiple data entries with the same target performance in the structured extraction results, then retain the first data entry with the same target performance and delete the other data entries with the same target performance. S8. Based on the verified structured extraction results, the feature data file is obtained through post-processing. S9. Based on the aforementioned feature data file, a feature data distribution map is obtained through visualization processing.

2. The method for automatic extraction of material document elements based on a large language model according to claim 1, characterized in that, In step S1, based on multi-source heterogeneous document data, a hierarchical document dataset with semantic structure is obtained through processing and parsing, including: S11. Based on multi-source heterogeneous literature data, obtain the document format by identifying the data format; S12. Based on the document format, a hierarchical document dataset with semantic structure is obtained through judgment. The judgment rules are as follows: Judgment Rule 1: If the document format is HTML, it shall be processed and parsed in HTML format; Judgment Rule 2: If the document format is XML, it shall be processed and parsed using XML format. Judgment Rule 3: If the document format is PDF, it shall be processed and parsed using PDF format.

3. The method for automatic extraction of material document elements based on a large language model according to claim 2, characterized in that, In step S12, rule 1 is used: if the document format is HTML, it is processed and parsed using HTML format to obtain a hierarchical document dataset with semantic structure, including: S1211. If the document format is HTML, by constructing a feature keyword dictionary, identifying and removing the tags and content of the feature keyword dictionary, a hierarchical document dataset of HTML in the first stage is obtained. S1212. Based on the hierarchical document dataset of HTML from the first stage, the publication metadata is matched and filtered using regular expressions to obtain a hierarchical document dataset with semantic structure.

4. The method for automatic extraction of material document elements based on a large language model according to claim 2, characterized in that, In step S12, rule 2 is used to determine if the document format is XML. If so, XML format processing and parsing are employed to obtain a hierarchical document dataset with semantic structure, including: S1221. If the document format is XML, the hierarchical document dataset of the first stage XML is obtained by constructing an XML document tree structure. S1222. Based on the hierarchical document dataset of the first-stage XML, the hierarchical document dataset of the second-stage XML is obtained by recursively traversing the text extraction of all XML element nodes. S1223. Based on the hierarchical document dataset of the second stage XML, the text content is concatenated to obtain a hierarchical document dataset with semantic structure.

5. The method for automatic extraction of material document elements based on a large language model according to claim 2, characterized in that, In step S12, rule 3 is used to determine if the document format is PDF. If the document is in PDF format, it is parsed using PDF processing to obtain a hierarchical document dataset with semantic structure, including: S1231. If the document format is PDF, the text blocks and corresponding location datasets are obtained through a dual parsing strategy based on physical layout and semantic rules. S1232. Based on the text block and the corresponding position dataset, a document dataset with restored paragraph structure is obtained through intelligent splicing and line break processing. S1233. Based on the document dataset with restored paragraph structure, a hierarchical document dataset with semantic structure is obtained by identifying and removing page metadata information.

6. The method for automatic extraction of material document elements based on a large language model according to claim 1, characterized in that, In step S3, based on the table information dataset and the hierarchical document dataset, a feature context dataset is obtained through textification and concatenation processing, including: S31. Based on the table information dataset, obtain a textualized table dataset by texturing; S32. Based on the textualized tabular dataset, the feature context dataset is obtained by concatenating it with the hierarchical document dataset.

7. The method for automatic extraction of material document elements based on a large language model according to claim 1, characterized in that, In step S7, by judging rule 2: if the structured extraction result exceeds the limit, after supplementary processing, the verified structured extraction result is obtained, including: S721. If the structured extraction results exceed the limit, the remaining extraction results are obtained by constructing supplementary prompt words. S722. Based on the remaining extraction results and the extraction results exceeding the limit, a spliced ​​extraction result is obtained through splicing. S723. Based on the spliced ​​extraction results, the structured extraction results are verified through integrity checks and blank filling.

8. The method for automatic extraction of material document elements based on a large language model according to claim 1, characterized in that, In step S9, based on the feature data file, a feature data distribution map is obtained through visualization processing, including: S91. Based on the aforementioned element data file, the first-stage visualization analysis dataset is obtained by merging multiple file data into a unified data frame structure. S92. Based on the visualization analysis dataset of the first stage, identify the element columns by column name pattern matching to obtain a standardized component data matrix; S93. Based on the standardized component data matrix, a performance parameter-unit mapping dataset is obtained through column name analysis and data type identification; S94. Based on the performance parameter-unit mapping dataset, a feature data distribution map is obtained through visualization processing.

9. An automatic extraction system for material document elements based on a large language model, used to implement the automatic extraction method for material document elements based on a large language model as described in any one of claims 1-8, characterized in that, The system includes: The hierarchical document acquisition module is used to obtain a hierarchical document dataset with semantic structure by processing and parsing multi-source heterogeneous literature data. The table information acquisition module is used to obtain the table information dataset by filtering based on rule-based keyword matching according to the hierarchical document dataset; The feature context acquisition module is used to obtain the feature context dataset by textualization and concatenation processing based on the table information dataset and the hierarchical document dataset. The target extraction module is used to obtain the target extraction dataset based on the element context dataset and guided by prompt words. The performance dataset acquisition module is used to obtain the performance dataset by calling the prompt word model based on the extracted target dataset and the hierarchical document dataset. The structured extraction result acquisition module is used to obtain structured extraction results based on the performance dataset and the feature context dataset, through standard output prompts. The structured extraction result verification module obtains the verified structured extraction result by judging the structured extraction result. The feature data file acquisition module is used to obtain the feature data file through post-processing based on the verified structured extraction results. The feature visualization module is used to obtain a feature data distribution map based on the feature data file through visualization processing.

Images