Energy storage field knowledge classification method and system based on large language model driven unsupervised mode

Abstract

The application discloses a kind of energy storage field knowledge classification method and system based on large language model driven unsupervised mode, by collecting field literature data and the field literature obtained is uniformly formatted and text cleaning processing, different sources, different format literature is converted into analyzable pure text form;Do a good job in literature classification preparation work. Utilize large language model to each literature carries out semantic understanding, and according to the preset core semantic template output literature core semantic information;According to the core semantic information of literature, the semantic classification of literature is carried out;Through training, the optimized large language model is obtained;Single theme label is converted into multidimensional semantic feature representation by semantic classification, to provide fine basis for subsequent classification, improve the accuracy of literature classification result, improve the precision of research direction description.

Description

Technical Field

[0001] This invention belongs to the field of document classification technology, specifically relating to a knowledge classification method and system for the energy storage field based on a large language model-driven unsupervised mode. Background Technology

[0002] As academic research continues to deepen, research outputs such as academic literature, technical reports, and patent documents are rapidly increasing across various fields, especially in interdisciplinary and emerging technology fields, characterized by their vast quantity, frequent updates, and highly overlapping research themes. When conducting field research or systematic learning, we typically acquire large-scale literature data through methods such as web scraping and database retrieval to systematically study and deeply analyze a specific research area. However, the literature obtained in this way is characterized by diverse sources, inconsistent structures, high overlap, and strong implicitness, making it difficult to efficiently and accurately identify research directions and placing higher demands on existing literature classification and management technologies.

[0003] Current methods for literature classification and research direction identification primarily rely on keyword matching or manually constructed rules, categorizing literature using pre-set thesaurus or classification rules. However, since research topics are often implicitly embedded in the context of long texts in diverse ways, keyword matching struggles to accurately reflect the true research intent. Furthermore, as research directions become increasingly specialized and domain terminology evolves, manually constructed rules and thesaurus require frequent maintenance, limiting their applicability and scalability. With a larger volume of literature or deeper overlap in research directions, these methods can easily lead to inconsistent classification results, affecting the accuracy and reliability of the classification.

[0004] To improve automation, some existing technologies have introduced document clustering and topic modeling methods based on statistical features or traditional machine learning, analyzing documents through word frequency or vector space models. However, these methods mainly rely on shallow statistical features and lack the ability to understand contextual semantics and research logic, making it difficult to accurately distinguish documents with similar research questions but different technical approaches or application scenarios. At the same time, most methods require manual pre-setting of the number of topics or classification boundaries, making it difficult to adapt to the dynamic evolution of research topics when research directions are not yet stable or new directions are constantly emerging.

[0005] Existing literature classification and research direction identification technologies are insufficient in terms of semantic modeling capabilities, classification granularity, and adaptability. They can usually only perform single topic tag division, ignoring the differences in multi-dimensional semantics of literature in terms of research questions, technical methods, and application scenarios. They are difficult to identify literature on interdisciplinary research directions and potential new research directions, and cannot meet the actual needs of refined research direction analysis and knowledge extraction of large-scale domain literature.

[0006] A new document classification method is needed to solve the above-mentioned technical problems. Summary of the Invention

[0007] The purpose of this invention is to provide a knowledge classification method for the energy storage field based on a large language model-driven unsupervised mode, which solves the technical problems of inaccurate literature classification results and single classification granularity in the prior art.

[0008] The present invention also aims to provide a literature classification system that employs a knowledge classification method for the energy storage field based on a large language model-driven unsupervised mode.

[0009] The technical solution of this invention to solve its technical problem is as follows: A knowledge classification method for the energy storage domain based on large language model-driven unsupervised patterns includes the following steps: S1: Collect domain literature data and perform format standardization and text cleaning on the acquired domain literature, converting literature from different sources and in different formats into parsable plain text. S2: Use a large language model to perform semantic understanding on each document and output the core semantic information of the document according to the preset core semantic template; perform semantic classification on the document according to the core semantic information; and obtain an optimized large language model through training. S3: After converting the documents to be classified and the set of references in the same field into parsable plain text, input them into a large language model to align the core semantic information of the documents and perform semantic hierarchical analysis, and output the classification results of the documents to be classified.

[0010] Preferably, the preset core semantic template includes: research object, research question, technical method, and application scenario.

[0011] Preferably, the semantic classification specifically includes the research question level, the technical method level, and the application scenario level.

[0012] Preferably, step S3 specifically comprises: S3.1: Input the documents to be classified and the set of references in the same field into the large language model to align the core semantic information of the documents; compare the differences between the documents to be classified and the references in the same field at the research question level, technical method level, and application scenario level, and calculate the similarity between the documents to be classified and the references in the same field in terms of research direction; S3.2: Output the classification results of the documents to be classified based on similarity.

[0013] Preferably, the similarity is divided into complete consistency, partial consistency, and inconsistency; the corresponding literature classification results are: literature with a clear research direction, literature with an interdisciplinary direction, and literature with a potential new research direction.

[0014] Preferably, the optimization process includes model reasoning and consistency verification.

[0015] A document classification system employing a knowledge classification method for the energy storage field based on a large language model-driven unsupervised model includes: The document semantic parsing and standardization module collects domain literature data and performs format unification and text cleaning on the acquired domain literature, converting documents from different sources and in different formats into parsable plain text. The document semantic classification modeling module uses a large language model to perform semantic understanding on each document and outputs the core semantic information of the document according to a preset core semantic template; it performs semantic classification on the document according to the core semantic information; and it obtains an optimized large language model through training. The document semantic comparison and classification module uses a large language model to align the core semantic information of documents and perform semantic hierarchical analysis. The classification result stabilization and output module outputs the classification results of the documents to be classified.

[0016] Preferably, the preset core semantic template includes: research object, research question, technical method, and application scenario.

[0017] Preferably, the semantic classification specifically includes the research question level, the technical method level, and the application scenario level.

[0018] Preferably, the optimization process includes model reasoning and consistency verification.

[0019] The reliability and stability of classification results are improved through model inference and consistency verification, reducing errors caused by fluctuations in single-round inference. The classification results are stored in a structured form, which can support literature screening, system learning and knowledge organization for specific research directions.

[0020] The beneficial effects of this invention are as follows: By collecting domain literature data and performing format unification and text cleaning on the acquired domain literature, documents from different sources and in different formats are transformed into parsable plain text; this prepares the groundwork for literature classification. A large language model is used to perform semantic understanding on each document and output the core semantic information of the document according to a preset core semantic template; the documents are semantically graded according to their core semantic information; an optimized large language model is obtained through training; and semantic grading is used to transform single topic tags into multi-dimensional semantic feature representations, providing a refined basis for subsequent classification, improving the accuracy of literature classification results, and enhancing the precision of research direction characterization. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the document classification method of this invention; Figure 2 This is a system structure diagram of the document classification system of this invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0023] like Figure 1 As shown, this invention discloses a knowledge classification method for the energy storage domain based on unsupervised mode driven by a large language model, comprising the following steps: S1: Collect domain literature data and perform format standardization and text cleaning on the acquired domain literature, converting literature from different sources and in different formats into parsable plain text. S2: Utilizing a large language model, semantic understanding is performed on each document, and the core semantic information of the document is output according to a pre-defined core semantic template. The documents are then semantically categorized based on this core semantic information. An optimized large language model is obtained through training. The pre-defined core semantic template includes: research object, research question, technical method, and application scenario. Semantic categorization specifically includes research question level, technical method level, and application scenario level. The optimization process includes model inference and consistency verification. These processes improve the reliability and stability of the classification results, reducing errors caused by fluctuations in single-round inference. The classification results are stored in a structured format, supporting document selection, system learning, and knowledge organization for specific research directions.

[0024] S3: After converting the documents to be classified and the set of references in the same field into parsable plain text, input them into a large language model to align the core semantic information of the documents and perform semantic hierarchical analysis, and output the classification results of the documents to be classified.

[0025] Specifically: S3.1: Input the documents to be classified and the set of references in the same field into the large language model to align the core semantic information of the documents; compare the differences between the documents to be classified and the references in the same field at the research question level, technical method level, and application scenario level, and calculate the similarity between the documents to be classified and the references in the same field in terms of research direction; S3.2: Output the classification results of the documents to be classified based on similarity. Similarity is categorized as completely identical, partially identical, or inconsistent; the corresponding classification results are: documents with a clear research direction, documents with interdisciplinary research directions, and documents with potential new research directions. The specific classification results are stored in a structured format, supporting document screening, systematic learning, and knowledge organization for specific research directions.

[0026] By collecting domain literature data and performing format standardization and text cleaning on the acquired literature, documents from different sources and formats are transformed into parsable plain text; preparations are made for literature classification. A large language model is used to perform semantic understanding on each document, and the core semantic information of the document is output according to a preset core semantic template; the documents are semantically classified according to their core semantic information; an optimized large language model is obtained through training; and the semantic classification is used to transform single topic tags into multi-dimensional semantic feature representations, providing a refined basis for subsequent classification, improving the accuracy of literature classification results, and enhancing the precision of research direction characterization.

[0027] like Figure 2 As shown, a document classification system for the energy storage domain, employing a knowledge classification method based on a large language model-driven unsupervised model, includes: The document semantic parsing and standardization module collects domain literature data and performs format unification and text cleaning on the acquired domain literature, converting documents from different sources and in different formats into parsable plain text. The document semantic classification modeling module utilizes a large language model to perform semantic understanding on each document and outputs the core semantic information of the document according to a preset core semantic template. It then performs semantic classification of the documents based on this core semantic information. An optimized large language model is obtained through training. The preset core semantic template includes: research object, research question, technical method, and application scenario. Semantic classification specifically includes research question level, technical method level, and application scenario level. The optimization process includes model inference and consistency verification.

[0028] The document semantic comparison and classification module uses a large language model to align the core semantic information of documents and perform semantic hierarchical analysis. The classification result stabilization and output module outputs the classification results for the documents to be classified. The specific process involves filtering documents based on their similarity to references in the same research direction, categorizing similarity into complete agreement, partial agreement, and disagreement. The corresponding classification results are: documents with a clear research direction, documents in interdisciplinary areas, and documents with potential new research directions. The classification results are stored in a structured format, supporting document filtering, systematic learning, and knowledge organization for specific research directions.

[0029] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

Claims

1. A knowledge classification method for the energy storage domain based on unsupervised mode driven by a large language model, characterized in that, Includes the following steps: S1: Collect domain literature data and perform format standardization and text cleaning on the acquired domain literature, converting literature from different sources and in different formats into parsable plain text. S2: Use a large language model to perform semantic understanding on each document and output the core semantic information of the document according to the preset core semantic template; perform semantic classification on the document according to the core semantic information; and obtain an optimized large language model through training. S3: After converting the documents to be classified and the set of references in the same field into parsable plain text, input them into a large language model to align the core semantic information of the documents and perform semantic hierarchical analysis, and output the classification results of the documents to be classified.

2. The knowledge classification method for energy storage based on large language model-driven unsupervised mode as described in claim 1, characterized in that: The preset core semantic template includes: research object, research question, technical method, and application scenario.

3. The knowledge classification method for energy storage based on large language model-driven unsupervised mode as described in claim 2, characterized in that: The semantic classification specifically includes the research question level, the technical method level, and the application scenario level.

4. The knowledge classification method for energy storage based on large language model-driven unsupervised mode as described in claim 3, characterized in that: Step S3 specifically involves: S3.1: Input the documents to be classified and the set of references in the same field into the large language model to align the core semantic information of the documents; Compare the differences between the literature to be classified and references in the same field at the levels of research questions, technical methods, and application scenarios, and calculate the similarity between the literature to be classified and references in the same field in terms of research direction; S3.2: Output the classification results of the documents to be classified based on similarity.

5. The knowledge classification method for energy storage based on large language model-driven unsupervised mode as described in claim 4, characterized in that: The similarity is divided into complete consistency, partial consistency, and inconsistency; the corresponding literature classification results are: literature with a clear research direction, literature with an interdisciplinary direction, and literature with a potential new research direction.

6. The document classification method based on a large language model according to claim 5, characterized in that: The optimization process includes model reasoning and consistency verification.

7. A document classification system employing the knowledge classification method for energy storage based on large language model-driven unsupervised mode as described in any one of claims 1-6, characterized in that, include: The document semantic parsing and standardization module collects domain literature data and performs format unification and text cleaning on the acquired domain literature, converting documents from different sources and in different formats into parsable plain text. The document semantic classification modeling module uses a large language model to perform semantic understanding on each document and outputs the core semantic information of the document according to a preset core semantic template; it performs semantic classification on the document according to the core semantic information; and it obtains an optimized large language model through training. The document semantic comparison and classification module uses a large language model to align the core semantic information of documents and perform semantic hierarchical analysis. The classification result stabilization and output module outputs the classification results of the documents to be classified.

8. The document classification system based on the knowledge classification method for energy storage driven by a large language model according to claim 7, characterized in that: The preset core semantic template includes: research object, research question, technical method, and application scenario.

9. The document classification system based on the knowledge classification method for energy storage driven by a large language model according to claim 7, characterized in that: The semantic classification specifically includes the research question level, the technical method level, and the application scenario level.

10. The document classification system based on the knowledge classification method for energy storage driven by a large language model according to claim 7, characterized in that: The optimization process includes model reasoning and consistency verification.

Images