High polymer material modification optimization method and system based on RAG

Abstract

The invention discloses an RAG-based high polymer material modification optimization method and system. The method comprises the following steps: constructing a high polymer material modification professional knowledge system; constructing a retrieval enhancement generation system, acquiring intermediate data from the professional knowledge system according to a demand input by a user, evaluating the intermediate data generated by the retrieval enhancement generation system, designing an evaluation index system, and performing natural language understanding on the intermediate data by using a large language model, analyzing the matching degree between the data content and the query requirement; inputting a modification demand or an optimization demand of a high polymer material into the retrieval enhancement generation system to obtain matching content; and performing answer generation through the matching content by using a large language model to obtain a high polymer material modification scheme or optimization scheme. According to the high polymer material modification optimization method provided by the invention, on the premise of meeting the requirements of different application scenes, the research and development efficiency is effectively improved, the number of unnecessary tests is reduced, and experimental materials and labor cost are saved.

Description

Technical Field

[0001] The present invention belongs to the technical field of polymer material modification, and in particular relates to a polymer material modification optimization method and system based on RAG. Background Art

[0002] Polymers have been widely used in many fields such as industry, medicine, electronics, aerospace, etc. Their excellent properties such as light weight, corrosion resistance, and easy processing make them an important part of modern materials science. However, in practical applications, a single polymer material is often difficult to meet the complex needs of specific application scenarios, and its performance needs to be improved through modification methods, such as enhancing mechanical strength, improving heat resistance, increasing conductivity or UV resistance, etc. Traditional polymer material modification methods include blending, copolymerization, grafting modification, filling modification, surface modification, etc., but these methods usually rely on the experience of experimenters and a large number of experiments, which is time-consuming and costly.

[0003] Traditional information retrieval methods in the field of materials science mainly rely on keyword matching and simple statistical models. With the development of artificial intelligence and big data technologies, some methods based on machine learning and natural language processing are used to improve the effects of information retrieval and knowledge discovery. These methods include but are not limited to: deep learning models for text classification and entity recognition, knowledge graph technology for building material science knowledge networks, reinforcement learning algorithms for optimizing search strategies, and large language models for understanding and generating professional texts.

[0004] The Chinese invention patent application with publication number CN118297275A discloses a data-driven intelligent process generation method and system, whose scheme is: marking and cleaning the process-related data in the knowledge base, and using the marked and cleaned process-related data to fine-tune the language model LLM; collecting process business data, corresponding design data and production data in the user's production system, combining and assembling them with process as the granularity, and storing them independently according to the process type, and embedding them into an independent knowledge base; processing and optimizing the data, combining the processing results of the language model LLM, and generating the content of the intelligent process; at the same time, collecting feedback data, using the feedback data as annotations, and using it for fine-tuning the language model LLM.

[0005] However, the existing polymer material modification scheme design still faces the following challenges: data in the field of polymer material modification are often scattered in different research papers, patent documents and experimental reports, lacking systematization and standardization, and are difficult to be directly used in the generation of modification schemes; factors such as the chemical structure, physical properties and processing conditions of different polymer materials are highly complex, and the design of modification schemes needs to consider the interaction between multiple variables; existing modification scheme generation methods mainly rely on static rules or empirical formulas, which are difficult to cope with the diverse requirements for material performance in different application scenarios. Therefore, there is an urgent need for a data-driven intelligent system that can efficiently integrate and utilize massive data in the field of polymer material modification, automatically generate innovative and efficient modification schemes to meet the needs of different application scenarios, and promote the further development and application of polymer materials. Summary of the invention

[0006] The present invention provides a RAG-based polymer material modification optimization method and system, aiming to solve the problems of insufficient versatility and low efficiency in the prior art in polymer material modification design.

[0007] In order to solve the above technical problems, the present invention proposes a RAG-based polymer material modification optimization method, comprising the following steps:

[0008] Build a professional knowledge system for polymer material modification;

[0009] Constructing a search enhancement generation system, obtaining intermediate data from the professional knowledge system according to the requirements input by the user, evaluating the intermediate data generated by the search enhancement generation system, designing an evaluation index system, using a large language model to perform natural language understanding on the intermediate data, and analyzing the matching degree between the data content and the query requirements;

[0010] Inputting polymer material modification requirements or optimization requirements into the search enhancement generation system to obtain matching content;

[0011] A large language model is used to generate answers based on the matching content to obtain a polymer material modification plan or optimization plan.

[0012] Preferably, after the method completes the construction of the polymer material modification professional knowledge system, the professional knowledge system is also evaluated, and a prompt is issued when any evaluation index is lower than a preset value. The evaluation indexes include:

[0013] New material coverage, assessing the knowledge base's coverage of the latest material categories and checking the completeness of the collection of cutting-edge synthesis methods and characterization techniques;

[0014] Accuracy of material performance data: random sampling to verify key material performance data and cross-validate with the experimental database;

[0015] The integrity of material structure information, evaluates the integrity of crystal structure, electronic structure, and defect information.

[0016] Preferably, the method for constructing the knowledge system is specifically as follows:

[0017] Collect raw data related to polymer material modification, clean and format the raw data, including noise removal, terminology unification, word segmentation, part-of-speech tagging and entity recognition operations, and convert the processed data into a structured knowledge representation form;

[0018] Vectorize the processed raw data using a pre-trained language model to construct a high-dimensional vector space;

[0019] Extract entity relationships from vectorized raw data and construct a knowledge network for polymer material modification;

[0020] A large language model is used to generate key concept explanations and technical principle elaborations based on the knowledge network to obtain a knowledge system including entities, relationships and attributes.

[0021] Preferably, the original data includes professional journals, patent databases, scientific research papers, experimental reports, material manuals, technical standards and industry reports.

[0022] Preferably, the evaluation of the intermediate data generated by the search enhancement generation system is specifically performed as follows:

[0023] Constructing an evaluation index system, wherein the evaluation index system includes data accuracy, timeliness and relevance;

[0024] Use a large language model to understand natural language, analyze the matching degree between data content and query requirements, and use a BERT-based model to calculate the semantic similarity between the query and the search results; use the text implication recognition module to determine whether the search results contain the information required for the query;

[0025] Generate an assessment report, mark high-quality data and potentially problematic data, build a scoring weighting algorithm, and combine various indicators to derive an overall score.

[0026] Preferably, the timeliness score of the evaluation index system is calculated as:

[0027] f(t)=e -λ(T-t)

[0028] Where f(t) is the timeliness score, t is the data release date, T is the current date, and λ is the set attenuation coefficient.

[0029] Preferably, the timeliness score of the evaluation index system is calculated as:

[0030]

[0031] Where f(t) is the timeliness score, t is the data release date, T is the current date, and β is the set attenuation coefficient.

[0032] Preferably, the retrieval enhancement generation includes data preprocessing, vector indexing, semantic similarity calculation and context-related retrieval strategy;

[0033] The data preprocessing is used for data cleaning and data distributed processing;

[0034] The vector index selects HNSW as the core index structure and uses the Faiss library to implement the HNSW index;

[0035] The semantic similarity calculation uses cosine similarity, Jaccard similarity or Euclidean distance to calculate the semantic similarity;

[0036] The context-dependent retrieval strategy uses WordNet or a domain-specific ontology library for query expansion and uses an attention mechanism to perform multiple rounds of retrieval to narrow the scope of results.

[0037] Preferably, after the search enhancement generation system is constructed, the search module of the search enhancement generation system is also evaluated, and the indicators include:

[0038] The ability to identify professional material terms, and the accuracy of the search enhancement generation system in identifying specific terms, chemical formulas, and structural descriptors in the new material field;

[0039] Multimodal search capability, the ability of the search enhancement generation system to handle multimodal input, including text, chemical structure diagrams, and material spectra;

[0040] Associative retrieval capability, the ability of the retrieval enhancement generation system to associate material properties, synthesis methods, and application fields.

[0041] In another aspect of the present invention, a RAG-based polymer material modification optimization system is also provided, wherein the system implements the above-mentioned optimization method, including:

[0042] Data collection module, used to collect raw data from professional journals, patent databases, scientific research papers, experimental reports, material manuals, technical standards and industry reports related to polymer material modification;

[0043] A knowledge base module, used to construct a professional knowledge system of polymer material modification based on the original data;

[0044] A search and query module, used to query matching content from the professional knowledge system according to the input modification requirements or optimization requirements;

[0045] A large language model is used to construct prompt words according to the matching content, and generate answers using the prompt words.

[0046] Compared with the prior art, the present invention has the following technical effects:

[0047] 1. The polymer material modification optimization method proposed in the present invention constructs a professional knowledge system of polymer material modification, combines retrieval enhancement generation with the use of a large language model to generate modification plans for user reference, and can efficiently integrate and utilize massive data in the field of polymer material modification, and automatically generate innovative and efficient modification plans to meet the needs of different application scenarios.

[0048] 2. The polymer material modification optimization method proposed in the present invention provides R&D personnel with modification schemes based on the latest data and knowledge, reduces the number of experiments, and accelerates the R&D process; by accurately generating efficient modification schemes and experimental design suggestions, the system can help researchers reduce the number of unnecessary experiments, save experimental materials and labor costs; and provide intelligent decision-making support in material selection and process adjustment to improve decision-making efficiency and accuracy.

[0049] 3. The polymer material modification optimization method proposed in the present invention can keep the knowledge system up to date by continuously collecting the latest research results and data, ensuring that the system can refer to the latest scientific research progress and technological trends when generating modification plans.

[0050] 4. The polymer material modification optimization method proposed in the present invention is combined with the self-learning ability of the large language model. The system can continuously adjust and optimize the generated model as the user usage increases, so that the output modification plan is more accurate and meets user needs.

[0051] 5. The polymer material modification optimization method proposed in the present invention can quickly and efficiently retrieve knowledge related to user needs by establishing a professional knowledge system of polymer material modification. Compared with traditional literature retrieval, this method is more targeted and can directly locate the key information required for polymer modification research. Semantic similarity and structured data in the knowledge system can be used to effectively filter out redundant information and improve the accuracy and efficiency of information retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the process of the polymer material modification optimization method of the present invention. DETAILED DESCRIPTION

[0053] In order 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 combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0054] Embodiment 1

[0055] This embodiment is a polymer material modification optimization method based on RAG, where RAG is the abbreviation of Retrieval-augmented Generation, which means retrieval-augmented generation in Chinese. Figure 1 As shown, the steps include steps 1 to 4:

[0056] Step 1: Construct a professional knowledge system for polymer material modification. The specific method for constructing the knowledge system is as follows:

[0057] Collect raw data related to polymer material modification, including professional journals, patent databases, scientific research papers, experimental reports, material manuals, technical standards and industry reports. Clean and format the raw data, including noise removal, terminology unification, word segmentation, part-of-speech tagging and entity recognition operations, and convert the processed data into a structured knowledge representation form;

[0058] Specifically, the latest academic papers related to polymer material modification can be captured through professional database API interfaces, such as Scopus, Web of Science and other database APIs. And by implementing an incremental data request mechanism plus a keyword combination strategy, such as "polymer modification" AND (nanocomposite OR "surface treatment"), the latest published papers can be automatically updated every week. Another example is collecting industry reports, technical white papers and other document materials, and establishing a directory for industry websites and report publishing platforms, such as Frost&Sullivan, MarketsandMarkets, etc. Use OCR (Optical Character Recognition) technology to process reports in PDF format, extract text content, implement an automated monitoring mechanism, and obtain the latest industry information in a timely manner. The above data is subjected to regular expressions to remove noise such as HTML tags and special characters, and then NLTK (Natural Language Toolkit) is used for text segmentation, part-of-speech tagging and named entity recognition.

[0059] The processed raw data is vectorized using a pre-trained language model to construct a high-dimensional vector space. In this embodiment, a BERT-Base or SciBERT model is used and fine-tuned for the field of materials science, and the Hugging Face Transformers library is used to load and apply the model.

[0060] Entity relationships are extracted from the vectorized raw data to construct a knowledge network for polymer material modification. In this embodiment, SpaCy is used for named entity recognition to identify entities such as material names, attributes, and methods. A relationship extraction algorithm based on dependency syntactic analysis is used to identify semantic relationships between entities. The Neo4j graph database is used to store and manage knowledge graphs. Knowledge representation based on graph embedding is implemented, such as the TransE or RotatE algorithm.

[0061] A large language model is used to generate key concept explanations and technical principle explanations based on the knowledge network to obtain a knowledge system containing entities, relationships and attributes. Specifically, GPT-3.5 or similar large-scale language models, such as PaLM or OPT, are used; by designing a specific prompt template, the model is guided to generate high-quality concept explanations, a few-shot learning mechanism is implemented, and examples written by domain experts are provided to improve the generation quality. In other embodiments of the present invention, a post-processing module is also included, which is used to check and correct potential errors in the generated content.

[0062] In a preferred embodiment of the present invention, after the professional knowledge system for polymer material modification is constructed, the professional knowledge system is also evaluated, and a prompt is issued when any evaluation index is lower than a preset value. The evaluation indexes include:

[0063] New material coverage, assessing the knowledge base's coverage of the latest material categories and checking the completeness of the collection of cutting-edge synthesis methods and characterization techniques;

[0064] Accuracy of material performance data: random sampling to verify key material performance data and cross-validate with the experimental database;

[0065] The integrity of material structure information, evaluates the integrity of crystal structure, electronic structure, and defect information.

[0066] When collecting data in step one, an automated data collection pipeline is established to ensure the timeliness and comprehensiveness of the data; key indicators include data volume, update frequency, source diversity, and data integrity to ensure the breadth and depth of the knowledge base and provide a data foundation for subsequent analysis.

[0067] Step 2: Build a search enhancement generation system, obtain intermediate data from the professional knowledge system according to the user's input requirements, evaluate the intermediate data generated by the search enhancement generation system, design an evaluation index system, use a large language model to perform natural language understanding on the intermediate data, analyze the matching degree between the data content and the query requirements, and retain the data content that meets the matching requirements.

[0068] The evaluation of the intermediate data generated by the retrieval enhancement generation system is as follows:

[0069] Constructing an evaluation index system, which includes data accuracy, timeliness and relevance, and calculating scores for each index;

[0070] Use a large language model to understand natural language, analyze the matching degree between intermediate data and query requirements, and use the BERT-based model to calculate the semantic similarity between the query and the search results; use the text implication recognition module to determine whether the search results contain the information required for the query; and give the calculation scores for each item;

[0071] Generate an evaluation report, build a scoring weighting algorithm, combine various indicators to get an overall score, and finally mark high-quality data and potential problem data, where high-quality data can be used for the next step of processing, and potential problem data can be manually reviewed and corrected. In one embodiment of the present invention, in order to realize an automated report generator, a library such as matplotlib is used to visualize the evaluation results.

[0072] The data accuracy index may be a consistency score with references (0-1 points).

[0073] In one embodiment of the present invention, the timeliness score for evaluating the intermediate data generated by the search enhancement generation system is calculated as:

[0074] f(t)=e -λ(T-t)

[0075] Where f(t) is the timeliness score, t is the data release date, T is the current date, and λ is the set attenuation coefficient.

[0076] In another embodiment of the present invention, the timeliness score for evaluating the intermediate data generated by the retrieval enhancement generation system is calculated as:

[0077]

[0078] Where f(t) is the timeliness score, t is the data release date, T is the current date, and β is the set attenuation coefficient.

[0079] Among them, the relevance calculation method combines TF-IDF (Term Frequency–Inverse Document Frequency) and semantic similarity. In TF-IDF calculation, first construct the TF-IDF matrix, convert the document set (or text set) into a TF-IDF matrix, where each document or text is a vector, and use these vectors to represent the characteristics of the document; then calculate the cosine similarity, and use the cosine similarity to calculate the TF-IDF similarity of two documents or texts. In semantic similarity calculation, use pre-trained language models (such as BERT, Sentence-BERT, GloVe, etc.) to convert text into fixed-size embedding vectors, which capture the semantic information of the text; then calculate the cosine similarity of the embedding vectors, and use the cosine similarity to calculate the semantic similarity between the two embedding vectors. Through the TF-IDF algorithm and semantic similarity calculation, the TF-IDF similarity and semantic similarity are combined according to certain weights, and the calculation method is as follows:

[0080] CS=α·TFIDF+(1-α)·SS

[0081] Where CS is the similarity after fusion, α is the fusion parameter, which takes a value of [0,1], TFIDF is the calculated TF-IDF similarity, and SS is the calculated semantic similarity. The fusion parameter can be optimized based on experience or through cross-validation.

[0082] Step 3: Input the polymer material modification requirements or optimization requirements into the search enhancement generation system to obtain matching content.

[0083] Step 4: Use the large language model to generate answers through the matching content to obtain polymer material modification or optimization solutions. Integrate the retrieved matching content into structured information (such as key points, summaries, key information lists, etc.). Convert the retrieval results represented by structured information into a prompt format that can be understood by the large language model, usually by prepending the retrieved knowledge as context information or background knowledge before the prompt input by the user. Or, the results retrieved from the database are structured and used as one of the inputs of the model, and are plugged in in the form of a prompt. The large language model generates a modification plan or technical guidance that meets the requirements based on the prompt and user query.

[0084] In step 2, the retrieval enhancement generation includes data preprocessing, vector indexing, semantic similarity calculation and context-related retrieval strategy;

[0085] The data preprocessing is used for data cleaning and data distributed processing;

[0086] The vector index selects HNSW (Hierarchical Navigable Small World) as the core index structure, and uses the Faiss library to implement HNSW index; uses the Faiss library to implement HNSW index, sets the maximum outgoing edge of each node; optimizes index parameters, such as setting the efConstruction parameter to 400 to improve index quality. It is also possible to build a distributed index and use multi-process parallelization to accelerate the indexing process of large-scale data.

[0087] The semantic similarity calculation uses cosine similarity, Jaccard similarity or Euclidean distance to calculate the semantic similarity; the cosine similarity calculation is implemented using the NumPy library. In addition to cosine similarity, other similarities such as Jaccard similarity and Euclidean distance can also be introduced. A suitable similarity algorithm can be selected according to the query features.

[0088] The context-related retrieval strategy uses WordNet or a domain-specific ontology library to expand queries, and uses an attention mechanism to perform multiple rounds of retrieval to narrow the scope of results; query expansion modules can be constructed to expand query keywords using WordNet or domain ontology, and domain knowledge rules such as range constraints of material performance indicators can be integrated to improve retrieval accuracy.

[0089] In other embodiments of the present invention, after the search enhancement generation system in step 2 is constructed, the search module of the search enhancement generation system is also evaluated. If the evaluation fails, the search enhancement generation system needs to be manually rebuilt or repaired. The indicators include:

[0090] The ability to identify professional material terms, and the accuracy of the search enhancement generation system in identifying specific terms, chemical formulas, and structural descriptors in the new material field;

[0091] Multimodal search capability, the ability of the search enhancement generation system to handle multimodal input, including text, chemical structure diagrams, and material spectra;

[0092] Associative retrieval capability, the ability of the retrieval enhancement generation system to associate material properties, synthesis methods, and application fields.

[0093] Among them, when evaluating the ability to recognize professional material terminology, the following methods can be used for intervention and repair: pre-training of material science literature using deep learning models to improve the system's understanding of rare and emerging terms; establishing a dynamically updated material terminology knowledge graph to capture new words and terminology evolution in the field in real time; using contextual information and semantic analysis technology to improve the accurate recognition of polysemous words and synonyms.

[0094] When evaluating multimodal retrieval capabilities, the following methods can be used for intervention and repair: develop a cross-modal attention mechanism to achieve deep fusion of text, image and spectrogram data; design a multimodal encoder to map information of different modalities into a unified vector space; introduce contrastive learning methods to enhance the semantic consistency between multimodal data; construct a multimodal knowledge graph to capture the implicit relationship between data of different modalities.

[0095] When evaluating the association retrieval capability, the following methods can be used for intervention and repair: use the graph neural network model to build a multi-level knowledge graph of materials, performance, and applications; develop an association analysis module based on causal reasoning to explore the causal relationship between material properties and performance; introduce transfer learning technology to transfer knowledge in the field of known materials to the association analysis of new materials; design a multi-task learning framework to simultaneously optimize material performance prediction, synthesis path planning, and application scenario matching.

[0096] Embodiment 2

[0097] This embodiment is a polymer material modification optimization system based on RAG, and the system implements the optimization method as described in the first embodiment, including:

[0098] Data collection module, used to collect raw data from professional journals, patent databases, scientific research papers, experimental reports, material manuals, technical standards and industry reports related to polymer material modification;

[0099] A knowledge base module, used to construct a professional knowledge system of polymer material modification based on the original data;

[0100] A search and query module, used to query matching content from the professional knowledge system according to the input modification requirements or optimization requirements;

[0101] A large language model is used to construct prompt words according to the matching content, and generate answers using the prompt words.

[0102] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, which all fall within the protection scope of the present invention.

Claims

1. A RAG-based polymer material modification optimization method, characterized in that: The following steps are involved: Build a professional knowledge system for polymer material modification; Constructing a search enhancement generation system, obtaining intermediate data from the professional knowledge system according to the requirements input by the user, evaluating the intermediate data generated by the search enhancement generation system, designing an evaluation index system, using a large language model to perform natural language understanding on the intermediate data, and analyzing the matching degree between the data content and the query requirements; Inputting polymer material modification requirements or optimization requirements into the search enhancement generation system to obtain matching content; A large language model is used to generate answers based on the matching content to obtain a polymer material modification plan or optimization plan.

2. The RAG-based polymer material modification optimization method according to claim 1, characterized in that: After the method completes the construction of the polymer material modification professional knowledge system, the professional knowledge system is also evaluated, and a prompt is issued when any evaluation index is lower than a preset value. The evaluation indexes include: New material coverage, assessing the knowledge base's coverage of the latest material categories and checking the completeness of the collection of cutting-edge synthesis methods and characterization techniques; Accuracy of material performance data: random sampling to verify key material performance data and cross-validate with the experimental database; The integrity of material structure information, evaluates the integrity of crystal structure, electronic structure, and defect information.

3. The RAG-based polymer material modification optimization method according to claim 1, characterized in that: The method for constructing the knowledge system is specifically as follows: Collect raw data related to polymer material modification, clean and format the raw data, including noise removal, terminology unification, word segmentation, part-of-speech tagging and entity recognition operations, and convert the processed data into a structured knowledge representation form; Vectorize the processed raw data using a pre-trained language model to construct a high-dimensional vector space; Extract entity relationships from vectorized raw data and construct a knowledge network for polymer material modification; A large language model is used to generate key concept explanations and technical principle elaborations based on the knowledge network to obtain a knowledge system including entities, relationships and attributes.

4. The RAG-based polymer material modification optimization method according to claim 3, characterized in that: The original data include professional journals, patent databases, scientific research papers, experimental reports, material manuals, technical standards and industry reports.

5. The RAG-based polymer material modification optimization method according to claim 1, characterized in that: The evaluation of the intermediate data generated by the retrieval enhancement generation system is as follows: Constructing an evaluation index system, wherein the evaluation index system includes data accuracy, timeliness and relevance; Use a large language model to understand natural language, analyze the matching degree between data content and query requirements, and use a BERT-based model to calculate the semantic similarity between the query and the search results; use the text implication recognition module to determine whether the search results contain the information required for the query; Generate an assessment report, mark high-quality data and potentially problematic data, build a scoring weighting algorithm, and combine various indicators to derive an overall score.

6. The RAG-based polymer material modification optimization method according to claim 5, characterized in that: The timeliness score of the evaluation index system is calculated as: f(t)=e-4(Tt) Where f(t) is the timeliness score, t is the data release date, T is the current date, and λ is the set attenuation coefficient.

7. The RAG-based polymer material modification optimization method according to claim 5, characterized in that: The timeliness score of the evaluation index system is calculated as: Where f(t) is the timeliness score, t is the data release date, T is the current date, and β is the set attenuation coefficient.

8. The RAG-based polymer material modification optimization method according to claim 1, characterized in that: The retrieval enhancement generation includes data preprocessing, vector indexing, semantic similarity calculation and context-related retrieval strategy; The data preprocessing is used for data cleaning and data distributed processing; The vector index selects HNSW as the core index structure and uses the Faiss library to implement the HNSW index; The semantic similarity calculation uses cosine similarity, Jaccard similarity or Euclidean distance to calculate the semantic similarity; The context-dependent retrieval strategy uses WordNet or a domain-specific ontology library for query expansion and uses an attention mechanism to perform multiple rounds of retrieval to narrow the scope of results.

9. The RAG-based polymer material modification optimization method according to claim 1, characterized in that: After the search enhancement generation system is built, the search module of the search enhancement generation system is also evaluated, and the indicators include: The ability to identify professional material terms, and the accuracy of the search enhancement generation system in identifying specific terms, chemical formulas, and structural descriptors in the new material field; Multimodal search capability, the ability of the search enhancement generation system to handle multimodal input, including text, chemical structure diagrams, and material spectra; Associative retrieval capability, the ability of the retrieval enhancement generation system to associate material properties, synthesis methods, and application fields.

10. A polymer material modification optimization system based on RAG, characterized in that: The system implements the optimization method according to any one of claims 1 to 9, including: Data collection module, used to collect raw data from professional journals, patent databases, scientific research papers, experimental reports, material manuals, technical standards and industry reports related to polymer material modification; A knowledge base module, used to construct a professional knowledge system of polymer material modification based on the original data; A search and query module, used to query matching content from the professional knowledge system according to the input modification requirements or optimization requirements; A large language model is used to construct prompt words according to the matching content, and generate answers using the prompt words.

Images