International patent classification method and Chinese library classification method conversion method

Abstract

The invention belongs to the technical field of patent classification method conversion, and particularly relates to a method for converting an international patent classification method and a Chinese library classification method. The method comprises the following steps of: 1, constructing an international patent classification semantic vector set I: fusing various target word vector sets Iw, concept vector sets Ic and phrase vector sets Ip of an international patent classification (IPC); 2, constructing a Chinese library classification semantic vector set C: fusing various target word vector sets Cw, concept vector sets Cc and phrase vector sets Cp of a Chinese library classification method CLC; step 3, performing IPC and CLC category mapping; and 4, verifying the mapping result by using an IPC and CLC category mapping sample set established by an expert. According to the method, the patent and the scientific and technical literature can be retrieved and analyzed uniformly in the precise subdivision field limited by the same classification number, switching operation in different databases is not needed, information separation is avoided, retrieval and analysis efficiency is improved, comprehensive and accurate intelligence support is provided, and integrity and continuity of intelligence information are guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of patent classification conversion technology, specifically involving a method for converting the International Patent Classification (IPC) and the Chinese Library Classification (CLC). Background Technology

[0002] Patents and non-patent documents such as journals and dissertations are all important products of scientific and technological research, reflecting the research results of scientific research and technological invention respectively. The knowledge flow between the two jointly determines the evolutionary path of scientific innovation in various fields. Revealing the knowledge transformation path and synergistic relationship between scientific research and technological application provides effective support for intellectual property management, innovation discovery, and scientific research planning.

[0003] Linking knowledge between patents and scientific papers hinges on achieving interoperability between the two, enabling effective integration and unified retrieval and analysis of both. Patent resources and academic papers each have their own independent classification criteria: internationally, the International Patent Classification (IPC) is widely used for patents; domestically, the Chinese Library Classification (CLC) is generally employed. Currently, neither domestically nor internationally, there is a classification system applicable to both patents and academic papers. Therefore, establishing a classification mapping between IPC and CLC can help identify patents and papers within the same research field, promote knowledge flow and information sharing between patents and papers, and break down knowledge silos between science and technology. Summary of the Invention

[0004] The purpose of this invention is to provide a method for converting between the International Patent Classification (IPC) and the Chinese Library Classification (CLC). It proposes methods for constructing word sets, concept sets, and segment vector sets for each category in IPC and CLC, and employs models to vectorize word, sentence, and segment data. Based on weighted fusion, similarity calculation, and ranking filtering, a linking method for constructing the IPC-CLC mapping relationship is developed, significantly improving the coverage and accuracy of the matching results. This enables the establishment of a correspondence between IPC and CLC classification numbers and the associated querying of documents under different classification systems. Furthermore, it allows for unified retrieval and analysis of patents and scientific literature within precise sub-fields limited by the same classification number, eliminating the need to switch between different databases, avoiding information separation, improving retrieval and analysis efficiency, providing comprehensive and accurate intelligence support, and ensuring the integrity and consistency of intelligence information.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC):

[0007] Step 1: Construct the semantic vector set I of the International Patent Classification (IPC): Integrate the word vector set Iw, concept vector set Ic, and segment vector set Ip of each category in the IPC.

[0008] Step 2: Construct the semantic vector set C of the Chinese Library Classification: integrate the word vector set Cw, concept vector set Cc, and segment vector set Cp of each category in the Chinese Library Classification (CLC).

[0009] Step 3: Mapping IPC and CLC categories;

[0010] Step 4 uses an expert-established IPC to CLC category mapping sample set to verify the above mapping results.

[0011] Step 1 specifically includes:

[0012] Step 1.1 Extraction of patent document sample set: X non-duplicate patents are extracted from each IPC classification number in descending order of publication time to form a patent document sample set;

[0013] Step 1.2 Construct an international patent classification terminology set and use a model for vectorization processing; the terminology set includes, but is not limited to, subject terms corresponding to IPC categories, machine-marked keywords, expert indexing keywords, and keywords extracted by the large language model from the patent document sample set corresponding to IPC categories;

[0014] Step 1.3 The International Patent Classification Concept Set Ic includes the overarching concepts, ontology, knowledge graph, and terminology definitions for the International Patent Classification number categories;

[0015] Step 1.4 International Patent Classification Segment Vector Set Ip: The segments in the sample set are vectorized using a semantic vector model, features are fused, and the feature mean vector of each category of patent sample set is calculated.

[0016] After completing step 1.5 of constructing the word vector set Iw, concept vector set Ic, and segment vector set Ip for the International Patent Classification, the semantic information of these three levels is integrated into a unified, strongly expressive composite semantic vector through a dynamic weighted fusion algorithm.

[0017] Step 1.1: X is a dynamic value, X = min(1000, total number of patents under this IPC category * 0.7), that is, 1000 or 70% of the patents under this category are selected, whichever is less, and duplicates are eliminated based on the patent publication number.

[0018] Step 1.2: IPC subject terms are directly derived from the official category titles and annotations of the IPC classification table; machine-generated keywords are automatically extracted from the titles, abstracts, and claims of the sample set patents using the TF-IDF algorithm, retaining the top-K words with the highest weights; expert indexing keywords: keywords manually indexed by experts are imported from the database; keywords are extracted using a large language model: a finely tuned large language model is used, with the prompt: Please extract the core technical keywords from the following patent text [title / abstract], separated by commas, for batch processing.

[0019] After lowercase conversion, word form restoration, and stop word removal of the vocabulary from the above sources, deduplication is performed based on the stem, retaining m words. The model is used for vectorization processing, and each word w_i generates a vector v_word_i. The vectorized representation of each IPC category word set is Iw={v_word_1,v_word_2,...,v_word_m}. The comprehensive word vector of IPC category a is Iw_avg=(1 / m)*Σv_word_1.

[0020] Step 1.3: The higher-level concepts are automatically obtained through the IPC classification tree structure, and each concept node is associated with its official description; the ontology and knowledge graph map IPC categories to relevant concept entities from external knowledge bases through entity linking technology; the terminology definitions are obtained from the official IPC guidelines, technical dictionaries, or category technical definitions generated by large language models.

[0021] For each concept c_j, all its textual description information is concatenated into a comprehensive text Text(c_j). The domain-adapted semantic vector model is used to encode Text(c_j) to obtain the concept vector v_concept_j. The International Patent Classification concept set Ic is represented as the set of all concept vectors: Ic = {v_concept_1, v_concept_2, ..., v_concept_n}. The center vector of the relevant concept vectors of IPC category a is Ic_cen = (1 / n)*Σv_concept_j.

[0022] Step 1.4: A unified semantic model fine-tuned by the domain is adopted to ensure the consistency of the vector space. For a specific IPC category a, there are X patent samples under it. The segment vector set Ip_a of this category is calculated, which is the set of all segment vectors. The feature mean vector V_ipc_a of this category is calculated as the semantic center of this category. The category mean vector is: Ip_cen=(1 / X)*Σv_patent_m, where v_patent_m is the document vector of the m-th patent under category a. The international patent classification segment vector set Ip is represented as Ip={V_ipc_1,V_ipc_2,...,V_ipc_n}.

[0023] Step 1.5: Configure the fusion weight parameters α,β,γ, where α,β,γ≥0 and α+β+γ=1, and the composite semantic vector Ia of category a of international patent classification number is α*Normalize(Iw_avg)+β*Normalize(Ic_cen+γ)*Normalize(Ip_cen), where Normalize() represents L2 normalization, which maps the vector to a unit hypersphere to make the cosine similarity calculation more stable.

[0024] Step 2 specifically includes:

[0025] Step 2.1 Extraction of non-patent literature sample set: X non-duplicate documents are extracted from each CLC classification number in descending order of publication time to form a non-patent literature sample set;

[0026] Step 2.2 Construct a Chinese Library Classification Thesaurus and use a model for vectorization; the thesaurus includes, but is not limited to, subject terms corresponding to CLC categories, machine-marked keywords for non-patent documents, author indexing keywords, expert indexing keywords, and keywords extracted by the large language model corresponding to CLC categories;

[0027] Step 2.3 The Chinese Library Classification Concept Set Cc includes the superordinate concepts, ontology, knowledge graph, and terminology definitions of the Chinese Library Classification Number category;

[0028] Step 2.4 The Chinese Library Classification Segment Vector Set Cp uses a semantic vector model to vectorize the segments in the sample set, fuses features, and calculates the feature mean vector of each category of document sample set.

[0029] Step 2.5 generates a composite semantic vector for CLC category b, Cb=α*Normalize(Cw_avg)+β*Normalize(Cc_cen)+γ*Normalize(Cp_cen), with the weight parameters α,β,γ fused, where α,β,γ≥0 and α+β+γ=1.

[0030] Step 2.1: X is a dynamic value, X = min(1000, total number of documents under this CLC category * 0.7), that is, extract 1000 documents or 70% of the documents under this category, whichever is less, and deduplicate based on the document DOI.

[0031] Step 2.2: CLC subject terms are taken from category names, annotations, and the "Chinese Classification Subject Headings"; machine-generated keywords are automatically extracted from sample documents using the TF-IDF algorithm, retaining the top-K words with the highest weights; author indexing keywords are directly taken from the author keywords in the document metadata; expert indexing keywords are extracted from the standardized keywords of high-quality journals; the large language model extracts keywords using a finely tuned large language model with the prompt: Please extract the keywords that best represent the core research content from the following scientific and technological documents [title / abstract], requiring both professionalism and universality.

[0032] After converting simplified and traditional Chinese characters and unifying full-width and half-width characters for the vocabulary from the above sources, and standardizing the terminology, deduplication is performed, retaining m words. The model is used for vectorization processing, and each word w_i generates a vector v_w_i. The vectorized representation of each category word set is Cw={v_w_1,v_w_2,...,v_w_m}. The comprehensive word vector of CLC category b is Cw_avg=(1 / m)*Σv_w_1.

[0033] Step 2.3: The higher-level concepts are generated through the CLC tree structure; the ontology and knowledge graph map the CLC categories to the relevant concept entities from the external knowledge base through entity linking technology; the terminology definitions are obtained from the category technical definitions generated by the user manual, technical dictionary, or large language model.

[0034] For each concept c_j, all its textual description information is concatenated into a comprehensive text Text(c_j). The domain-adapted semantic vector model is used to encode Text(c_j) to obtain the concept vector v_concept_j. The CLC classification concept set Cc is represented as the set of all concept vectors: Cc = {v_concept_1, v_concept_2, ..., v_concept_n}. The category concept center vector Cc_cen = (1 / n)*Σv_concept_j.

[0035] Step 2.4: A unified semantic model with domain fine-tuning is adopted to ensure the consistency of the vector space; for a specific CLC category b, there are X document samples under it. Calculate the segment vector set Cp_b of this category, that is, the set of all segment vectors. The category center vector of category b is: Cp_cen=(1 / X)*Σv_doc_m, where v_doc_m is the document vector of the m-th patent under category b.

[0036] Step 3 specifically involves: calculating the composite semantic vector for each classification number, sequentially calculating the similarity between Ia of each IPC category and Cb of each CLC category, Sim(I,C)=CosineSimilarity(Ia,Cb), and determining that IPC classification number a and CLC classification number b are similar categories when the similarity threshold is greater than λ. The Top-N CLC categories are used as the final mapping result, and this process generates a mapping from CLC to IPC.

[0037] Step 4: Optimize and adjust the parameters α, β, γ, and λ.

[0038] The beneficial effects achieved by this invention are as follows:

[0039] Multi-granularity fusion: It overcomes the limitations of single-word or text representation models by fusing word vectors with paragraphs, capturing the semantics of terminology details, conceptual categories, and overall discourse, thereby improving the accuracy of mapping.

[0040] Domain Adaptive: By using the same data type to represent both patents and non-patents as vectors, this adapts to the same domain technologies across different document types, improving the comprehensiveness of the mapping.

[0041] Dynamic weights: Automatically adjust the fusion weights to make the model more adaptable to the semantic representation of categories with different granularities. Detailed Implementation

[0042] The present invention will now be described in detail with reference to specific embodiments.

[0043] A method for converting between the International Patent Classification and the Chinese Library Classification includes the following steps:

[0044] Step 1: Construct the semantic vector set I of the International Patent Classification (IPC): Integrate the word vector set Iw, concept vector set Ic, and segment vector set Ip of each category in the International Patent Classification (IPC).

[0045] Step 1.1 Patent Document Sample Set Extraction: For each IPC classification number, X non-duplicate patents are extracted in descending order of publication date to form a patent document sample set. Here, X is a dynamic value, X = min(1000, total number of patents under this IPC category * 0.7), meaning 1000 patents or 70% of the patents under this classification (whichever is less) are extracted, and duplicates are removed based on the patent publication number.

[0046] Step 1.2 Constructs an International Patent Classification (IPC) terminology set and uses a model for vectorization. The terminology set includes, but is not limited to, subject terms corresponding to IPC categories, machine-defined keywords, expert-indexed keywords, and keywords extracted by a large language model from the patent document sample set corresponding to IPC categories. IPC subject terms are directly derived from the official category titles and annotations of the IPC classification table. Machine-defined keywords: Automatically extracted from the titles, abstracts, and claims of the patents in the sample set using the TF-IDF algorithm, retaining the top-K words with the highest weights. Expert-indexed keywords: Imported keywords manually indexed by experts from the database. Keywords extracted by the large language model: Using a finely tuned large language model with the prompt: "Please extract the core technical keywords from the following patent text [title / abstract], separated by commas.", batch processing is performed. After lowercaseing, word form restoration, and removal of stop words from the above sources, deduplication is performed based on stemming, retaining m words. The model is used for vectorization processing. Each word w_i generates a vector v_word_i, and the vectorized representation of each IPC category word set is Iw={v_word_1,v_word_2,...,v_word_m}. The comprehensive word vector of IPC category a is Iw_avg=(1 / m)*Σv_word_1.

[0047] Step 1.3 The International Patent Classification (IPC) concept set Ic includes the superordinate concepts, ontology, knowledge graph, and terminology definitions of the IPC category. Superordinate concepts: Automatically obtained through the IPC classification tree structure; each concept node can be associated with its official description. Ontology and knowledge graph: Mapped from external knowledge bases to relevant concept entities using entity linking technology. Terminology definitions: Obtained from IPC official guidelines, technical dictionaries, or category technical definitions generated by large language models. For each concept c_j, all its textual description information is concatenated into a comprehensive text Text(c_j). Using the same domain-adapted semantic vector model as in Step 1.2, Text(c_j) is encoded to obtain the concept vector v_concept_j. The IPC concept set Ic can be represented as the set of all concept vectors: Ic = {v_concept_1, v_concept_2, ..., v_concept_n}. The central vector of the relevant concept vectors for IPC category a is Ic_cen = (1 / n)*Σv_concept_j.

[0048] Step 1.4: The International Patent Classification (IPC) segment vector set Ip uses a semantic vector model to vectorize the segments in the sample set, fuses features, and calculates the feature mean vector for each category of patent sample set. A domain-adjusted semantic model, consistent with steps 1.2 and 1.3, is used to ensure consistency in the vector space. For a specific IPC category a, there are X patent samples. Calculate the segment vector set Ip_a for this category, which is the set of all segment vectors. Calculate the feature mean vector V_ipc_a for this category, serving as the semantic center of this category. Category mean vector: Ip_cen=(1 / X)*Σv_patent_m, where v_patent_m is the document vector of the m-th patent under category a. The IPC segment vector set Ip can be represented as Ip={V_ipc_1,V_ipc_2,...,V_ipc_n}.

[0049] Step 1.5: After completing steps 1.1 to 1.4 and constructing the word vector set (Iw), concept vector set (Ic), and segment vector set (Ip) for the International Patent Classification (IPC), a dynamic weighted fusion algorithm is designed to integrate the semantic information from these three levels into a unified, strongly expressive composite semantic vector. The configurable fusion weight parameters α, β, γ are provided, where α, β, γ ≥ 0 and α + β + γ = 1. The composite semantic vector Ia for IPC category a is defined as: Ia = α * Normalize(Iw_avg) + β * Normalize(Ic_cen + γ) * Normalize(Ip_cen), where Normalize() represents L2 normalization, mapping the vector to a unit hypersphere to make cosine similarity calculation more stable.

[0050] Step 2: Construct the semantic vector set C of the Chinese Library Classification: integrate the word vector set Cw, concept vector set Cc, and paragraph vector set Cp of each category in the Chinese Library Classification (CLC).

[0051] Step 2.1 Non-Patent Literature Sample Set Extraction: For each CLC classification number, X non-duplicate documents are extracted in descending order of publication time to form the non-patent literature sample set. Here, X is a dynamic value, X = min(1000, total number of documents under this CLC category * 0.7), meaning 1000 documents or 70% of the documents under this category (whichever is less) are extracted, and duplicates are eliminated based on the document DOI.

[0052] Step 2.2 Constructs a Chinese Library Classification (CLC) thesaurus and uses a model for vectorization. The thesaurus includes, but is not limited to, subject terms corresponding to CLC categories, machine-generated keywords for non-patent documents, author-indexed keywords, expert-indexed keywords, and keywords extracted by the large language model. CLC subject terms are derived from category names, annotations, and the *Chinese Classification Thesaurus*. Machine-generated keywords: automatically extracted from sample documents using the TF-IDF algorithm. The top-K keywords with the highest weights are retained. Author-indexed keywords: directly used from the author keywords in the document metadata. Expert-indexed keywords: extracted from standardized keywords in high-quality journals. Keyword extraction by the large language model: using a finely tuned large language model with the prompt: "Please extract the keywords that best represent the core research content from the following scientific and technological documents [title / abstract], requiring both professionalism and generality." The vocabulary from the above sources is converted from traditional Chinese to simplified Chinese, unified with full-width and half-width characters, and standardized in terms of terminology before deduplication, retaining m keywords. The same model as in step 1.2 is used for vectorization processing. Each word w_i generates a vector v_w_i, and the vectorized representation of each category word set is Cw={v_w_1,v_w_2,...,v_w_m}. The comprehensive word vector of CLC category b is Cw_avg=(1 / m)*Σv_w_1.

[0053] Step 2.3 The CLC classification concept set Cc includes the superordinate concepts, ontology, knowledge graph, and terminology definitions of the CLC classification categories. Superordinate concepts: generated through the CLC tree structure. Ontology and knowledge graph: mapped from external knowledge bases to relevant concept entities using entity linking technology. Terminology definitions: obtained from user manuals, technical dictionaries, or category technical definitions generated by large language models. For each concept c_j, all its textual description information is concatenated into a comprehensive text Text(c_j). Using the same domain-adapted semantic vector model as in Step 2.2, Text(c_j) is encoded to obtain the concept vector v_concept_j. The CLC classification concept set Cc can be represented as the set of all concept vectors: Cc = {v_concept_1, v_concept_2, ..., v_concept_n}. The category concept center vector Cc_cen = (1 / n)*Σv_concept_j.

[0054] Step 2.4: The segment vector set Cp of the Chinese Library Classification uses a semantic vector model to vectorize the segments in the sample set, fuses features, and calculates the feature mean vector of each category's document sample set. A semantic model consistent with steps 2.2 and 2.3, but finely tuned for the domain, is used to ensure consistency of the vector space. For a specific CLC category b, there are X document samples. Calculate the segment vector set Cp_b for this category, which is the set of all segment vectors. The category center vector of category b: Cp_cen=(1 / X)*Σv_doc_m, where v_doc_m is the document vector of the m-th patent under category b.

[0055] Step 2.5 generates a composite semantic vector for CLC category b, Cb = α*Normalize(Cw_avg) + β*Normalize(Cc_cen) + γ*Normalize(Cp_cen), where α, β, and γ are the same as in step 1.5.

[0056] Step 3: Mapping IPC and CLC Categories. Based on the composite semantic vector of each classification number calculated in steps 1.5 and 2.5, the similarity between each IPC category Ia and CLC category Cb is calculated sequentially: Sim(I,C) = CosineSimilarity(Ia,Cb). When the similarity threshold is greater than λ, IPC classification number a and CLC classification number b are considered similar categories. The Top-N CLC categories are used as the final mapping results. At the same time, this process generates a mapping from CLC to IPC.

[0057] Step 4 uses an expert-established IPC and CLC category mapping sample set to verify the above mapping results and to debug and optimize the parameters α, β, γ, and λ.

Claims

1. A method for converting between the International Patent Classification and the Chinese Library Classification, characterized in that: Step 1: Construct the semantic vector set I of the International Patent Classification (IPC): Integrate the word vector set Iw, concept vector set Ic, and segment vector set Ip of each category in the IPC. Step 2: Construct the semantic vector set C of the Chinese Library Classification: integrate the word vector set Cw, concept vector set Cc, and segment vector set Cp of each category in the Chinese Library Classification (CLC). Step 3: Mapping IPC and CLC categories; Step 4 uses an expert-established IPC to CLC category mapping sample set to verify the above mapping results.

2. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 1, characterized in that: Step 1 specifically includes: Step 1.1 Extraction of patent document sample set: X non-duplicate patents are extracted from each IPC classification number in descending order of publication time to form a patent document sample set; Step 1.2 Construct an international patent classification terminology set and use a model for vectorization processing; the terminology set includes, but is not limited to, subject terms corresponding to IPC categories, machine-marked keywords, expert indexing keywords, and keywords extracted by the large language model from the patent document sample set corresponding to IPC categories; Step 1.3 The International Patent Classification Concept Set Ic includes the overarching concepts, ontology, knowledge graph, and terminology definitions for the International Patent Classification number categories; Step 1.4 International Patent Classification Segment Vector Set Ip: The segments in the sample set are vectorized using a semantic vector model, features are fused, and the feature mean vector of each category of patent sample set is calculated. After completing step 1.5 of constructing the word vector set Iw, concept vector set Ic, and segment vector set Ip for the International Patent Classification, the semantic information of these three levels is integrated into a unified, strongly expressive composite semantic vector through a dynamic weighted fusion algorithm.

3. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 2, characterized in that: Step 1.1: X is a dynamic value, X = min(1000, total number of patents under this IPC category * 0.7), that is, 1000 or 70% of the patents under this category are selected, whichever is less, and duplicates are eliminated based on the patent publication number.

4. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 2, characterized in that: Step 1.2: IPC subject terms are directly derived from the official category titles and annotations of the IPC classification table; machine-generated keywords are automatically extracted from the titles, abstracts, and claims of the sample set patents using the TF-IDF algorithm, retaining the top-K words with the highest weights; expert indexing keywords: keywords manually indexed by experts are imported from the database; keywords are extracted using a large language model: a finely tuned large language model is used, with the prompt: Please extract the core technical keywords from the following patent text [title / abstract], separated by commas, for batch processing.

5. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 4, characterized in that: After lowercase conversion, word form restoration, and stop word removal of the vocabulary from the above sources, deduplication is performed based on the stem, retaining m words. The model is used for vectorization processing, and each word w_i generates a vector v_word_i. The vectorized representation of each IPC category word set is Iw={v_word_1,v_word_2,...,v_word_m}. The comprehensive word vector of IPC category a is Iw_avg=(1 / m)*Σv_word_1.

6. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 2, characterized in that: Step 1.3: The higher-level concepts are automatically obtained through the IPC classification tree structure, and each concept node is associated with its official description; the ontology and knowledge graph map IPC categories to relevant concept entities from external knowledge bases through entity linking technology; the terminology definitions are obtained from the official IPC guidelines, technical dictionaries, or category technical definitions generated by large language models.

7. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 6, characterized in that: For each concept c_j, all its textual description information is concatenated into a comprehensive text Text(c_j). The domain-adapted semantic vector model is used to encode Text(c_j) to obtain the concept vector v_concept_j. The International Patent Classification concept set Ic is represented as the set of all concept vectors: Ic = {v_concept_1, v_concept_2, ..., v_concept_n}. The center vector of the relevant concept vectors of IPC category a is Ic_cen = (1 / n)*Σv_concept_j.

8. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 2, characterized in that: Step 1.4: A unified semantic model fine-tuned by the domain is adopted to ensure the consistency of the vector space. For a specific IPC category a, there are X patent samples under it. The segment vector set Ip_a of this category is calculated, which is the set of all segment vectors. The feature mean vector V_ipc_a of this category is calculated as the semantic center of this category. The category mean vector is: Ip_cen=(1 / X)*Σv_patent_m, where v_patent_m is the document vector of the m-th patent under category a. The international patent classification segment vector set Ip is represented as Ip={V_ipc_1,V_ipc_2,...,V_ipc_n}.

9. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 2, characterized in that: Step 1.5: Configure the fusion weight parameters α,β,γ, where α,β,γ≥0 and α+β+γ=1, and the composite semantic vector Ia of category a of international patent classification number is α*Normalize(Iw_avg)+β*Normalize(Ic_cen+γ)*Normalize(Ip_cen), where Normalize() represents L2 normalization, which maps the vector to a unit hypersphere to make the cosine similarity calculation more stable.

10. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 1, characterized in that: Step 2 specifically includes: Step 2.1 Extraction of Non-Patent Literature Sample Set: X non-duplicate documents are extracted from each CLC classification number in descending order of publication time to form a non-patent literature sample set; Step 2.2 Construct a Chinese Library Classification Thesaurus and use a model for vectorization; the thesaurus includes, but is not limited to, subject terms corresponding to CLC categories, machine-marked keywords for non-patent documents, author indexing keywords, expert indexing keywords, and keywords extracted by the large language model corresponding to CLC categories; Step 2.3 The Chinese Library Classification Concept Set Cc includes the superordinate concepts, ontology, knowledge graph, and terminology definitions of the Chinese Library Classification Number category; Step 2.4 The Chinese Library Classification Segment Vector Set Cp uses a semantic vector model to vectorize the segments in the sample set, fuses features, and calculates the feature mean vector of each category of document sample set. Step 2.5 generates a composite semantic vector for CLC category b, Cb=α*Normalize(Cw_avg)+β*Normalize(Cc_cen)+γ*Normalize(Cp_cen), with the weight parameters α,β,γ fused, where α,β,γ≥0 and α+β+γ=1.

11. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 10, characterized in that: Step 2.1: X is a dynamic value, X = min(1000, total number of documents under this CLC category * 0.7), that is, extract 1000 documents or 70% of the documents under this category, whichever is less, and deduplicate based on the document DOI.

12. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 10, characterized in that: Step 2.2: CLC subject terms are taken from category names, annotations, and the "Chinese Classification Subject Headings"; machine-generated keywords are automatically extracted from sample documents using the TF-IDF algorithm, retaining the top-K words with the highest weights; author indexing keywords are directly taken from the author keywords in the document metadata; expert indexing keywords are extracted from the standardized keywords of high-quality journals; the large language model extracts keywords using a finely tuned large language model with the prompt: Please extract the keywords that best represent the core research content from the following scientific and technological documents [title / abstract], requiring both professionalism and universality.

13. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 12, characterized in that: After converting simplified and traditional Chinese characters and unifying full-width and half-width characters for the vocabulary from the above sources, and standardizing the terminology, deduplication is performed, retaining m words. The model is used for vectorization processing, and each word w_i generates a vector v_w_i. The vectorized representation of each category word set is Cw={v_w_1,v_w_2,...,v_w_m}. The comprehensive word vector of CLC category b is Cw_avg=(1 / m)*Σv_w_1.

14. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 10, characterized in that: Step 2.3: The higher-level concepts are generated through the CLC tree structure; the ontology and knowledge graph map the CLC categories to the relevant concept entities from the external knowledge base through entity linking technology; the terminology definitions are obtained from the category technical definitions generated by the user manual, technical dictionary, or large language model.

15. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 14, characterized in that: For each concept c_j, all its textual description information is concatenated into a comprehensive text Text(c_j). The domain-adapted semantic vector model is used to encode Text(c_j) to obtain the concept vector v_concept_j. The CLC classification concept set Cc is represented as the set of all concept vectors: Cc = {v_concept_1, v_concept_2, ..., v_concept_n}. The category concept center vector Cc_cen = (1 / n)*Σv_concept_j.

16. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 10, characterized in that: Step 2.4: A unified semantic model with domain fine-tuning is adopted to ensure the consistency of the vector space; for a specific CLC category b, there are X document samples under it. Calculate the segment vector set Cp_b of this category, that is, the set of all segment vectors. The category center vector of category b is: Cp_cen=(1 / X)*Σv_doc_m, where v_doc_m is the document vector of the m-th patent under category b.

17. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 1, characterized in that: Step 3 specifically involves: calculating the composite semantic vector for each classification number, sequentially calculating the similarity between Ia of each IPC category and Cb of each CLC category, Sim(I,C)=CosineSimilarity(Ia,Cb), and determining that IPC classification number a and CLC classification number b are similar categories when the similarity threshold is greater than λ. The Top-N CLC categories are used as the final mapping result, and this process generates a mapping from CLC to IPC.

18. The method for converting the International Patent Classification (IPC) to the Chinese Library Classification (CLC) according to claim 17, characterized in that: Step 4: Optimize and adjust the parameters α, β, γ, and λ.