Language Model-Based Algorithm for Constructing Knowledge Graphs from Patent Data

Gavrilov, Nikita

doi:10.1007/978-3-031-97648-3_15

Language Model-Based Algorithm for Constructing Knowledge Graphs from Patent Data

2026-01-01

0

PRE

AI

DOI：10.1007/978-3-031-97648-3_15

Original

Original request for help

Save

Abstract

En 中文

This paper presents a method for constructing a knowledge graph based on patent data, which facilitates the identification of hidden relationships between patents and the organization of information for subsequent analysis. The method involves extracting key textual fields from patent documents and vectorizing them using state-of-the-art transformer models, and building a graph where the nodes represent individual documents, and the edges reflect their semantic proximity. A clustering algorithm is employed to group the patents, ensuring high internal coherence within clusters and reducing the original graph to a compact representation. The resulting clusters are summarized using language models, enabling automatic extraction of significant terms for cluster descriptions. Experimental research conducted on a large corpus of patent data demonstrates the efficacy of the proposed approach, which is confirmed by the relevant partitioning quality metrics. The proposed method improves the interpretation of patent information, facilitating the identification of implicit relationships and structural patterns, which is of great importance for analyzing scientific achievements and managing intellectual property.

Keywords:

Knowledge Graph

Patent Data

Text Vectorization

Clustering

Language Model-Based Algorithm for Constructing Knowledge Graphs from Patent Data

Abstract

Journal

Organization

Cited Papers