Graph-Based Document Retrieval for Conceptual Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional document retrieval methods struggle to accurately capture the conceptual similarities between documents, particularly in fields like patents and contracts, where similar words are heavily used, leading to a need for a retrieval technology that considers document concepts beyond just word usage.
Innovation Solution
A document retrieval system that creates graph structures from documents, incorporating morphological and modification analyses, part-of-speech tagging, and vectorization using Weisfeiler-Lehman Kernels to evaluate conceptual similarities between documents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If word-based retrieval methods (PageRank, thesaurus, Jaccard coefficient, cosine similarity) are used, then retrieval can be performed using basic word matching, but precision deteriorates because conceptual similarities are not captured
Solution Approach 1:
The patent segments the document into sentences, then segments each sentence into clauses, and finally segments clauses into graph nodes representing concepts. This multi-level segmentation allows the system to capture conceptual relationships while maintaining manageable processing complexity through hierarchical decomposition.
Solution Approach 2:
The patent introduces a graph structure as an intermediary representation between raw text and retrieval queries. The graph nodes and edges serve as mediators that encode conceptual relationships, allowing the system to bridge the gap between word-based processing and concept-based retrieval without requiring direct complex analysis of entire documents.
2Measurement precision
If graph structure creation with morphological and modification analysis is performed, then conceptual similarity capture is improved, but processing time increases
Solution Approach 1:
The patent performs morphological analysis and modification analysis as preliminary actions during the graph structure creation phase. By completing these analyses upfront when building the graph representation, the system avoids repeated processing during retrieval operations, thereby reducing query time while maintaining high conceptual similarity measurement accuracy.
3Measurement precision
If vectorization using Weisfeiler-Lehman Kernels is applied, then conceptual similarity evaluation is enhanced, but computational complexity increases
Solution Approach 1:
The patent applies parameter changes by using Weisfeiler-Lehman Kernels to transform graph structures into vector representations. This parameter transformation enables efficient similarity evaluation through vector operations while capturing complex conceptual relationships, balancing computational complexity with improved similarity evaluation accuracy.
Data Source
AI summary
A document retrieval system that retrieves documents, with concepts of the documents taken into account, is provided. The document retrieval system (100) includes an input unit (101), a first processing unit (102), a storage unit (105), a second processing unit (103), and an output unit (104). The input unit (101) has a function of inputting a first document (20), the first processing unit (102) has a function of creating a first graph structure (21) from the first document (20), the storage unit (105) has a function of storing a second graph structure (11), the second processing unit (103) has a function of calculating a similarity between the first graph structure (21) and the second graph structure (11), the output unit (104) has a function of supplying information, the first processing unit (102) has a function of dividing the first document (20) into a plurality of tokens, a node and an edge of the first graph structure (21) have a label, and the label includes the plurality of tokens.