Year: 2025
Type: dissertation
Abstract: The rapid development of augmented retrieval generation (RAG) methods has significantly improved the quality of text generation in knowledge-intensive tasks. However, traditional RAG systems often face challenges related to retrieval accuracy, semantic complexity, and multi-step reasoning. This paper proposes an approach to improve RAG by integrating knowledge graphs (KGs), which provide a structured representation of entities and their relationships. The proposed method leverages KGs to improve the quality of information retrieval and decompose complex queries into smaller, more manageable subqueries. The approach involves constructing a knowledge graph from domain-specific data, performing graph-based search to identify relevant entities, and incorporating the extracted information into the generation process using graph-based hints. Experiments are conducted on publicly available datasets to evaluate the effectiveness of the approach in query-oriented abstract summarization (QFAS) tasks. The results show that the proposed method improves retrieval accuracy, answer consistency, and multi-step reasoning capabilities compared to baseline RAG systems. This work highlights the potential of knowledge graphs to improve the interpretability and robustness of RAG systems, paving the way for future research on combining knowledge graphs and large language models in natural language generation tasks.
Authors: Измалкин, А. А., Izmalkin, A. A.
Language: Russian
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7115037696