Question answering system based on knowledge graph
Zhang, Zherui (2025)
Kandidaatintyö
2025
School of Engineering Science, Tietotekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2025052353388
https://urn.fi/URN:NBN:fi-fe2025052353388
Tiivistelmä
In this paper, we propose and implement a document Q&A system that fuses knowledge graph and retrieval augmented generation (RAG) mechanism.The system takes full advantage of the unique advantage of knowledge graph in structured semantic expression, and significantly improves the accuracy and interpretation ability of large language models in vertical domain Q&A tasks.The system architecture takes LlamaIndex as the core to build the indexing system, and realises inference support and semantic mapping with the language and embedding models provided by Knowledge Graph AI.
At the implementation level, this paper systematically develops an in-depth analysis of the overall design concept, module coordination mechanism, key technology paths and system validation process.The experimental results clearly show that the system has the ability to efficiently identify entity nodes and semantic relationships in documents, and can construct a knowledge graph with an inherent logical structure, on the basis of which it can realise targeted and contextually appropriate Q&A interactions.
It is worth emphasising that, compared with the traditional retrieval method based on semantic vector matching, this system shows better reasoning ability and response accuracy when coping with tasks such as multi-hop reasoning and complex relationship chain parsing.The results not only validate the feasibility of the Knowledge Graphenhanced RAG framework, but also explore a potential technology path for intelligent Q&A of documents in professional domains.
At the implementation level, this paper systematically develops an in-depth analysis of the overall design concept, module coordination mechanism, key technology paths and system validation process.The experimental results clearly show that the system has the ability to efficiently identify entity nodes and semantic relationships in documents, and can construct a knowledge graph with an inherent logical structure, on the basis of which it can realise targeted and contextually appropriate Q&A interactions.
It is worth emphasising that, compared with the traditional retrieval method based on semantic vector matching, this system shows better reasoning ability and response accuracy when coping with tasks such as multi-hop reasoning and complex relationship chain parsing.The results not only validate the feasibility of the Knowledge Graphenhanced RAG framework, but also explore a potential technology path for intelligent Q&A of documents in professional domains.