The application provides an intelligent question and answer
system and method based on hierarchical vector retrieval and literature tracing, which splits the literature by taking the
paragraph as the minimum unit through the
data processing layer, binds the literature meta information, enables the
retrieval result to be accurately positioned to the
specific knowledge point instead of the whole document, significantly improves the
granularity and verifiability of the answer, the hierarchical vector index structure of the clustering center index and the clustering internal index is constructed through the semantic index layer, the retrieval efficiency under the large-scale
knowledge base is improved, the similarity retrieval is performed in the local cluster through the semantic retrieval module, the candidate paragraphs are filtered and sorted, the low-quality and irrelevant content is effectively eliminated, and the context purity of the input large
language model is improved, the reference
paragraph and the literature meta information are constructed as a structured context through the question and answer generation module, the final answer generated naturally carries the literature source information, the
strong binding of the answer content and the original evidence is realized, and the requirements of the professional field for the result credibility and
traceability are met.