Reservoir standard intelligent question and answer intent recognition and clause positioning navigation method and system

Abstract

A reservoir standard intelligent question and answer intention recognition and clause positioning navigation method and system solve the problems of inaccurate intention understanding, fuzzy clause positioning and poor engineering application continuity in the current reservoir field. The method comprises the following steps: constructing a reservoir standard knowledge base and a professional intention dictionary, the intention dictionary containing four categories of core intentions and three-dimensional subdivision labels of "engineering type + technical theme + applicable stage"; performing clause-level fragmentation indexing and vectorization processing on the reservoir standard text, establishing a four-level index of "intention label-standard chapter-clause number-core semantics", and fusing the inverted index and the knowledge graph; performing intention recognition based on the fine-tuning BERT model combined with the intention dictionary, and extracting key entities; remembering the context through dialogue state tracking technology to realize multi-round intention association; matching the intention with the index, screening the applicable clauses through multi-dimensional sorting, and generating a jump link; and outputting a one-stop result of integrated answers, applicable scenarios and links.

Images