Method and system for automatically evaluating answer quality of tobacco question-answering system

The invention discloses a method and a system for automatically evaluating the answer quality of a tobacco question-answering system. The method comprises the following steps: acquiring a question-answering pair to be evaluated; performing semantic coherence detection on questions and answers in the question and answer pairs by adopting the trained large language model; the method comprises the steps of performing question type classification on question pairs passing semantic coherence detection, extracting entities and attributes in questions and answers according to the obtained question types, and querying corresponding information in a pre-constructed standard database according to the entities and the attributes to perform reverse coverage detection so as to judge whether the answers are correct or not. The answers are evaluated jointly in two aspects of sentence-level semantic coherence and word-level information accuracy, and the automatic evaluation efficiency and the system question and answer evalu