FAQ question Answering method based on semantic similarity matching

Frequently asked questions (FAQ) is a typical limited domain-oriented retrieval question answering system, and semantic similarity matching is the key technology of FAQ question answering. Aiming at the FAQ problems of insufficient deep semantic representation and extraction and synonyms and polysemy, this paper proposes a cascade matching model combining coarse-grained recall and fine-grained sorting. Firstly, an improved Okapi BM25 algorithm based on inverse word frequency is proposed. This paper proposes a new solution to FAQ answer based on the traditional text matching and retrieval pipeline, and gives a new semantic similarity matching model. Additionally, we further analyze the shortcomings of semantic matching and the advantages of data-driven semantic similarity matching model. Finally, the proposed hybrid matching model is evaluated. Experimental results show that the proposed method not only ensures the retrieval speed of question and answer, but improves its performance