Keyword-Bert-based Academic Intelligent Question Answering System

Abstract:
Since the pre-training model has matured, natural language processing has ushered in unprecedented development, and deep semantic matching has also achieved better results. Traditional intelligent question answering cannot accurately find the most similar among several questions. Therefore, an accurate answer cannot be returned. Therefore, we propose a question-and-answer system based on information retrieval-based deep semantic matching and deep learning ranking with contextual understanding, and apply it to the academic evaluation platform for middle school students, which can quickly and accurately return the best answer. First, construct the question-and-answer pairs for academic assessment and store them in the Mysql database. Then, entity recognition is carried out based on rules and dictionary, and secondly, BM25 and Word2vec are used to recall question sentences. Finally, use Keyword-Bert to sort the recalled questions again, and then select the best quality answer. For questions that cannot be answered, we have built an intelligent knowledge base, which can efficiently answer questions and distribute annotations, and complete the life cycle management and optimization of question and answer knowledge.