基于知识图谱、TF-IDF和BERT模型的冬奥知识问答系统

传统信息检索技术已经不能满足人们对信息获取效率的要求，智能问答系统应运而生，并成为自然语言处理领域一个非常重要的研究热点。本文针对中文的冬奥问答领域，提出了基于知识图谱、词频-逆文本频率指数 (term frequency-inverse document frequency,TF-IDF)和自注意力机制的双向编码表示(bidirectional encoder representation from transformers,BERT)的3种冬奥问答系统模型。本文首次构建了冬奥问答数据集，并将上述3种方法集成在一起，应用于冬奥问答领域，用户可以使用本系统来快速准确地获取冬奥内容相关的问答知识。进一步，对3种模型的效果进行了测评，测量了3种模型各自的回答可接受率。实验结果显示BERT模型的整体效果略优于知识图谱和TDIDF模型，BERT模型对3类问题的回答可接受率都超过了96%，知识图谱和TDIDF模型对于复合统计问答对的回答效果不如BERT模型。

Winter Olympic Q & A system based on knowledge map,TF-IDF and BERT model

With the advent of the information age, traditional information retrieval technology can no longer meet people’s requirements for the efficiency in information acquisition, so intelligent question answering systems are proposed and have become a very important research hotspot in natural language processing. This paper proposes three Winter Olympics Q&A system models based on knowledge graph, TFIDF and BERT for the Chinese Winter Olympics Q&A, constructing the Winter Olympics Q&A data set for the first time and integrating the above three methods into the Winter Olympics Q&A. Users can use this system to quickly and accurately obtain the Q&A knowledge related to the Winter Olympics content. Furthermore, this paper evaluates the effects of the three models and measures the acceptance rate of each model. The experimental results show that overall the BERT model is slightly better than the knowledge graph and TDIDF model. The acceptance rate of the BERT model for each of the three types of questions exceeds 96%. The knowledge graph and TDIDF model are not so effective as the BERT model for the answer to the composite statistical question and answer pair.