[4Q2-GS-9] Natural language processing, information retrieval: Q&A system

[4Q2-GS-9-05] Collecting Chinese Know-How Question Answering Examples and Application of Neural Machine Comprehension Model

〇Tengyang Chen¹, Zechang Qian², Hongyu Li¹, Takehito Utsuro¹, Yasuhide Kawada³ (1. University of Tsukuba, 2. Tokyo University of Technology, 3. Logworks Co., Ltd.)

Keywords:question answering, machine comprehension, tip, BERT, Chinese

In the recent research, the state-of-the-art technique has achieved much gain in the field of question answering,
and the result is comparable to the human-being’s. On the contrary, there are limited research focusing on the field
of non-factoid question answering. This thesis studies how to develop a dataset for training Chinese tip question
answering models. In this thesis, then the Chinese tip questions and answers dataset is used to train one of
the state-of-the-art machine comprehension models. This thesis also compares the trained tip question answering
model to a Chinese factoid question answering model which is trained with a Chinese factoid questions and answers
dataset. Evaluation result shows that the model trained with the mixture of tip and factoid question and answers
datasets achieve the best performance.