Automatic Answering Method Considering Word Order for Slot Filling Questions of University Entrance Examinations

Ryo Tagami, Tasuku Kimura, Hisashi Miyamori

Abstract: Recently, automatic answering technologies such as question answering have attracted attention as a technology to satisfy various information requests from users. In this paper, we propose an automatic answering method considering word order for the slot filling questions in the university entrance examination world history problems. In particular, when in analyzing the question sentence, the answer category is estimated from the surrounding words of the filling slot and used for extracting the answer candidates. Also, these candidates are evaluated by introducing the indicator using the consistency with the category and the occurrence situation of the surrounding words. In the experiment, we first compare the accuracy of the word prediction models. Then, we compare the proposed method with the baseline method and clarify what kind of change is observed in the correct answer rate.