An Auto Question Answering System for Tree Hole Rescue

Abstract

This paper introduces an automatic question answering system which aimed to provide online how-to instructions for volunteers of Tree Hole Rescue–a Chinese online suicide rescue organization. When a volunteer needs to make sure how to deal with a rescue task professionally, he/she could ask this system via its WeChat public account other than reading a rescue instruction menu book. Firstly, a Tree Hole Rescue question-answer knowledge graph was constructed to manage Tree Hole Rescue question-answer knowledge and its relationship. Then, based on this semantic technology, a question in Chinese natural language was parsed into a machine-readable logical language through the question preprocessing and entity mapping process. And, a candidate question set was generated through a hierarchical information retrieval strategy. Finally, an exact or close answer and recommended similar questions were sent to the asker after calculating words sequence similarity via an algorithm which combined word form and semantic features. If the system could not match an answer for a question, the question would be added to unsolved question list and the system would alert administrator to deal with it. System testing shew that the Q&A system has a high accuracy rate in response of Tree Hole Rescue questions. Meanwhile, this system provides a series of methods to improve the update capability of the Q&A library and the scalability of the system.