WabiQA: A Wikipedia-Based Thai Question-Answering System
作者：
Highlights：
• We propose WabiQA, a novel Thai QA system that implements a BM25F-based document retriever and a bi-directional LSTM document reader.
• We compare different document retrieval methods, including Google Search API, TF-IDF, and BM25F, and also study the impacts of different features in the bi-directional LSTM reader.
• WabiQA outperforms other Thai QA systems by 31.48% in terms of document retrieval accuracy, and 198.83% in terms of answer prediction F1.
• WabiQA is developed as part of a prototype mobile application aiming to facilitate users with visual impairment.
摘要
•We propose WabiQA, a novel Thai QA system that implements a BM25F-based document retriever and a bi-directional LSTM document reader.•We compare different document retrieval methods, including Google Search API, TF-IDF, and BM25F, and also study the impacts of different features in the bi-directional LSTM reader.•WabiQA outperforms other Thai QA systems by 31.48% in terms of document retrieval accuracy, and 198.83% in terms of answer prediction F1.•WabiQA is developed as part of a prototype mobile application aiming to facilitate users with visual impairment.
论文关键词：Question Answering System,Deep Learning,Creative Language Processing
论文评审过程：Received 7 July 2020, Revised 5 October 2020, Accepted 31 October 2020, Available online 25 November 2020, Version of Record 25 November 2020.
论文官网地址：https://doi.org/10.1016/j.ipm.2020.102431
