A Model of Vietnamese Person Named Entity Question Answering System

Copyright 2012 by Mai-Vu Tran, Duc-Trong Le, Xuan- Tu Tran, and Tien-Tung Nguyen 26th Pacific Asia Conference on Language,Information and Computation pages 325–332 A Model of Vietnamese Person Named Entity Question Answering System Mai-Vu Tran, Duc-Trong Le KTLab, University of Engineering and Technology – Vietnam National University, Hanoi {vutm, trongld}@vnu.edu.vn Xuan- Tu Tran, Tien-Tung Nguyen KTLab - University of Engineering and Technology – Vietnam National University, Hanoi {tutx_52, tungnt_5}@vnu.edu.vn Abstract In this paper, we proposed a Vietnamese named entity question answering (QA) model. This model applies an analytical question method using CRF machine learning algorithm combined with two automatic answering strategies: indexed sentences database-based and Google search engine-based. We gathered a Vietnamese question dataset containing about 2000 popular “Who, Whom, Whose” questions to evaluate our question chunking method and QA model. According to experiments, question chunking phase acquired the average F1 score of 92.99%. Equally significant, in our QA evaluation, experimental results illustrated that our approaches were completely reasonable and realistic with 74.63% precision and 87.9% ability to give the answers. Keywords: Vietnamese question, QA, VPQA, question analysis, answer extraction, question parser 1 Introduction Numerous researches about Question Answering (QA) systems have been discussed in recent years. Initially, they only answered simple questions; however, currently researches have been focused on methods for more complex questions. Those methods analyze and parse complex questions to various simple questions before using existed techniques to respond. [1] Automatic question answering – the ability of computers to answer simple or complex questions, posed in ordinary human language – is the most exciting. Building the question answering system is a difficult issue in terms of natural language processing tasks. Presently, automatic question answering systems are revolutionizing the processing of textual information. By coordinating complex natural language processing techniques, sophisticated linguistic representations and advanced machine learning methods, automatic question answering systems can detect exact responses from a wide variety of natural language questions in unstructured texts. Recent researches demonstrated that the increasing in performance of systems is dependent on the number of probable answers in documents. The exact answer detection is one of the most significant problems in QA systems. For this purpose, our model utilized CRF [5] machine learning algorithm to parse natural questions and some IR strategies to extract answers. The model works on closed domain by extracting human names based on knowledge warehouse and search engines. If answers are not found in database, the question will push into Google search engine. The QA system just supports questions (such as “Who?”, “Whom?”, “Whose?”) in factoid form or one sentence. The aim of this paper is to design and implement a new classification model, reformulation and answer validation in a QA system. The methodology in our system is to discover correct answer in person domain with NLP techniques, CRF model to parse question, and some strategies to extract answer: knowledge- based, search engine-based and hybrid method. The primary reason of an answer validation component in the system concerns the difficulty of picking up from a document …

Content truncated to 500 words