SAGE - SMART ANSWERS GENERATOR OF EVERYTHING

Abstract

In recent years, the problem of Open-Domain Question Answering has been widely discussed and rapidly developed. However, most of the Question Answering systems are only available in English. In this thesis, we present a system which focuses on processing Vietnamese natural language, exclusively for Vietnamese users. Our system, Smart Answers Generator of Everything (SAGE), is inspired by the Open Question Answering System – or the OQA System (Anthony Fader, Luke Zettlemoyer, Oren Etzioni, 2014). Similar to the Open Question Answering system, SAGE approaches the problem of Open-Domain Question Answering with a divide-and-conquer strategy. However, asides from traditional difficulties faced by other Question Answering systems, a Vietnamese Question Answering System has to overcome a challenge that only applies to a frontier system: the limited availability of Vietnamese knowledge bases and other data corpuses, which also leads to the impossibility of creating reliable operators sets for subproblems which would normally be mined from the knowledge base. Thus, SAGE has to present a unique tactic to solve each of these sub-problems. To create SAGE, we have studied about state-of-the-art Question Answering Systems like PARALLEX, OQA, as well as learning the basic steps of building a Question Answering system. We proposed SynGraph, a method to improve recall when processing a query. A clean, well-prepared Knowledge Base extracted from DBPedia was created along the way to ensure the operation of SAGE.