Tamil Question Answering System Using Machine Learning

Introduction

Internet provides high source of data but English is the dominant language. Though the people get information from search engine it may or may not be useful and relevant. They spend much of time to get the relevant answer. The key idea of this work is to find and represent the exact answer to the user’s query. The difficulty of finding and validating the right answer makes a QAS complex than the general information retrieval task.

The system processes Tamil questions and provides answers in Tamil to the user in the agriculture domain. Uzhavan mobile application for agriculture domain in Tamil provides information regarding agriculture schemes and subsidy. But it does not give concise answers to the farmer queries. To address such problems Tamil QAS system gives exact answers to user queries.

Table 1 shows the different type of questions in Tamil. In this work, the user queries in the agriculture domain is processed by the proposed model to get answer in Tamil. The key idea of this system is to find agriculture domain based question answering system in Tamil language. Stop words in Tamil are collected manually and the Tamil words are POS tagged. Locality sensitive hashing is used to retrieve relevant documents. Finally the answers are ranked using Jaccard similarity.

Table 2 shows the different types of QASs. QAS are classified based on application domain, analysis done and techniques used. Based on application domain it is classified as open and closed domain. Based on analysis QAS is further classified as semantic QAS, syntactic QAS and morphological QAS. Based on technique used it is classified as web based and ontology based.