HYBRID MODEL FOR THE CLASSIFICATION OF QUESTIONS EXPRESSED IN NATURAL LANGUAGE

Abstract

Question-answering systems rely on an unstructured text corpora or a knowledge base to answer user questions. Most of these systems store knowledge in multiple repositories including RDF. To access this type of repository, SPARQL is the most convenient formal language. It is a complex language, it is therefore necessary to transform the questions expressed in natural language by users into a SPARQL query. As this language is complex, several approaches have been proposed to transform the questions expressed in natural language by users into a SPARQL query.However, the identification of the question type is a serious problem. Questions classification plays a potential role at this level. Machine learning algorithms including neural networks are used for this classification. With the increase in the volume of data, neural networks better perform than those obtained by machine learning algorithms, in general. That is, neural networks, machine learning algorithms also remain good classifiers. For more efficiency, a combination of convolutional neural network with these algorithms has been suggested in this paper. The BICNN-SVM combination has obtained good score not only with small dataset with a precision of 96.60% but also with a large dataset with 94.05%.