Relation Detection of Persian Questions by Combining Direct and Indirect Methods
Subject Areas : electrical and computer engineeringAbbas Shahini Shamsabadi 1 , Reza Ramezani 2 , Hadi Khosravi farsani 3 , Mohammadali nematbakhsh 4
1 - Department of Software Engineering, Faculty of Computer Engineering, University of Isfahan, Iran
2 - Department of Software Engineering, Faculty of Computer Engineering, University of Isfahan, Iran
3 - Department of Computer Engineering, Shahrekord University, Shahrekord, Iran
4 - Department of Software Engineering, Faculty of Computer Engineering, University of Isfahan, Iran
Keywords: Persian question-answering, relation detection, knowledge base, natural language processing.,
Abstract :
In this study, for the problem of answering Persian questions using linked data, the sub-problem of relation detection for single-relation questions has been investigated in detail. In these questions, the answer is extracted from a triple in the form of <subject, predicate, object>. This process has two main steps: entity linking and relation detection. In the first step, the entity identified in the question is mapped to a subject or object of a triple, and in the second step, a predicate is selected for the semantic relation in the question. In most existing methods, after entity linking, all relations of that entity in the knowledge base are considered as candidate relations, and finally one of them is selected as the final relation. In these methods, if there is an error in the entity linking step, it is propagated to the relation detection step. In this study, to solve this dependency, the hierarchical structure of relations is used in order to directly extract the relation of the question. The accuracy of the proposed method in Persian is 72% for direct relation detection and 90% for selecting the best candidate relation (indirect). The accuracy has increased to 94% by combining direct and indirect methods.
