Topic oriented semantic parsing

Abstract

Semantic parsing is still a challenging problem for open domain question answering. In semantic parsing, questions are mapped with their meaning representations. These representations are matched with feasible answers in knowledge bases. In Knowledge bases (e.g. Freebase), knowledge is stored in the form of Topics. For a successful answer extraction from Freebase, it is required to correctly identify the Topic node (or Topic word) of the question and retrieve every type and property associated with this Topic node. In this paper, a Topic Node Identification (TNI) algorithm is proposed for correctly identifying question Topic and Domain Word Identification (DWI) algorithm is proposed for correctly identifying domain of the Topic node. After domain identification the Topic node is further expanded for its all types and properties. Out of all types identified, one of the type and associated property is likely to be an answer of the question. TWI and DWI algorithms use techniques i.e. proposed rulebased and machine learning approach with the help of question dependency parser. Results of proposed approach outperform state of art approaches.