Most studies on why-question answering system usually  used  the  keyword-based  approaches.  They  rarely involved domain ontology in capturing the semantic of the document contents, especially in detecting the presence of the causal relations. Consequently, the word mismatch problem usually  occurs  and the  system often retrieves  not  relevant answers. For solving this problem, we propose an answer extraction method by involving the semantic similarity measure, with selective causality detection. The selective causality detection is  applied because not  all  sentences  belonging  to  an  answer contain  causality.  Moreover,  the  motivation  of  the  use  of semantic similarity measure in scoring function is to get more moderate results about the presence of the semantic annotations in a sentence, instead of 0/1. The semantic similarity measure employed is based on the shortest path and the maximum depth of the ontology graph. The evaluation is conducted by comparing the proposed method against the comparable ontology-based methods, i.e., the sentence extraction with Monge-Elkan with 0/1 internal similarity function. The proposed method shows the improvements in  term of  MRR (16%, 0.79-0.68), P@1  (15%, 0.76-0.66), P@5 (14%, 0.8-0.7), and Recall (19%, 0.86-0.72).
Copyrights © 2017