Towards an Approach Based on Passage Enrichment and the Decision Tree Model in an Arabic Question Answer System

Abstract:

The goal of a question answer system is to extract and present answers to questions asked in natural language. These answers are the candidate sentences available after the question analysis, document retreival and / or passage. Extracting a specific answer serves to examine the list of selected passages and select the most appropriate sentence to the question. For this reason, the selection of answers in search of precise information necessarily implements a matching of passages with the question. We propose a method based on two contributions: the first is used to rapproach passages of text to pairs (questions, candidate answers) in an Arabic question answer system. These are selected by lexical expansion using Arabic WordNet and word vectors. After, a support passage is extracted from the selected passage set. It is enriched by means of an external Arabic resource including Ontology that it combines lexical coverage and semantic relationships between words existing in Arabic WordNet and verbs in Arabic VerbNet.The second contribution presents the clustering of answers which is realized by using the decision tree model. The objective of the model is to validate the correct answers, and to invalidate the false answers which it must be able to assign the correct class to each answer choice. In our work, we are interested in dealing with a different type of question compared to the majority of search in Arabic.