Arbaaeen, Ammar Fuad OAmmar Fuad OArbaaeen2024-10-082024-10-082022https://studentrepo.iium.edu.my/handle/123456789/9346The tremendous growth in the field of data science and the widespread usage of information retrieval techniques has enabled users to retrieve accurate information. The diverse data availability in various Knowledge Base (KB) formats introduces several challenges to deliver concise and precise information corresponding to human queries. This requires a user to be familiar with the structures of KB and use a formal query language for the system to effectively understand the query. Question Answering (QA) systems have been introduced to enable users to post questions in Natural Language (NL) and infer specific answers instead of lists of documents. Such a system requires the capability of both critical analysis on questions and inference on answers selection. NL question analysis module is a fundamental step that impacts the QA system performance. It aims to transform users’ NL questions into representations of a structured format suitable to query across KBs. Literature showed, the major challenge in NL question transformation is language ambiguity that may occur at a lexical-semantic level. Moreover, various challenging questions require handling ambiguities based on a certain condition such as seeking instructions or advice. Therefore, the motivation of this study is to propose a Knowledge-based Sense Disambiguation (KSD) method for resolving the problem of lexical ambiguity associated with NL questions. This algorithm is designed by incorporating question’s metadata (date/GPS), context knowledge, and domain ontology, into a shallow NL processor. It aims at enhancing the accuracy of the word sense disambiguation process in questions analysis module to effectively returns potential answers corresponding to questions posed in QA systems. This work explores the use of the proposed KSD method to support pilgrims in expressive queries to obtain accurate information via a mobile QA application. Therefore, the validity of the proposed solution has been supported by two experiments to evaluate the accuracy performance. First, in vitro experiment was carried out as a standalone task to evaluate the KSD as a word sense disambiguation method in comparison with the baselines WordNet Most Frequent Sense (MFS) and the simplified version of the Lesk method on the same condition and test dataset. Second, in vivo experiment was performed to evaluate the effectiveness of the KSD method in a QA application in comparison to the MFS in the context of a pilgrimage domain. The results obtained from both experiments have revealed the feasibility of the proposed solution to effectively cope with lexical ambiguity in NL questions as well as to contribute to QA system performance improvement. “We don’t need more information. We need more meaning.” Paul SalopekenNatural language processing (Computer science)Semantics -- Data processingDoctoral Thesis