Query disambiguation approach using triple-filter

In order to effectively deal with structured information, some technical skills are needed. However, most users do not possess these skills. Natural Language Interfaces (NLIs) are therefore built to provide everyday users who lack the needed technical knowledge, with some means of gaining access to the information stored in knowledge bases. They are designed to deal with the natural language articulation of what the user wants, and then transform it into a computer language that specifies how to accomplish it. Each NLI system also has a scope and limitation which everyday users are unaware of. It is therefore understandable that the users may not see errors in their queries or even know how to write appropriate queries (according to the system’s limitation and scope) in order to retrieve the correct information. The effective retrieval of any piece of information depends wholly on the correct mapping of queries made in natural language to machine understandable form. However, most of the existing NLIs are lacking in terms of being able to provide support for users to formulate their queries. Once queries are wrongly formulated, there is tendency for the system to retrieve wrong answers. As a result, a user may have to reformulate his query severally before the required answer is retrieved (if at all). In this thesis therefore, it is proposed that the best way to formulate appropriate queries is by guiding the user through the query writing process and helping the user to resolve ambiguities by providing suggestions that are easy to understand. The proposed approach, referred to as triple-filter query disambiguation approach, has been implemented into a prototype as a proof of concept. The prototype, referred to as QuFA (Query Formulation Assistant), is intended to serve as an upper layer for NLIs. It is equipped with an authoring service that guides the user to write his query, a disambiguation module that resolves ambiguities in order to ascertain the user’s intention and finally, a query rewriting module that transforms the user’s input query into an intermediate query that will suit the underlying search system’s perspective of the user’s question. Extensive experimental evaluations were conducted in order to validate the proposed approach, using the developed prototype. The proposed triple-filter query disambiguation approach was directly compared with the approach in FREyA (Feedback, Refinement and Extended vocabulary Aggregation) that also provides support to users when formulating queries. The evaluation was based on the usability and performance of the approaches. In terms of usability, the results show that the proposed approach has the potential of being more acceptable in the field; and in terms of effectiveness, it also shows a high performance based on precision and recall. The proposed approach helps users to conceive and articulate more effective queries, and facilitates information search activities. The main contributions of this research work include the introduction of an approach that enables users without knowledge of formal computer languages to formulate useful queries while effectively expressing themselves using natural language. The approach utilizes the effectiveness of human-computer dialogue to effectively retrieve desired information from ontologies. The proposed triple-filter disambiguation approach inculcates a learning mechanism that continues to automatically learn from user queries and continuously improves its performance capability. The triple-filter disambiguation algorithm was also developed, along with two documents (terms equivalence catalogue (TEC) and the enhanced concepts store (ECS)) that represent the thesaurus and the lexicon for use with the Mooney Geoquery dataset. All of these are available for use by other researchers.

Preview Text FSKTM 2018 11 - IR.pdf Download (2MB) | Preview

Actions (login required)

View Item