Discovering query patterns with features for query recommendations

Submission note: A thesis submitted in total fulfilment of the requirements for the degree of Master of Science by Research to the Department of Computer Science and Computer Engineering, La Trobe University, Bundoora.

With rapid development of Internet technology, the recommender system has become the most important filtering tool to solve information overload problems. Recommender system could help users discover useful information from massive data quickly and effectively. Although the research on query recommendations has been conducted extensively, most existing solutions are working well only on specific applications with various assumptions and constraints. Especially, those advanced applications such as query pattern discovery for recommendations in OLAP systems have not been deeply and broadly investigated yet. More accurate and effective recommendation systems are required further for their full potential. An advanced query recommendation system should include novel methods for describing users’ behaviours, integrating various information, and providing an efficient and effective process modelling for accurate recommendations. This thesis has endeavoured to make the advanced recommendation system satisfied. Two approaches to improve OLAP systems for providing the aforementioned features are proposed. First, a novel framework called feature-based recommendations in OLAP system is proposed. It aims at analysing the query logs for mining users’ preferences, interests and behaviours. Experiments have been done to show the superiority of the proposed framework. Second, a novel query-flow graph for mining and recommending useful information from query logs is designed. The approach is based on concepts of mutual information and confidence between queries. The research work has introduced a novel ranking function using skyline dominance semantics to evaluate relevant queries. Extensive experiments have been conducted. The results have demonstrated that the proposed architecture is efficient for generating effective query recommendations.