Generation and Ranking of Candidate Networks of Relations for Keyword Search over Relational Databases

Abstract

Several systems proposed for processing keyword queries over relational databases rely on the generation and evaluation of Candidate Networks (CNs), i.e., networks of joined database relations that, when processed as SQL queries, provide a relevant answer to the input keyword query. Although the evaluation of CNs has been extensively addressed in the literature, problems related to efficiently generating meaningful CNs have received much less attention. To generate useful CNs is necessary to automatically locating, given a handful of keywords, relations in the database that may contain relevant pieces of information, and determining suitable ways of joining these relations to satisfy the implicit information need expressed by a user when formulating her query. In this thesis, we present two main contributions related to the processing of Candidate Networks. As our first contribution, we present a novel approach for generating CNs, in which possible matchings of the query in database are efficiently enumerated at first. These query matches are then used to guide the CN generation process, avoiding the exhaustive search procedure used by current state-of-art approaches. We show that our approach allows the generation of a compact set of CNs that leads to superior quality answers, and that demands less resources in terms of processing time and memory. As our second contribution, we initially argue that the number of possible Candidate Networks that can be generated by any algorithm is usually very high, but that, in fact, only very few of them produce answers relevant to the user and are indeed worth processing. Thus, there is no point in wasting resources processing useless CNs. Then, based on such an argument, we present an algorithm for ranking CNs, based on their probability of producing relevant answers to the user. This relevance is estimated based on the current state of the underlying database using a probabilistic Bayesian model we have developed. By doing so we are able do discard a large number of CNs, ultimately leading to better results in terms of quality and performance. Our claims and proposals are supported by a comprehensive set of experiments we carried out using several query sets and datasets used in previous related work and whose results we report and analyse here.