| Abstract |
Preference modelling and management has attracted considerable attention in the areas of Databases, Knowledge Bases and Information Retrieval Systems in recent years. This interest stems from the fact that a rapidly growing class of untrained lay users confront vast data collections, usually through the Internet, typically lacking a clear view of either content or structure, moreover, not even having a particular object in mind. Rather, they are attempting to discover potentially useful objects, in other words, objects that best suit their preferences. A modern information system, consequently, should enable users to quickly focus on the k best object according to their preferences. In this thesis, modelling preferences as binary relations, we introduce efficient algorithms for the evaluation of the top-k objects. Previous related work treated preference expressions as black boxes and dealt with the idea of exhaustively applying dominance tests among database objects in order to determine the best ones, resulting in quadratic costs. On the contrary, we advocate a query ordering based approach. Our key idea is to exploit the semantics of the input preference expression itself, in terms of both the operators and the preferences involved, to define an ordering over those queries, whose evaluation is necessary for the retrieval of the top-k objects. We introduce two novel algorithms, LBA and TBA. LBA defines an ordering over queries which are essentially conjunctions of atomic selection conditions, containing all attributes that the user preferences involve. The algorithm ensures that the way and order in which objects are fetched respect user preferences, avoiding any dominance testing, and accessing only the top-k objects, each of them only once. From a different angle, TBA defines an order of queries which are disjunctions of atomic selection conditions over single attributes, and uses appropriate threshold values to signal object fetching termination, ensuring that all remaining objects are worse than those fetched. Dominance tests are performed only for already retrieved objects. Analytical study and experimental evaluation show that our algorithms outperform existing ones under all problem instances.
|
Top-k Αλγόριθμοι Βασισμένοι Σε Διατάξεις Επερωτήσεων Για Ποιοτικώς Καθορισμένες Προτιμήσεις
