Εξόρυξη και Διαχείριση Περιεχομένου Παραγόμενο από Χρήστες και Προτιμήσεων

Abstract

Ιn this thesis, we present techniques to manage the results of expressive queries, such as skyline, and mine online content that has been generated by users. Given the numerous scenarios and applications where content mining can be applied, we focus, in particular, to two cases: review mining and social media analysis. More specifically, we focus on preference queries, where users can query a set of items, each associated with an attribute set. For each of the attributes, users can specify their preference on whether to minimize or maximize it, e.g., "minimize price", "maximize performance", etc. Such queries are also know as "pareto optimal", or "skyline queries". A drawback of this query type is that the result may become too large for the user to inspect manually. We propose an approach that addresses this issue, by selecting a set of diverse skyline results. We provide a formal definition of skyline diversification and present efficient techniques to return such a set of points. The result can then be ranked according to established quality criteria. We also propose an alternative scheme for ranking skyline results, following an information retrieval approach.