Abstract

Existing Web search services fail in helping users with information needs that are broad, vague, or hard to express through a set of keywords. This dissertation investigates the use of retrieval techniques based on inter-document similarity, either measured through the textual contents or the linkage between documents. Unlike traditional retrieval approaches, which match documents against keywords and produce one-dimensional ranked lists of results, techniques based on inter-document similarity offer better support for results visualization, as well as alternative ways of expressing information needs. A Portuguese Web search engine has been extended with two inter-document similarity algorithms: result set clustering and related pages. The system was evaluated in a user survey, which has shown that both algorithms are well accepted.

Citations