Abstract:
We present a three-stage system for information exploration and retrieval. Given a collection of texts, the first stage extracts a set of index terms describing each text. The second stage, named GALOIS, exploits the indices determined in the earlier stage to build a particular lattice of clustered texts. The resulting structure supports the third stage, a visual retrieval interface named ULYSSES. ULYSSES provides multiple and integrated retrieval strategies; the user may browse or query the clustered search space, or, in addition, he may restrict it by specifying constraints that the desired texts have to satisfy. The system, as a whole, has two main advantages. The first is that it is fully automatic and works in unrestricted environments. The second is that it combines browsing capabilities with good retrieval performance. In fact, the experiments conducted so far suggest that it can be seen as an alternative to more conventional systems even for retrieval tasks such as subject searching and for databases of reasonable size.