Implementing ranking strategies using text signatures

Reviewer: Jane B. Grimson

The dramatic growth of office information systems in recent years has given a tremendous boost to research in information retrieval. This paper addresses the important issue of providing efficient and effective methods for the storage and retrieval of text and document files. Its particular focus is on signature files. Signature files aim to improve the efficiency, rather than the effectiveness, of retrieval from text files. The addition of ranking strategies improves effectiveness but imposes a storage and efficiency penalty. The authors examine the trade-offs and evaluate a number of different techniques for the implementation of probabilistic ranking strategies with both sequential and bit-sliced signature files. They compare these strategies to ranking techniques that use term-based document representatives and inverted files. They conclude that little choice exists between term-based and signature versions for sequential organizations. If an inverted or bit-sliced organization is selected to improve efficiency, then the term-based file would give significantly better performance at the cost of increased storage. This is a well-written and -presented paper. For those not familiar with the field, the terms used are all clearly explained. The authors draw on results from the Esprit MULTOS project, which has developed a multimedia document server. Such a background ensures that the authors focus on the practical issues rather than on the underlying mathematical theory and results in a very readable paper.