Information filtering and information retrieval: two sides of the same coin?

Reviewer: Stephen Charles Smithson

The institutional barriers between information retrieval research (traditionally carried out in schools of library or information science) and the more mainstream computing and business information systems research are being slowly dismantled, thanks to papers like this. The authors, who come from the information retrieval research community, have done a good job of comparing recently introduced filtering systems, which operate on streams of unstructured or semi-structured data (such as news feeds and electronic mail), with information retrieval systems, where research stretches back 20 or 30 years. The authors clearly demonstrate a considerable similarity between the two types of systems and go on to argue that researchers and developers of information filtering systems could learn a lot from studying previous research into information retrieval. They focus on a probabilistic model, the inference net model, which they apply to information filtering. They emphasize the importance of indexing and the estimation of probabilities, as well as noting the efficiency problems of implementing such a model. This paper is a timely and useful comparison, involving the currently popular research area of information filtering. The argument that the use of information retrieval models, such as the probabilistic approach, can provide useful insight into the issues involved is sound. In cataloguing information retrieval research, however, the authors neglect to mention that most of the indexing and matching schemas they mention still have not been used in operational systems of any size and the performance results are mostly based on small-scale laboratory experiments.

Reviewer: Robert G Crawford

The issues are wonderfully placed in perspective in this lead paper in a CACM special section on information filtering, providing a useful framework for consideration of this subject. Three aspects of the paper give it strength. First, models for retrieval and filtering are presented and described. The authors strip away the terminology to get at the heart of what is really going on with retrieval and filtering. Second, the lessons from information retrieval research that apply to filtering are categorized and discussed. Third, the paper includes a good list of aspects of the general problem for which filtering requires further research. Filtering is discussed in the context of a specific probabilistic retrieval model, the inference net model. The analysis is clear and useful.