Effectiveness of Weighted Searching in an Operational IR Environment

We describe an experiment to compare the effectiveness of Boolean retrieval and weighted retrieval on a commercial data collection that contains millions of documents. The experiment was carried out via a front-end connected to an operational conventional host-oriented IR environment. In contrast to previous experiments where weighted retrieval had to be simulated on the host, our host was equipped with a built-in weighted retrieval algorithm. The results of the experiment clearly show that weighted retrieval performs significantly better than Boolean retrieval. On the other hand, no difference in the performance between manually weighted queries and automatically weighted queries could be detected.