Snippet-based result merging in federated search

Abstract

In federated search, the central broker simultaneously forwards the search query to multiple resources. The returned results from those resources are then merged into a single ranked list. An autonomous resource in a federated search system usually does not provide scores for the retrieved documents; even if some of them do, scores from different resources are incomparable due to the heterogeneity in many aspects of those resource involved such as retrieval models and corpus statistics. Many results merging approaches have been proposed in the literature to deal with this problem. However, to the best of our knowledge, none of them has utilised snippets. This article proposes a snippet-based weighting scheme for the query terms involved. It quantifies the importance of each query term from two angles: the frequency of the term and the part in which the term occurs inside a snippet. Three parts – which are URL, title, and description – are given different weights. Experiments are conducted with the TREC 2013 FedWeb data set. The results show that the proposed methods consistently outperform several baseline models. We also find in many instances, a further small slight performance improvement is achievable by an extra measure of weighting each of the resources involved, which can be done in the phase of resource selection.