NovaSearch at TREC 2013 Federated Web Search Track: Experiments with rank fusion

We propose an unsupervised late-fusion approach for the results merging task, based on combining the ranks from all the search engines. Our idea is based on the known pressure for Web search engines to put the most relevant documents at the very top of their ranks and the intuition that relevance of a document should increase as it appears on more search engines [9]. We performed experiments with state-of-the-art rank fusion algorithms: RRF and Condorcet Fuse and our proposed method: Inverse Square Rank (ISR) fusion algorithm. Rank fusion algorithms have low computational complexity and do not need engines to return document scores nor training data. Inverse Square Rank is a novel fully unsupervised rank fusion algorithm based on quadratic decay and on logarithmic document frequency normalization. The results achieved in the competition were very positive and we were able to improve them further post-TREC.