Hide/Show Apps

Search result diversification for selective search

Küçükoglu, Emre Can
Our work explores the performance of result diversification methods in the selective search scenario, where the underlying document collection is topically partitioned across several nodes and the search is conducted only at a subset of these nodes. In particular, we investigate whether diversification at each node is superior to previous approaches in the literature, i.e., diversification at the broker node applied before the resource selection or after the result merging stages. We also compare performance of different centralized sample indexes to show their effect on diversification. Fi- nally, we explore the impact of recently introduced query expansion techniques using word embeddings to improve the effectiveness of diversification applied at the broker node, and subsequently, overall diversification. Our experiments reveal that for im- plicit diversification methods, expanding queries with diversified terms and applying diversification during the resource selection stage yield the best performance. In con- trary, for explicit diversification methods, diversifying merged results at the broker is the best solution.