RC-Index: Diversifying Answers to Range Queries

Query result diversification is widely used in data exploration, Web search, and recommendation systems. The problem of returning diversified query results consists of finding a small subset of valid query answers that are representative and different from one another, usually quantified by a diversity score. Most existing techniques for query diversification first compute all valid query results and then find a diverse subset. These techniques are inefficient when the set of valid query results is large. Other work has proposed efficient solutions for restricted application settings, where results are shared across multiple queries. In this paper, our goal is to support result diversification for general range queries over a single relation. We propose the RC-Index, a novel index structure that achieves efficiency by reducing the number of items that must be retrieved by the database to form a diverse set of the desired size (about 1 second for a dataset of 1 million items). Further, we prove that an RC-Index offers strong approximation guarantees. To the best of our knowledge, this is the first index-based diversification method with a guaranteed approximation ratio for range queries.