Balancing the Trade-Offs Between Diversity and Precision for Web Image Search Using Concept-Based Query Expansion

Even though Web image search queries are often ambiguous, traditional search engines retrieve and present results solely based on relevance ranking, where only the most common and popular interpretations of the query are considered. Rather than assuming that all users are interested in the most common meaning of the query, a more sensible approach may be to produce a diversified set of images that cover the various aspects of the query, under the expectation that at least one of these interpretations will match the searcher’s needs. However, such a promotion of diversity in the search results has the side-effect of decreasing the precision of the most common sense. In this paper, we evaluate these competing factors in the context of a method for explicitly diversifying image search results via concept-based query expansion using Wikipedia. Experiments with controlling the degree of diversification illustrate this trade-off between diversity and precision for both ambiguous and more specific queries. As a result of these experiments, an automatic method for tuning the diversification parameter is proposed based on the degree of ambiguity of the original query.