PROJECT TITLE :

Search Result Diversity Evaluation Based on Intent Hierarchies - 2018

ABSTRACT:

Search result diversification aims at returning diversified document lists to hide different user intents of a question. Existing diversity measures assume that the intents of a question are disjoint, and don't think about their relationships. In this Project, we tend to introduce intent hierarchies to model the relationships between intents, and present four weighing schemes. Based mostly on intent hierarchies, we propose several hierarchical measures that take under consideration the relationships between intents. We demonstrate the feasibility of hierarchical measures by employing a new take a look at assortment based on TREC Internet Track 2009-2013 diversity take a look at collections and by using NTCIR-11 IMine check assortment. Our main experimental findings are: (one) Hierarchical measures are additional discriminative and intuitive than existing measures. In terms of intuitiveness, it is preferable for hierarchical measures to use the whole intent hierarchies than to use solely the leaf nodes. (two) The types of intent hierarchies used have an effect on the discriminative power and intuitiveness of hierarchical measures. We tend to counsel the most effective type of intent hierarchies to be used per whether or not the nonuniform weights are available. (3) To live the benefits of the diversification algorithms that use automatically mined hierarchical intents, it's important to use hierarchical measures rather than existing measures.