Dynamic diversification for interactive complex search

Many real-world searches are examples of complex information needs such as exploratory, comparative or survey oriented searches. In these search scenarios, users engage interactively with search systems to tackle their information needs. On one hand, user interactions can be leveraged to induce search intents and reformulate queries. On the other hand, the nature of these scenarios introduces constraints in the search process. For instance, systems are expected to satisfy the information needs earlier in the interaction. This research investigates a dynamic diversification approach that observes user interactions and dynamically changes its behaviour in response. In this research, we investigated how dynamic diversification methods should be evaluated. In that regard, we studied and analysed a wide range of offline metrics that model topical relevance novelty and user effort. In addition, this research investigates how to exploit user interactions to develop dynamic diversification methods. In particular, we study the impact of the different dimensions of user relevance feedback, the internal components of relevance feedback algorithms and diversification methods on the overall performance of dynamic diversification methods. Lastly, we intend to measure user satisfaction with these methods using a controlled user study.