Mining information interaction behavior Academic papers and enterprise emails

In the first part of the thesis, we uncover how academic searchers interact with information objects. We start by characterizing academic search queries, and find that they are different from general web search queries. We also find that it is possible to suggest good query recommendations to help users that encounter query failures, using query characteristics and session information. Next, we examine user behavior over a longer period, in particular, query reformulation and topic shift. We identify multiple query reformulation strategies, and find that revisiting queries is especially common. We look for correlations between query reformulation and topic shift. Surprisingly, we find little correlation in the long term. Topic shift in the short term is correlated with certain reformulation types such as submitting new queries. We also study how users download papers. We characterize their download behavior, and observe different patterns among disciplines. Based on the observations, we propose to predict user downloads, using LSTM-based models in combination with user segmentations. In another scenario, we study a recommender that sends out paper recommendations through newsletters and propose the task of reranking the recommendations, using a hybrid reranking model that considers both content and behavior.
In the second part of the thesis, we focus on how users read their enterprise emails, and characterize user reading time. We find that reading time is correlated with many contextual factors. The results improve our understanding of user behavior on email platforms, and also shed light on system improvements to make email reading more efficient.