Distant Supervision for Information Retrieval
Search engines often require large-scale training data to improve their services. My Ph.D. dissertation at UMass (expected to finish in Summer 2018) studies automatic methods for generating such training data using public and unlabeled corpora. The proposed techniques make it possible to train cutting-edge retrieval techniques without Google-scale data.
Faithful Evaluation of Search and Conversational Systems
I develop techniques to evaluate user experience in search and other systems automatically in real-time based on behavioral signals and user modeling. My work has been applied to evaluate user satisfaction with Bing and Microsoft Cortana.
Interactive Search Behavior & Systems
I study human behavior in search and conversational systems based on search logs or laboratory user studies. I also design intelligent interactive search systems that are adaptive to users' potential behavior, including both promoting beneficial activities and avoiding harmful ones.