Learning to rank (LTR) is a core problem in information retrieval and machine learning with numerous applications in web search, recommender systems and ad placement The goal of LTR is to present a list of K documents out of L that maximizes the satisfaction of the user This problem has been traditionally solved by training supervised learning models on manually annotated relevance judgments However, strong evidence suggests that users click feedback can lead to major improvements over supervised LTR methods In addition, billions of users interact daily with commercial LTR systems, and it is finally feasible to interactively and adaptive maximize the satisfaction of these users from clicks These observations motivated numerous papers on online LTR methods, which utilize user feedback to improve the quality of ranked lists These methods can be divided into two groups: learning the best ranker in a family of rankers, and learning the best list under some model of user interaction with the list, such as a click model The click model is a stochastic model of how the user examines and clicks on a list of items This talk will focus on online LTR in click models and present the algorithms from the setting of specific click models, like cascade model, to general click models

Shuai Li is currently a 4th year PhD student in the Chinese University of Hong Kong. She received Bachelor’s Degree in Math from Chu Kochen Honors College, Zhejiang University and received Master’s Degree in Math from Institute of Mathematics. During PhD, she has visited MSRA, Huawei Noah’s Ark Lab, University of Alberta (CA) and UC Berkeley (US), and interned at Adobe Research (US) and DeepMind (UK). She is a recipient of Google PhD Fellowship this year.