On Including the User Dynamic in Learning to Rank
Accepted at SIGIR ’17: ACM Conference on Research and Development in Information Retrieval .
Abstract. Ranking query results eﬀectively by considering user past behaviour and preferences is a primary concern for IR researchers both in academia and industry. In this context, LtR is widely believed to be the most eﬀective solution to design ranking models that account for user-interaction features that have proved to remarkably impact on IR eﬀectiveness. In this paper, we explore the possibility of integrating the user dynamic directly into the LtR algorithms. Speciﬁcally, we model with Markov chains the behaviour of users in scanning a ranked result list and we modify Lambda-Mart, a state-of-the-art LtR algorithm, to exploit a new discount loss function calibrated on the proposed Markovian model of user dynamic. We evaluate the performance of the proposed approach on publicly available LtR datasets, ﬁnding that the improvements measured over the standard algorithm are statistically signiﬁcant.