Diversifying recommendations on sequences of sets

Diversifying recommendations on a sequence of sets (or sessions) of items captures a variety of applications. Notable examples include recommending online music playlists, where a session is a channel and multiple channels are listened to in sequence, or recommending tasks in crowdsourcing, where a session is a set of tasks and multiple task sessions are completed in sequence. Item diversity can be defined in more than one way, e.g., as a genre diversity for music, or as a function of reward in crowdsourcing. A user who engages in multiple sessions may intend to experience diversity within and/or across sessions. Intra session diversity is set-based, whereas Inter session diversity is naturally sequence-based. This novel formulation gives rise to four bi-objective problems with the goal of minimizing or maximizing Inter and Intra diversities. We prove hardness and develop efficient algorithms with theoretical guarantees. Our experiments with human subjects on two real datasets show that our diversity formulations do serve different user needs and yield high user satisfaction. Our large-scale experiments on real and synthetic data empirically demonstrate that our solutions satisfy our theoretical bounds and are highly scalable, compared to baselines.