Rank-sensitive proportional aggregations in dynamic recommendation scenarios

In this paper, we focus on the problem of rank-sensitive proportionality preservation when aggregating outputs of multiple recommender systems in dynamic recommendation scenarios. We believe that individual recommenders may provide complementary views on the user’s preferences or needs, and therefore, their proportional (i.e. unbiased) aggregation may be beneficial for the long-term user satisfaction. We propose an aggregation framework (FuzzDA) based on a modified D’Hondt’s algorithm (DA) for proportional mandates allocation. Specifically, we adjusted DA to register fuzzy membership of items and modified the selection procedure to balance both relevance and proportionality criteria. Furthermore, we propose several iterative votes assignment strategies and negative implicit feedback incorporation strategies to make FuzzDA framework applicable in dynamic recommendation scenarios. Overall, the framework should provide benefits w.r.t. long-term novelty of recommendations, diversity of recommended items as well as overall relevance. We evaluated FuzzDA framework thoroughly both in offline simulations and in online A/B testing. Framework variants outperformed baselines w.r.t. click-through rate (CTR) in most of the evaluated scenarios. Some variants of FuzzDA also provided the best or close-to-best iterative novelty (while maintaining very high CTR). While the impact of the framework variants on user-wise diversity was not so extensive, the trade-off between CTR and diversity seems reasonable.

Online convex combination of ranking models

As a task of high importance for recommender systems, we consider the problem of learning the convex combination of ranking algorithms by online machine learning. First, we propose a stochastic optimization algorithm that uses finite differences. Our new algorithm achieves close to optimal empirical performance for two base rankers, while scaling well with an increased number of models. In our experiments with five real-world recommendation data sets, we show that the combination offers significant improvement over previously known stochastic optimization techniques. The proposed algorithm is the first effective stochastic optimization method for combining ranked recommendation lists by online machine learning. Secondly, we propose an exponentially weighted algorithm based on a grid over the space of combination weights. We show that the algorithm has near-optimal worst-case performance bound. The bound provides the first theoretical guarantee for non-convex bandits using limited number of evaluations under very general conditions.