M2: Mixed Models with Preferences, Popularities and Transitions for Next-Basket Recommendation

Next-basket recommendation considers the problem of recommending a set of items into the next basket that users will purchase as a whole. In this paper, we develop a novel mixed model with preferences, popularities and transitions (M2) for next-basket recommendation. This method explicitly models three important factors in next-basket generation process: 1) users' general preferences, 2) items' global popularities and 3) transition patterns among items. We also propose a simple encoder-decoder based framework (ed-Trans) to better model the transition patterns among items. We compared M2 with 5 state-of-the-art next-basket recommendation methods on 4 public benchmark datasets. Our experimental results demonstrate that M2 significantly outperforms the state-of-the-art methods on all the datasets, with an improvement as much as 19.0% at recall@5. We also compared M2 with these baseline methods in recommending the second next and third next baskets. Our experimental results demonstrate that M2 could consistently outperform the baseline methods in all these tasks, with an improvement as much as 14.4% at recall@5. In addition, we conducted a comprehensive ablation study to verify the effects of the different factors. The results show that learning all the factors together could significantly improve the recommendation performance compared to learning each of them alone. The results also show that ed-Trans in learning item transitions among baskets could outperform recurrent neural network-based methods on the benchmark datasets, with an improvement as much as 20.4% at recall@5. We also have a thorough discussion on various experimental protocols and evaluation metrics for next-basket recommendation evaluation.