Next-basket recommendation (NBR) is prevalent in e-commerce and retail industry. In this scenario, a user purchases a set of items (a basket) at a time. NBR performs sequential modeling and recommendation based on a sequence of baskets. NBR is in general more complex than the widely studied sequential (session-based) recommendation which recommends the next item based on a sequence of items. Recurrent neural network (RNN) has proved to be very effective for sequential modeling, and thus been adapted for NBR. However, we argue that existing RNNs cannot directly capture item frequency information in the recommendation scenario. Through careful analysis of real-world datasets, we find that personalized item frequency (PIF) information (which records the number of times that each item is purchased by a user) provides two critical signals for NBR. But, this has been largely ignored by existing methods. Even though existing methods such as RNN based methods have strong representation ability, our empirical results show that they fail to learn and capture PIF. As a result, existing methods cannot fully exploit the critical signals contained in PIF. Given this inherent limitation of RNNs, we propose a simple item frequency based k-nearest neighbors (kNN) method to directly utilize these critical signals. We evaluate our method on four public real-world datasets. Despite its relative simplicity, our method frequently outperforms the state-of-the-art NBR methods-including deep learning based methods using RNNs-when patterns associated with PIF play an important role in the data.
|Original language||English (US)|
|Title of host publication||SIGIR 2020 - Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval|
|Publisher||Association for Computing Machinery, Inc|
|Number of pages||10|
|State||Published - Jul 25 2020|
|Event||43rd Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2020 - Virtual, Online, China|
Duration: Jul 25 2020 → Jul 30 2020
|Name||SIGIR 2020 - Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval|
|Conference||43rd Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2020|
|Period||7/25/20 → 7/30/20|
Bibliographical noteFunding Information:
Beyond this work, we believe that there are two directions that deserve to be explored. First, a direct extension is whether there are other commonly-used functions which are hard to be learned by existing widely-used deep models. This direction can help us better understand how to apply deep learning based methods in recommendation systems as we observe that recent publications  show a worry about the unclear progress in sequential/session-based recommendation. We believe different types of methods should have different advantages in different tasks and data sets. And a deep understanding about the boundary of the deep learning methods can bring benefits not only to recommendation systems but also to other machine learning areas. Second, another direct extension is to investigate if there are other patterns associated with PIF or other patterns that are associated with different types of item frequency, e.g., global item frequency, local item frequency (the item frequency associated with a small group of users or a small group of items), and inverse item frequency . Acknowledgement: This research was supported part by NSF under grants CNS-1814322, CNS-1831140, CNS-1901103, US DoD DTRA DTRA grant HDTRA1-14-1-0040, and National Natural Science Foundation of China (61972372, U19A2079). Also, thanks to the constructive suggestions from the reviewers.
© 2020 ACM.
- item frequency
- k-nearest neighbors
- next-basket recommendation
- recurrent neural networks