A general tail item representation enhancement framework for sequential recommendation

Recently advancements in deep learning models have significantly facilitated the development of sequential recommender systems (SRS). However, the current deep model structures are limited in their ability to learn high-quality embeddings with insufficient data. Meanwhile, highly skewed long-tail distribution is very common in recommender systems.

To solve the problems, a research team led by Qi Liu published their new research on 15 December 2024 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.

The team proposed a sequential recommendation framework named TailRec that enables contextual information of tail item well leveraged and greatly improves its corresponding representation. In contrast to previous works, this approach allows for the mining of contextual information from cross sequence behaviors to boost the performance of sequential recommendations. Such a light contextual filtering component is plug-and-play for a series of SRS models.

In the research, the core idea is to leverage contextual information of tail items to further improve their representation quality. The surrounding interaction records of each tail item are considered as contextual information without using any additional side information. Since co-occurring items are typically consumed by users with similar interests, this setting makes sense. Each tail item has different contextual information because any item can be consumed by multiple users. Therefore, we allow the contextual representation module to dynamically update during training to fully utilize this cooccurrence information. Notably, cross-sequence information can be effectively mined in sequential recommendation tasks. The representation quality of tail items can be significantly improved by incorporating contextual information, which enhances both the recommendation results and training efficiency. We evaluate TailRec over three benchmark recommender models for sequential recommendation tasks and demonstrate its effectiveness and efficiency. The experiments are conducted on multiple benchmark recommenders and datasets. The experimental results demonstrate that TailRec can not only produces more promising recommendation performances than baselines but also significantly accelerate the training process of benchmark recommender models.

Future work can focus on finding more suitable contextual information so as to achieve more effective recommendation performance for sequential recommender systems.

DOI: 10.1007/s11704-023-3112-y