Uniform Sequence Better: Time Interval Aware Data Augmentation for Sequential Recommendation

Sequential recommendation is an important task to predict the next-item access based on a sequence of interacted items. Most existing works learn user preference as transition pattern from previous item next one, ignoring time interval between these two However, we observe that in may vary significantly different, and thus result ineffectiveness modeling due issue drift. In fact, conducted empirical study validate this observation, found with uniformly distributed (denoted uniform sequence) more beneficial for performance improvement than greatly varying interval. Therefore, propose augment data perspective interval, which not studied literature. Specifically, design five operators (Ti-Crop, Ti-Reorder, Ti-Mask, Ti-Substitute, Ti-Insert) transform original non-uniform consideration variance intervals. Then, devise control strategy execute augmentation sequences different lengths. Finally, implement improvements state-of-the-art model CoSeRec our approach four real datasets. The experimental results show reaches better other 9 competing methods. Our implementation available: https://github.com/KingGugu/TiCoSeRec.