Dynamic Multi-Behavior Sequence Modeling for Next Item Recommendation

Sequential Recommender Systems (SRSs) aim to predict the next item that users will consume, by modeling user interests within their sequences. While most existing SRSs focus on a single type of behavior, only few pay attention multi-behavior sequences, although they are very common in real-world scenarios. It is challenging effectively capture because information about entangled throughout sequences complex relationships. To this end, we first address characteristics should be considered SRSs, and then propose novel methods for Dynamic Multi-behavior Sequence named DyMuS, which light version, DyMuS+, an improved considering characteristics. DyMuS encodes each behavior sequence independently, combines encoded using dynamic routing, dynamically integrates required final result from among many candidates, based correlations between furthermore, applies routing even encoding further at item-level. Moreover, release new, large up-to-date dataset recommendation. Our experiments DyMuS+ show superiority significance capturing