Time interval enhanced contrastive learning sequence recommendation

Existing sequence recommendation models usually ignore the time interval of the sequence when modeling long-term preferences and short-term interests, which may lead to bias in user interests and fail to model user interests well for recommendations. In this paper, Time interval enhanced contrastive learning sequence recommendation (TICLSR) is proposed. Specifically, when modeling long-term preferences, we change the non-uniform sequence into a uniform sequence by enhancing the time interval, so that the model can model long-term preferences more effectively. In addition, we decouple long-term preferences and short-term interests, introduce contrastive learning, and explicitly model long-term preferences and short-term interests to obtain more accurate preference information. Comparative experiments with baseline models show that our proposed model exhibits better performance.