MRIF: Multi-resolution Interest Fusion for Recommendation

In this paper, we introduce multi-resolution interest interests based on their historical behaviors. Most of recent advances fusion model (MRIF) composed of interest extraction layer, interest in recommender systems mainly focus on modeling users' aggregation layer, and attentional fusion structure, which addresses preferences accurately using deep learning based approaches. There the problem of extracting users' preferences at different temporalranges are two important properties of users' interests, one is that users' and combining multi-resolution interests effectively. The interests are dynamic and evolve over time, the other is that users' main contributions are: interests have different resolutions, or temporal-ranges to be precise, such as long-term and short-term preferences. Existing approaches - We design a new network structure to model the dynamic either use Recurrent Neural Networks (RNNs) to address changes and different temporal-ranges of users' interests, the drifts in users' interests without considering different temporalranges, which yields more accurate prediction results than extracting or design two different networks to model long-term and main interest directly from interaction sequences.