A Dynamic Meta-Learning Model for Time-Sensitive Cold-Start Recommendations

We present a novel dynamic recommendation model that focuses on users who have interactions in the past but turn relatively inactive recently. Making effective recommendations to these time-sensitive cold-start is critical maintain user base of recommender system. Due sparse recent interactions, it challenging capture users' current preferences precisely. Solely relying their historical may also lead outdated misaligned with interests. The proposed leverages and user-item dynamically factorizes user's (latent) preference into time-specific time-evolving representations jointly affect behaviors. These latent factors further interact an optimized item embedding achieve accurate timely recommendations. Experiments over real-world data help demonstrate effectiveness model.