MUEU: Meta-Learned User Preference Estimation for Cold-Start Recommendation via enriching the representations of users

In recommender systems, a long-standing challenge is the difficulty of capturing user preferences due to sparse user interactions, commonly referred to as the cold-start problem. For newly registered users, the lack of interaction data hinders the inference of their preferences; similarly, for recently registered users, limited interaction data may be insufficient to accurately identify their preferences. In recent years, meta-learning-based approaches have shown promise in alleviating this issue. These methods can fine-tune models with minimal interactions to capture user-specific characteristics. However, the preferences inferred through these methods are still suboptimal because, in the cold-start scenario, a few interactions are insufficient to fully represent a user’s characteristics. To address this challenge, we propose a new meta-learning model, MUEU. Considering the sparsity of user vectors, we propose a simple and efficient method for identifying similar users by combining intrinsic user attributes with historical interaction data, further optimizing similarity identification of user attributes through the Jaccard similarity coefficient to identify other users with similar characteristics. We then use information from these similar users to enrich the representation of the target user, while employing a weighted attention mechanism to integrate the similar users with the target user, thereby creating a more accurate user preference estimator. Experimental results demonstrate that our approach outperforms state-of-the-art methods.