Multi-interest Distribution Based Diversified Recommendation

Traditional sequential recommendations often utilize a single interest representation for users, limiting the modeling of their diverse interests and resulting in a lack of recommendation diversity. Therefore, multi-interest recommendations aim to enhance accuracy and diversity by considering users' multiple interests. However, existing methods fail to fully leverage these interests, which restricts diversity enhancement. To address this issue, we propose a diversified recommendation model called MIND-DR, which utilizes multiple interest distributions to enhance recommendation diversity. Specifically, during the reranking phase, we use KL divergence to align the interest distribution of the recommendation list with the average interest distribution of users, thus generating more diverse recommendations. Additionally, considering the interdependence and interference among different interests, we employ contrastive learning loss to encourage independence among users' multiple interest representations. Experimental results on three datasets demonstrate the effectiveness of the proposed model in improving recommendation diversity while maintaining high accuracy.