COMET: Convolutional Dimension Interaction for Collaborative Filtering

Abstract

Latent factor models play a dominant role among recommendation techniques. However, most of the existing latent factor models assume embedding dimensions are independent of each other, and thus regrettably ignore the interaction information across different embedding dimensions. In this paper, we propose a novel latent factor model called COMET (COnvolutional diMEnsion inTeraction), which provides the first attempt to model higher-order interaction signals among all latent dimensions in an explicit manner. To be specific, COMET stacks the embeddings of historical interactions horizontally, which results in two "embedding maps" that encode the original dimension information. In this way, users' and items' internal interactions can be exploited by convolutional neural networks with kernels of different sizes and a fully-connected multi-layer perceptron. Furthermore, the representations of users and items are enriched by the learnt interaction vectors, which can further be used to produce the final prediction. Extensive experiments and ablation studies on various public implicit feedback datasets clearly demonstrate the effectiveness and the rationality of our proposed method.

Attention over Self-attention:Intention-aware Re-ranking with Dynamic Transformer Encoders for Recommendation

“…We adopt the same evaluation metrics with PRM [7] to evaluate the performance of all methods: Precision (Pre@k) and Mean Average Precision (MAP@k), where k is the length of the recommendation list. Precision evaluates the fraction of correct recommendations in recommendation lists for all users, and MAP computes the mean average precision of all ranked lists cut off by k. In addition, we also evaluated the recommendation performance by normalized discounted cumulative gain (NDCG@k) which takes the position of correct recommendations into account [4], [5]. Note that higher metric values indicate a better recommendation performance.…”

“…I N the era of big data, recommender systems are widely adopted by the online platforms (e.g., Amazon and Youtube), so as to alleviate the problem of information overload [1], [2]. Accordingly, latent factor models, e.g., matrix factorization [3], [4], and deep learning models, e.g., NeuMF [5], have demonstrated their effectiveness to achieve personalized recommendations by learning user and item representations. Despite the great success, one fundamental assumption of the above solutions is that a global ranking model is designed to optimize the overall performance of item recommendations.…”

User-specific recommender systems: from data to model

“…Precision evaluates the fraction of correct recommendations in recommendation lists for all users, and MAP computes the mean average precision of all ranked lists cut off by 𝑘. In addition, we also evaluated the recommendation performance by the normalized discounted cumulative gain (NDCG@𝑘) which takes the position of correct recommendations into account [39,60]. Note that higher metric values indicate a better recommendation performance.…”

“…This chapter has been submitted to ACM Transactions on Intelligent Systems and Technology (TIST)[60].…”