CALRec: Counterfactual Active Learning for Sequential Recommendation

Sequential recommendation prevails among real-world applications like e-commerce, marketing, and entertainment. On the other hand, the problem of data sparsity in sequential recommendation has attracted much attention. To this issue, a popular solution is data augmentation. However, previous trivial augmentation methods, such as cropping and reordering, are introducing too much randomness due to a lack of guidance. In this paper, we leverage Active Learning strategies to help with model-based guided data augmentation. We propose a novel framework called CALRec(Counterfactual Active Learning for Sequential Recommendation). Experiments on several real-world datasets demonstrate our framework’s effectiveness and generalizability.