基于自监督学习的序列推荐算法

针对现有序列推荐算法易受数据稀疏影响以及对用户短期动态偏好建模不充分的问题，提出基于自监督学习的序列推荐算法。针对短期序列中的原始项目关系更易受到随机数据增强破坏的问题，对长短期序列使用不同的数据增强方法来构建更有效的自监督信号；利用对比式自监督学习框架对用户长期偏好和短期偏好进行多任务联合建模；针对现有自注意力机制无法建模序列中项目相对位置关系的问题，将自然语言处理领域中的解耦注意力机制引入到用户短期偏好学习过程中，充分捕获用户短期序列中项目的相对位置信息。实验结果证明了所提算法的有效性。

Sequence recommendation algorithm based on self-supervised learning

In order to relieve the problem that existing sequence recommendation algorithms are vulnerable to data sparsity, and the modeling of short-term dynamic preferences of users is insufficient, we propose a sequence recommendation algorithm based on self-supervised learning. To address the issue that the original item relations in short-term sequences are more susceptible to random data augmentation, we use different data augmentation methods for long and short-term sequences to construct more effective self-supervised signals. A multi-task joint modeling of users'' long-term and short-term preferences is carried out using a contrastive self-supervised learning framework. Because of the problem that the existing self-attention mechanism cannot model the relative positional relationship of items in the sequence, the disentangled attention mechanism in the field of natural language processing is introduced into the user''s short-term preference learning process to fully capture the relative position information of items in the user''s short-term sequence. The experimental results demonstrate the effectiveness of the proposed algorithm.