Accurate user interest modeling is vital for recommendation scenarios. One of the effective solutions is the sequential recommendation that relies on click behaviors, but this is not elegant in the video feed recommendation where users are passive in receiving the streaming contents and return skip or no-skip behaviors instead of active click behavior. Here skip and no-skip behaviors can be treated as negative and positive feedback, respectively. Indeed, skip and no-skip are not simply positive or negative correlated, so it is challenging to capture the transition pattern of positive and negative feedback. To do so, FeedRec has exploited a shared vanilla Transformer and grouped each feedback into different Transformers. Indeed, such a task may be challenging for the vanilla Transformer because head interaction of multi-heads attention does not consider different types of feedback. In this paper, we propose Dual-interest Factorization-heads Attention for Sequential Recommendation (short for DFAR) consisting of feedback-aware encoding layer, dual-interest disentangling layer and prediction layer. In the feedback-aware encoding layer, we first suppose each head of multi-heads attention can capture specific feedback relations. Then we further propose factorization-heads attention which can mask specific head interaction and inject feedback information so as to factorize the relation between different types of feedback. Additionally, we propose a dual-interest disentangling layer to decouple positive and negative interests before performing disentanglement on their representations. Finally, we evolve the positive and negative interests by corresponding towers whose outputs are contrastive by BPR loss. Experiments on two real-world datasets show the superiority of our proposed method against state-of-the-art baselines. Further ablation study and visualization also sustain its effectiveness.

翻译：精确的用户兴趣建模在推荐场景中至关重要。一种有效解决方案是基于点击行为的序列推荐，但在视频流推荐中该方法并不理想——用户被动接收流式内容并返回跳过或未跳过行为（而非主动点击行为）。在此场景中，跳过与未跳过行为可分别视作负反馈与正反馈。然而，这两类反馈并非简单的正负相关关系，因此捕获正负反馈的转移模式颇具挑战性。为此，FeedRec采用共享的标准Transformer架构，并将各类反馈分组至不同Transformer中处理。实际上，由于标准Transformer中多头注意力的头部交互未区分不同反馈类型，此类任务对其而言可能存在困难。本文提出面向序列推荐的双视角因子化多头注意力机制（简称DFAR），包含反馈感知编码层、双视角解耦层与预测层。在反馈感知编码层中，我们首先假设多头注意力的每个注意力头可捕获特定反馈关系；继而提出因子化多头注意力机制，通过掩码特定头部交互并注入反馈信息，实现不同类型反馈关系的因子化解耦。此外，我们提出双视角解耦层，在表征解耦前先分离正负兴趣。最终通过对应任务塔进化正负兴趣，其输出经BPR损失进行对比约束。在两个真实世界数据集上的实验表明，本方法显著优于现有最优基线模型。进一步的消融实验与可视化分析也验证了其有效性。