Debiased Causal Inference for Sequential Recommendation

Learning dynamic user preferences from sequential user behavior data is a hot topic in recommender system research. Common sequence learning methods include Recurrent Neural Networks (RNNs), Transformer, and Graph Neural Networks (GNNs). However, in real-world environments, user behavior data recorded on online platforms often suffer from various biases, such as selection bias, position bias, exposure bias, and popularity bias, which affect the representation ability of sequence learning. Although contrastive learning (CL) methods are commonly used to address noise and sparsity issues, we argue that CL-based methods have limitations in addressing biases and identifying users' real interests. In this paper, we propose a new Debiased Causal Inference for Sequential Recommendation (DCISRec), which models the counterfactual data distribution to learn accurate and robust user representations, uncover indispensable behaviors in the sequence, and reduce the impact of noisy behaviors. Simultaneously integrating sequence denoising encoding methods to enhance the learning ability of sequence relationships. Extensive experiments on three real-world datasets have demonstrated that our proposed method significantly outperforms state-of-the-art baselines, indicating its efficacy for recommendation.