CausalRec: A CausalBoost Attention Model for Sequential Recommendation

TL;DR

CausalRec introduces a causal-augmented attention framework for sequential recommendation, combining a Causal Discovery Block with identifiability guarantees and a CausalBooster that amplifies causally relevant signals in the attention mechanism. The model learns item-level causal graphs from sequences via covariance-based estimation under a linear SEM, imposes acyclicity and sparsity constraints, and integrates the learned causality into a transformer-style attention network. Empirical results on four real-world datasets show consistent improvements in NDCG@10 and HR@10 over state-of-the-art baselines, with notable gains on the Foursquare dataset, and ablations confirm the contributions of causal discovery, sparsity, and attention. The work advances the use of causality in recommender systems, offering improved predictive accuracy and explanations while addressing identifiability concerns and ensuring practical scalability.

Abstract

Recent advances in correlation-based sequential recommendation systems have demonstrated substantial success. Specifically, the attention-based model outperforms other RNN-based and Markov chains-based models by capturing both short- and long-term dependencies more effectively. However, solely focusing on item co-occurrences overlooks the underlying motivations behind user behaviors, leading to spurious correlations and potentially inaccurate recommendations. To address this limitation, we present a novel framework that integrates causal attention for sequential recommendation, CausalRec. It incorporates a causal discovery block and a CausalBooster. The causal discovery block learns the causal graph in user behavior sequences, and we provide a theory to guarantee the identifiability of the learned causal graph. The CausalBooster utilizes the discovered causal graph to refine the attention mechanism, prioritizing behaviors with causal significance. Experimental evaluations on real-world datasets indicate that CausalRec outperforms several state-of-the-art methods, with average improvements of 7.21% in Hit Rate (HR) and 8.65% in Normalized Discounted Cumulative Gain (NDCG). To the best of our knowledge, this is the first model to incorporate causality through the attention mechanism in sequential recommendation, demonstrating the value of causality in generating more accurate and reliable recommendations.

Figures (3)

• Figure 1: Illustrating the motivation for modeling causal relationships among user behaviors. Consider the example: although the cable and the phone shell frequently co-occur in the same user behavior sequence, it does not imply that they influence each other directly. Their co-occurrence arises from their shared causal relationship with the phone.
• Figure 2: Architecture of our proposed CausalRec. We propose a Causal Discovery Block to learn user behavior causality and incorporate it into our model with CausalBooster.
• Figure 3: Visualization of user interaction histories (left), recommended items (center), and CausalRec’s discovered causal graph (right). The film title in red on the left belongs to the same series as the target item in the middle (also shown in red), which is the user’s actual choice that the model seeks to recommend. Categories shown in blue indicate the target item’s category. Different colored boxes denote explanations from CausalRec (green), NARM (red), and SASRec (purple), while the orange box highlights the ground‐truth (target) item.

Theorems & Definitions (3)

• lemma 1: Identifiability Conditions for Linear SEMs Chen_2019JMLR-Park
• proposition 1
• proof