Continuous-Time Sequential Recommendation with Temporal Graph Collaborative Transformer (2108.06625v2)

Abstract: In order to model the evolution of user preference, we should learn user/item embeddings based on time-ordered item purchasing sequences, which is defined as Sequential Recommendation (SR) problem. Existing methods leverage sequential patterns to model item transitions. However, most of them ignore crucial temporal collaborative signals, which are latent in evolving user-item interactions and coexist with sequential patterns. Therefore, we propose to unify sequential patterns and temporal collaborative signals to improve the quality of recommendation, which is rather challenging. Firstly, it is hard to simultaneously encode sequential patterns and collaborative signals. Secondly, it is non-trivial to express the temporal effects of collaborative signals. Hence, we design a new framework Temporal Graph Sequential Recommender (TGSRec) upon our defined continuous-time bi-partite graph. We propose a novel Temporal Collaborative Trans-former (TCT) layer in TGSRec, which advances the self-attention mechanism by adopting a novel collaborative attention. TCT layer can simultaneously capture collaborative signals from both users and items, as well as considering temporal dynamics inside sequential patterns. We propagate the information learned fromTCTlayerover the temporal graph to unify sequential patterns and temporal collaborative signals. Empirical results on five datasets show that TGSRec significantly outperforms other baselines, in average up to 22.5% and 22.1%absolute improvements in Recall@10and MRR, respectively.

Continuous-Time Sequential Recommendation with Temporal Graph Collaborative Transformer

The paper "Continuous-Time Sequential Recommendation with Temporal Graph Collaborative Transformer" presents an innovative approach to enhancing recommendation systems by unifying sequential patterns and temporal collaborative signals. The central contribution of this work is the development of the Temporal Graph Sequential Recommender (TGSRec), a novel framework designed to capture the temporal dynamics inherent in user-item interactions.

Methodological Framework

The paper critiques existing sequential recommendation methods that primarily focus on modeling item transitions but overlook the integration of temporal collaborative signals. In response, the authors propose TGSRec, which operates on a continuous-time bipartite graph constructed from user-item interactions. This graph serves as the foundation for learning user and item embeddings that account for both sequential and temporal collaborative influences.

A key component of TGSRec is the Temporal Collaborative Transformer (TCT) layer. This layer advances traditional self-attention mechanisms by employing a novel collaborative attention model. The TCT layer simultaneously captures the temporal dynamics of sequential patterns and the collaborative signals from both users and items. By propagating the output of the TCT layer across the temporal graph, TGSRec effectively unifies these two critical components to improve recommendation quality.

Empirical Evaluation

The efficacy of TGSRec is demonstrated through empirical analysis on five distinct datasets. The results indicate that TGSRec significantly outperforms existing benchmarks, including state-of-the-art transformer-based models like SASRec, by up to 22.5% in Recall@10 and 22.1% in Mean Reciprocal Rank (MRR). These findings suggest that incorporating temporal collaborative signals into recommendation systems can noticeably enhance their performance.

Theoretical and Practical Implications

The introduction of TGSRec has both theoretical and practical implications. Theoretically, it expands the understanding of sequential recommendation by considering the temporal dimension alongside traditional sequential patterns. This dual consideration allows for more precise modeling of user preferences as they evolve over time.

Practically, TGSRec offers a robust framework for developing recommendation systems in various applications where user behavior is time-dependent. The framework's ability to generalize temporal embeddings beyond the training data further elevates its utility, making it highly suitable for real-world deployment in dynamic environments.

Future Directions

Looking forward, the research opens several avenues for further exploration:

1. Scalability: As the TGSRec model is deployed on larger datasets, further investigations into its scalability and performance optimization across more extensive interaction networks would be beneficial.
2. Hybrid Models: Integrating TGSRec with other recommendation paradigms, such as content-based filtering or knowledge graph embeddings, might reveal synergistic effects that could drive further improvements.
3. Time-sensitive Applications: The application of TGSRec in domains where timing is critical, such as real-time advertising or personalized news delivery, could yield valuable insights and commercial advantages.

In conclusion, the Temporal Graph Collaborative Transformer presents a substantial advancement in the field of sequential recommendation. By thoughtfully integrating temporal signals into the recommendation process, TGSRec stands as a promising solution poised to handle the intricacies of evolving user preferences with precision and clarity.