Date of Award


Publication Type


Degree Name



Computer Science


Recommendations, Recommender System, Transformer, User Interests


Ziad Kobti


Aznam Yacoub



Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


In e-commerce, a sequential recommender system is often used to predict the item that the user is likely to select next. This prediction can be used to create a recommender system to assist the user in making selections. However, when the user’s interests evolve over time, it becomes challenging to make such personalized recommendations. A more accurate recommender system thus needs to effectively interpret and adapt to a user’s changing interests by considering user’s long-term and short-term interests. Many attention-based methods focus on a user’s last clicked item to learn short-term interests. However, this approach may not consistently represent the user’s actual preferences, leading to irrelevant recommendations. Recent advancements in attention mechanisms have given rise to state-of-the-art models for addressing this task, focusing on temporal signals. In comparison, some models used a self-attention mechanism, considering the time interval between a user’s current and target interactions. These approaches aim to capture evolving short-term interests, utilizing a unidirectional method to learn user’s behavior. Transformers4Rec stands out by considering item recency to gain a better understanding of user’s short- term interests, proving to be highly effective among existing models. This model employs bidirectional attention to comprehend user’s complex behavior. Despite its advantages, Transformers4Rec does not distinguish between user’s short-term and long-term interests, emphasizing only short-term interests. Considering user’s both interests is equally essential to provide more personalized recommendations. Therefore, we introduced a model named XLNet4Rec, which captures user’s short- term interests by considering item recency and observing the complete user behavior sequence to capture long-term interests. Integrating both interests contributes to a final prediction, enabling a deeper understanding of user complex behavior and providing more personalized recommendations. Experimental evaluations on Movielens and REES46 datasets show that our model outperforms baseline models, demonstrating notable advancements in sequential based recommendation by providing accurate and personalized recommendations.