GSRec: A Hybrid Sequential Recommendation Model Combining GRU and SASRec
Author
Term
4. term
Education
Publication year
2025
Submitted on
2025-06-05
Pages
44
Abstract
Sequential recommendation is key in modern recommender systems to capture user behavior in data. Two approaches for this have proliferated: neural networks and transformers. That is why we propose a new hybrid model, GSRec, for sequential recommendation, which combines the strengths of the SASRec transformer and the GRU neural network. We integrate them in a sequential pipeline to capture both short-term and longterm user preferences. SASRec is first applied to extract recent patterns in user data using an attention layer, followed by GRU to model longer-term dependencies. We evaluate GSRec on two datasets, Amazon Beauty and MovieLens 1M to see performance in comparison to state-of-the-art models such as POP, GRU4Rec, SASRec, and BERT4Rec. While BERT4Rec outperforms GSRec in dense datasets, our model shows robustness in sparser environments.
Keywords
Documents