Graph convolutional network and self-attentive for sequential recommendation

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Sequential recommender systems (SRS) are crucial for personalized recommendations but struggle with suboptimal item embedding representations.
• Existing methods often suffer from item embedding vector concentration, limiting prediction accuracy.

Purpose of the Study:

• To propose a novel model, GSASRec, that improves item representation learning in SRS.
• To enhance the capture of complex user-item relationships and disperse item embeddings.

Main Methods:

• Utilizing a graph convolutional neural network (GCN) to model user-item interactions and learn node embeddings.
• Employing a self-attentive sequential model for outcome prediction based on user item embedding sequences.
• Incorporating instance-wise contrastive learning (ICL) and prototype contrastive learning (PCL) to refine representation learning.

Main Results:

• The proposed GSASRec model demonstrated superior performance across four diverse datasets.
• Ablation studies confirmed the effectiveness of the integrated GCN, self-attention, and contrastive learning components.
• The method successfully dispersed item embeddings, overcoming limitations of prior techniques.

Conclusions:

• GSASRec offers a significant advancement in sequential recommender systems by effectively addressing item embedding challenges.
• The combination of graph networks and contrastive learning provides a robust framework for representation learning in SRS.
• This approach leads to demonstrably better personalized recommendations.