Group attention for collaborative filtering with sequential feedback and context aware attributes.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommender systems are vital for predicting user preferences based on past behavior.
• Collaborative Filtering (CF) is a common method, but struggles with individual rating variations and complex item attributes.
• Existing attention-based methods lack reliability in capturing intricate item-attribute relationships and providing global explanations.

Purpose of the Study:

• To propose GCORec, a novel framework for enhancing recommender systems.
• To integrate short- and long-term user preferences effectively.
• To overcome limitations in modeling fine-grained item-attribute relationships and providing comprehensive explanations.

Main Methods:

• GCORec utilizes a Hierarchical Attention Network for complex item-attribute relationships.
• A Group-wise enhancement mechanism reduces noise and emphasizes important features.
• An Attentive Bi-Directional GRU models long-term user behaviors, complemented by Collaborative Multi Head Attention for attribute influence.

Main Results:

• GCORec demonstrated significant improvements over baseline methods on benchmark datasets.
• Achieved 3.03% and 1.49% improvements in Recall@20.
• Showcased 5.88% and 5.92% improvements in NDCG@20 on datasets with varying sparsity and domain features.

Conclusions:

• GCORec effectively integrates short- and long-term user preferences for improved recommendation accuracy.
• The framework's novel mechanisms successfully capture intricate relationships and provide robust explanations.
• GCORec represents a significant advancement in recommender system performance, particularly in handling complex user-item dynamics.