Bid Farewell to Seesaw: Towards Accurate Long-tail Session-based Recommendation via Dual Constraints of Hybrid Intents
Researchers have developed a new framework called HID (Hybrid Intent-based Dual Constraint Framework) to improve session-based recommendation systems. This framework addresses the common issue where promoting less popular, long-tail items can degrade recommendation accuracy. HID introduces hybrid intent learning, using attribute-aware spectral clustering to better map items to intents and distinguish session-irrelevant noise. It also incorporates an intent constraint loss that balances diversity and accuracy, leading to state-of-the-art performance in long-tail recommender systems. AI
IMPACT This framework could improve the performance and diversity of recommender systems, particularly for less popular items.