Hybrid contrastive multi-scenario learning for multi-task sequential-dependence recommendation.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Multi-scenario and multi-task learning are vital for industrial recommendation systems.
• Conventional models struggle with cross-scenario information and multi-step task data sparsity.

Purpose of the Study:

• To propose a novel framework, Hybrid Contrastive Multi-scenario learning for Multi-task Sequential-dependence Recommendation (HCM²SR).
• To enhance recommendation quality across diverse scenarios and mitigate challenges in multi-step conversion tasks.

Main Methods:

• Hybrid contrastive learning in the scenario layer captures shared and specific information.
• A scenario-aware multi-gate network evaluates cross-scenario relevance.
• An adaptive multi-task network facilitates knowledge transfer across sequential stages.

Main Results:

• HCM²SR demonstrates significant effectiveness on public and industrial datasets.
• Ablation studies confirm the positive contribution of individual components within the framework.

Conclusions:

• HCM²SR offers an effective solution for multi-scenario, multi-task recommendation systems.
• The framework successfully addresses cross-scenario information leveraging and data sparsity in sequential tasks.