DisenReason: Behavior Disentanglement and Latent Reasoning for Shared-Account Sequential Recommendation

Researchers have developed DisenReason, a new AI method for improving recommendations on shared accounts (like streaming services) by better identifying multiple users behind one account. The two-stage approach combines behavior analysis and latent reasoning to achieve up to 12.56% improvement in recommendation accuracy over existing methods.

• →DisenReason addresses the challenge of shared-account sequential recommendation where multiple users share streaming or e-commerce accounts.
• →The method uses a two-stage approach combining behavior disentanglement and latent user reasoning to identify the number of users behind an account.
• →Testing on four benchmark datasets showed consistent outperformance of existing methods with up to 12.56% improvement in MRR@5 metrics.
• →The approach shifts focus from inferring user preferences to inferring the actual users behind shared accounts.
• →The frequency-domain perspective creates unified account behavior representation for better recommendation accuracy.