The Facet-Aware Multi-Head Mixture-of-Experts Model (FAME) is a novel approach to sequential recommendation systems that aims to capture the multi-faceted nature of items and user preferences. Traditional sequential recommendation systems often use a single embedding vector to represent each item, which may not fully capture the diverse attributes of items, such as genres or actors in movies. FAME addresses this limitation by leveraging sub-embeddings from each head in the last multi-head attention layer to predict the next item separately. This approach allows the model to capture the potential multi-faceted nature of items without increasing model complexity. A gating mechanism is used to integrate recommendations from each head and dynamically determine their importance. Additionally, FAME introduces a Mixture-of-Experts (MoE) network in each attention head to disentangle various user preferences within each facet. Each expert within the MoE focuses on a specific preference, and a learnable router network computes the importance weight for each expert and aggregates them. Extensive experiments on public sequential recommendation datasets demonstrate the effectiveness of FAME over existing baseline models, highlighting its ability to capture complex user preferences and improve recommendation accuracy.
Multi-head attention, Mixture-of-Experts
Multi-head attention with Mixture-of-Experts
Public sequential recommendation datasets
Recommendation accuracy, User preference capture
Cloud-based, On-premises
Yes
Yes
Facet-aware recommendations, Dynamic gating mechanism, User preference disentanglement
Yes
Standard server hardware
Web, Mobile
Integrates with existing recommendation systems
Data encryption, User privacy protection
GDPR
N/A
No
Limited community support
Research institutions, AI companies
Large-scale datasets
Low latency for real-time recommendations
Optimized for efficiency
Facet-based explanations
User privacy, Bias in recommendations
Complexity in model training
Retail, Entertainment
Product recommendations, Content recommendations
E-commerce platforms, Streaming services
API integration, SDKs
Highly scalable
Vendor support, Professional services
Available
Web-based dashboard
Yes
Supports multiple languages
Subscription-based, Usage-based
Yes
AI technology partners
N/A
Compliant with industry standards
Latest
Software as a Service
Yes
RESTful API for integration
B2B, B2C
0.00
USD
Proprietary
01/01/1970
01/01/1970
N/A
Real-time analytics, Customizable recommendations
Yes