Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 24·5 min read
Adapting Knowledge Graphs for Behavior Denoising in Sequential Recommendation
TL;DR
AdaptedKG enhances sequential recommendation systems by using knowledge graphs to filter out less informative user interactions. This method improves the accuracy of predictions by focusing on relevant historical data.
✦ Why It Matters
Implement AdaptedKG to enhance your recommendation systems by filtering out noise from user interaction data.
Key Takeaways
- Use AdaptedKG to improve recommendation accuracy.
- Focus on relational paths in user interactions.
- Calibrate interaction relevance with knowledge graphs.
- Evaluate performance gains in your existing models.
How It Works
AdaptedKG operates by first comparing a user's observed interactions with structurally similar alternatives to identify significant relational paths. This creates a localized view of the knowledge graph that highlights relevant connections.
Each interaction is then evaluated against these reference items to determine its support within the context, allowing for the adjustment of historical data representations. The retention coefficients derived from this process help in reweighting the target losses during training, ensuring that the model focuses on the most informative interactions.