Breaking the Information Silo: Semantic Personas for Cross-Domain Recommendation

Researchers introduce SPHERE, a semantic-based system that enables recommendation knowledge transfer across completely separate digital platforms without requiring shared users or items. Using large language models to create behavioral semantic personas, the approach demonstrates consistent improvements over traditional recommendation algorithms across Amazon Books, Goodreads, and Steam, suggesting a new paradigm for breaking down information silos in cross-domain systems.

SPHERE addresses a fundamental limitation in modern digital ecosystems: recommendation systems operate in isolation, preventing platforms from leveraging behavioral insights across domains. Traditional cross-domain recommenders fail when platforms have no shared users or similar graph structures—a common scenario among independent services. This research reframes the problem by using language models to extract behavioral patterns, creating semantic personas that transcend platform boundaries. Rather than matching users by identity, SPHERE identifies behaviorally similar communities and transfers knowledge through semantic alignment.

The technical approach combines semantic signals from LLM-generated personas with collaborative filtering through a dual-tower architecture with dynamic fusion gating. This modularity enables integration with existing recommendation backbones like collaborative filtering, making adoption practical without full system overhauls. Empirical testing across three distinct domains—book retail, book discovery, and gaming—shows consistent improvements over NCF, SVD++, and LightGCN baselines.

The findings reveal nuanced insights: semantic proximity between domains doesn't solely determine transfer effectiveness. Instead, structural factors like target domain density and native predictive strength prove critical. This challenges assumptions that similar content domains naturally enable better knowledge transfer. For developers and platform operators, SPHERE offers a practical mechanism to enhance personalization without requiring federated data sharing or user account linking, addressing privacy and competition concerns.

The research signals emerging opportunities in AI-driven cross-platform analytics. As platforms increasingly recognize the business value of behavioral understanding while respecting privacy constraints, semantic approaches could drive the next generation of recommendation systems. This work demonstrates that language models can bridge domain gaps through abstraction rather than identity matching.

• →SPHERE enables cross-domain recommendations without shared users, items, or similar graph structures using LLM-generated semantic personas
• →Semantic proximity between domains is less important than structural density and native predictive strength of the target domain
• →The dual-tower architecture with dynamic fusion maintains interpretability and modularity, allowing integration with existing recommender systems
• →Testing across Amazon Books, Goodreads, and Steam demonstrates consistent performance improvements over baseline algorithms
• →Behavioral semantic alignment offers platforms a privacy-preserving alternative to user account linking for cross-domain personalization