An Intelligent Cross-Domain Recommendation Framework Using TF-IDF and Hybrid Fusion Scoring

Abstract

Cross-domain recommendation systems address the basic challenge of delivering meaningful item suggestions over heterogeneous product domains by influencing shared semantic representations as well as transfer learning principles. This research paper represents a unified cross-domain recommendation architecture that operates instantaneously over three distinct domains: Books, Movies, and Mobile Devices. The proposed research approach employs TF-IDF-based text embeddings created from a joint vocabulary of 5,000 features derived from 33,114 items, cosine similarity for semantic retrieval, and a domain-adaptive hybrid fusion scoring mechanism. An intelligent domain inference module automatically detects query intent and routes retrieval to target. The system is evaluated on three large-scale real-world datasets: Book-Crossing (271,360 books, 1,149,780 ratings), MovieLens (10,329 movies, 105,339 ratings), and Flipkart Mobiles (3,114 products). d or multi-domain modes accordingly. The hybrid scoring combines semantic similarity (α = 0.6) with domain-specific quality signals — collaborative ratings for books and movies (β = 0.4) and normalized popularity for mobiles (γ = 0.2). Results demonstrate effective cross-domain knowledge transfer, with the final scores reaching 2.240 for books, 2.041 for movies, and 0.558 for mobiles on representative queries.