STN-CDRS: Sentiment Transfer Network for Cross-Domain Recommendation Systems

• Published in: Computer assisted methods in engineering and science Vol. 31; no. 3
• Main Authors: Nikita Taneja, Hardeo Kumar Thakur
• Format: Journal Article
• Language: English
• Published: Institute of Fundamental Technological Research Polish Academy of Sciences 01.06.2024
• Subjects: cross-domain recommendations; deep learning; knowledge transfer; sentiment transfer network; user reviews
• ISSN: 2299-3649; 2956-5839
• DOI: 10.24423/cames.2024.900

In enterprise environments, the products may come from a variety of categories or domains. Users may engage with entities in one domain, but not in the others when they are presented with multiple domains. Such users are referred to as “cold-starters” in other domains. The primary difficulty in cross-domain recommendation systems is to efficiently transfer user’s latent information based on their engagements in one domain into the other domains. The advancements in recommendation systems have inspired us to develop review-driven recommendation models that utilize cross-domain knowledge transfer and deep learning models. This work proposes a sentiment transfer network specifically designed for providing recommendation in cross-domain (STN-CDRS). The novelty of the work lies in the user rating enrichment mechanism, which is done by extracting latent information from user review data to fill sparse rating matrix. This enrichment uses previously developed RNN-Core method for efficiently learning user reviews. The reviews provided by the users are used to enrich sparse data across domains. This enrichment allows two things: alleviates the cold start problem and allows more intersecting users across domains to bridge the gap while learning. This work empirically demonstrates its efficiency by iteratively updating over the baseline recommendation models in terms of MAE (mean absolute error), RMSD (root mean squared deviation), precision and recall measures with other state-of-the-art-review-aided cross-domain recommendation systems.