[Research News] New Deep Learning Framework Solves the Cold-Start Problem ｜Doshisha University

A new cross-domain recommendation framework leverages both positive and negative feedback to accurately predict the preferences of ‘cold-start’ users

Associate Professor Keiko Ono and her team have introduced a cross-domain recommendation framework designed to improve prediction accuracy for users with little or no activity on a target platform. The approach separately models high and low ratings and then adaptively integrates them using a gating network, resulting in substantially lower prediction errors. Experiments on Amazon review datasets demonstrated strong performance, showing that the framework enables more precise and flexible transfer of user preferences than existing methods.

Reference

Shimizu, Y., Ono, K., and Futagami, T. (2026). DUPGT-CDR: Deep User Preference Gating Transfer for Cross-Domain Recommendation, IEEE Access　(Volume 14).

DOI 10.1109/ACCESS.2026.3664871

For more details, please see the website of Organization for Research Initiatives and Development, Doshisha University. 
https://research.doshisha.ac.jp/news/news-detail-93/

This achievement has also been featured in the “EurekAlert!.”

https://www.eurekalert.org/news-releases/1119880

Title: Enhancing CDR with adaptive fusion of positive and negative feedback

Caption: The Deep User Preference Gating Transfer for Cross-Domain Recommendation (DUPGT-CDR) framework improves CDR by separately encoding high and low user feedback from the source domain and adaptively fusing these signals using a gating network. This mechanism enables more accurate transfer of user preferences to the target domain, improving recommendation performance in cold-start scenarios and outperforming existing CDR methods across multiple real-world datasets.

Credits: Associate Professor Keiko Ono from Doshisha University, Japan 

Image link:
https://ieeexplore.ieee.org/document/11396649


Image license type: CC BY-NC-ND 4.0 

Usage restrictions: Credit must be given to the creator. Only noncommercial uses of the work are permitted. No derivatives or adaptations of the work are permitted.

Title: A new CDR framework leverages both positive and negative feedback, improving both convergence speed and final accuracy compared to existing models

Caption: Deep User Preference Gating Transfer for Cross-Domain Recommendation (DUPGT-CDR), a newly developed CDR model, extracts high- and low-rating interaction vectors from the source domain, generates corresponding transformation vectors, and adaptively fuses them via a gating network. This allows the framework to achieve lower prediction errors than existing models and can aid in building highly integrated personalization in commerce, entertainment, and education.

Credits: Associate Professor Keiko Ono from Doshisha University, Japan 

Image link:
https://ieeexplore.ieee.org/document/11396649

Image license type: CC BY-NC-ND 4.0 

Usage restrictions: Credit must be given to the creator. Only noncommercial uses of the work are permitted. No derivatives or adaptations of the work are permitted.