Research Article
An Empirical Study on News Recommendation in Multiple Domain Settings
@INPROCEEDINGS{10.1007/978-3-030-94822-1_34, author={Shuichiro Haruta and Mori Kurokawa}, title={An Empirical Study on News Recommendation in Multiple Domain Settings}, proceedings={Mobile and Ubiquitous Systems: Computing, Networking and Services. 18th EAI International Conference, MobiQuitous 2021, Virtual Event, November 8-11, 2021, Proceedings}, proceedings_a={MOBIQUITOUS}, year={2022}, month={2}, keywords={Recommender system News recommendation Deep learning}, doi={10.1007/978-3-030-94822-1_34} }- Shuichiro Haruta
Mori Kurokawa
Year: 2022
An Empirical Study on News Recommendation in Multiple Domain Settings
MOBIQUITOUS
Springer
DOI: 10.1007/978-3-030-94822-1_34
Abstract
News recommendations using deep neural networks have been a hot research topic. However, most studies on news recommendations are based on the single domain setting. In this paper, we propose a news recommendation framework that uses freezing parameters and fine-tuning techniques for multiple domain settings. Since the model learned by data from multiple news platforms enables the representation of news articles to be much more robust, freezing the parameters of the news encoder effectively works in this setting. Moreover, the characteristics of domain-specific users are captured by fine-tuning the model on each domain data. Our empirical results with a real-world dataset demonstrate that using multiple domain data in the news recommendation results in a better performance. Despite its simplicity, the proposed framework works well, especially for domains where the number of data points is small. This framework has an AUC improvement of about 10% compared with the single domain setting.
