MULABS: Multi-Task Learning with Attention-Based Scoring for Click-Through Rate Prediction on Sparse Data in Healthcare Real-World Scenarios
Tu Tu, Jing Zhang, Yue Wang, Hui Jin, Megan Weihong He · 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · 2022
Personalized recommender systems have found wide applications in healthcare since it has the potential to inform individual users about the most needed medical resources. However, most recommender systems levering deep learning techniques demand a rich amount of data to learn patterns. In this paper, we aim to improve the performance of a recommender system when data are sparse and in low volume. In particular, we introduce a multi-task learning approach to alleviate the learning pressure and propose an attention-based scheme to boost the model performance by fully utilizing users’ historical data. In the experiments, we develop a novel dataset from a public dataset to mimic the real-world scenarios in healthcare. The results demonstrate that this multi-task learning with attention-based scoring (MULABS) aided recommender system is more accurate in predicting user preference with an AUC of 0.7541, requiring limited computation and memory resources.