Deep learning-based artificial intelligence applications have driven transformation across multiple industries, and recommendation systems based on deep learning have been widely adopted in the industry. For example, in clothing, dining, and TV show recommendations, recommendation systems can utilize big data to suggest items that users might like based on their behavioral data. They can also optimize the next recommendation results based on whether users accept or reject the recommendations. Multi-task learning refers to handling multiple tasks simultaneously during the modeling process, such as user clicks and user orders, or user likes and viewing duration. There are correlations among multiple tasks, and compared to training on a single task only, multi-task learning can significantly improve the effectiveness of each task. In this work, we propose a novel multi-task learning framework, the Independent Grouped Information Expert (IGIE) model. The IGIE model consists of two identical Multi-gate Mixture-of-Experts (MMoE) structures, with independent inputs for each part. The upper expert layer processes information based on different initialized embeddings, and after the two independent MMoE structures complete their information processing, the results are concatenated and passed to different task towers at the upper level.
A Wed, study studied this question.