Deep Multi-interaction Hidden Interest Evolution Network for Click-Through Rate Prediction

Click-Through Rate (CTR) prediction plays a crucial role in the field of recommendation systems. Some previous works treat the user's historical behavior as a sequence to uncover the hidden interests behind it. However, these works often ignore the dependencies and dynamic interests between different user behaviors evolving over time, as well as hidden information by user representation. To solve the above problems, we propose Deep Multi-Interaction Hidden Interest Evolution Network (MIHIEN). Specifically, we first design Hidden Interest Extraction Layer (HIE) to initially mine the hidden interests of users evolving over time from it, which can better reflect the user representation. The deeper interests of users are then explored in two types of interactions in the Item-to-Item Sub-network (IISN) and the User-to-Item Sub-network (UISN), respectively. The experimental results show that our proposed MIHIEN model outperforms other previous mainstream models.