Optimization of CTR Prediction in Recommendation with Rating-Time Decay

Abstract

The Click-Through-Rate (CTR) prediction is a significant concern in the advertising industry, and this research aims to address the three open aspects of learning, feature, and bias that require attention in CTR. The research conducted a literature review and identified appropriate methods from previous research to tackle these aspects. The proposed model optimizes learning time, prevents over-fitting using an early stopping strategy, and handles bias recommendations using Rating-Time decay. Testing on user interest in new and unpopular items provides promising results, indicating that the user’s latest preferences align with the latest event for CTR. The study’s findings demonstrate that the proposed model resolves CTR and over-fitting issues and optimizes the learning aspect of CTR models.