A Clickbait Identification Method for Unbiased Recommendation.

This article focuses on CI-UR, a novel Clickbait Identification method for Unbiased Recommendation that integrates news de-duplication and clickbait detection to improve news recommendation accuracy. CI-UR addresses two key challenges in news recommendation systems: reducing exposure bias caused by clickbait and eliminating redundant similar news articles without relying on auxiliary information or explicit exposure mechanisms. The method employs a relevance function based on cosine similarity between news titles and content for de-duplication, a dynamic sampling strategy for user-news embedding, and an unbiased learning framework to mitigate false positive noise during training. Experimental results on two real-world datasets, TikTok and MIND, demonstrate that CI-UR outperforms state-of-the-art baselines in Recall and NDCG metrics, validating its effectiveness in providing more accurate and diverse personalized news recommendations.