The prevalence of anomalous behavior on social media platforms like YouTube disrupts user experience and undermines trust. Identifying these patterns can be challenging, particularly when analyzing commenter activity on YouTube channels. We propose a semi-supervised model called Dual Similarity-based Propagation Scoring (DSPS) to identify channels with anomalous behavior on YouTube by leveraging association between channels across a shared co-commenter network. The co-commenter network represents interactions between users who comment together, providing a magnifying lens into the connectivity between users who interact with videos across channels. We evaluated our approach and model on data from 97 channels with over 640,000 videos, 12 million commenters, and 123 million comments. Additionally, we created synthetic data representing real-world situations for a rigorous evaluation. The model achieved an F1 score of 0.92 on the synthetic graph. The findings show that analyzing the affinities between suspended and active channels can reveal previously unidentified features useful for characterizing anomalous behavior. This research contributes to the broader effort to ensure genuine and reliable user interaction on digital platforms.