This study develops a technique to identify radical opinions in hate group web forums by combining machine learning with semantic approaches. It utilizes four types of text features—syntactic, stylistic, content-specific, and lexicon features—along with three classification techniques: SVM, Naïve Bayes, and Adaboost. The approach is tested on postings from two hate group forums, showing promising preliminary results. Cross-validation demonstrates that the technique is stable and adaptable to different timeframes, making it useful for broader sentiment classification tasks.