Text classification is widely used in tasks such as sentiment analysis, spam detection, and hate speech detection, but performance often suffers when models rely on only linguistic features or semantic embeddings. Single-representation methods struggle to capture both syntactic structure and deep semantic meaning, limiting robustness and generalization across tasks. This study proposes a hybrid feature fusion framework that combines interpretable linguistic features with semantic embeddings from Doc2Vec and transformer-based models. The approach is evaluated on five benchmark datasets covering fake news detection, Bloom’s taxonomy classification, and hate speech detection. Experiments with multiple classifiers show that the fused features consistently outperform single-feature baselines. The best results, using a BERT-based fusion approach, achieve accuracies of 81% for fake news detection, 67% for Bloom’s taxonomy classification, and 72% for hate speech detection, with improved precision, recall, and F1-score. Overall, the results demonstrate that integrating linguistic and semantic features provides a robust, domain-agnostic solution for improved text classification.