Violence occurs all around the developing and developed countries that harm people and damaging properties. Ranges of violence like physical, sexual, emotional, economical and traditional harm can occur in everyday life and people easily share their experiences on social media. So immediate analysis and categorization of violence text from social media can be useful to safeguard people. We proposed a novel violence text detection (Vio) model named VioEMBiL with BERT embedding for contextual findings (EM), TF-IDF for relevance and important terms and the BiLSTM classifier (BiL) for considering long dependency of text. The novel VioEMBiL model evaluates with English language dataset and succeeds with 97 F1 score and 97% accuracy. The proposed model presents achievable performance and that analyze with misclassification errors, contextual embedding models, deep learning, and traditional machine learning classifiers.