Gender-based violence (GBV) is a pervasive social and public health issue that increasingly manifests in digital communication platforms. This article presents a multidimensional framework, the Gender Discourse Violence Index ( \(GDVI_{AI}\) ), designed to detect and quantify violent discourse in WhatsApp conversations. The framework integrates four key dimensions: (i) toxicity detection using large language model prompts, (ii) sentiment analysis with BERT to capture emotional load and polarity, (iii) a weighted dictionary of over 2200 offensive expressions, and (iv) grammatical person identification to assess the directness of threats. By combining these components in a weighted formula, the \(GDVI_{AI}\) produces a score ranging from 0.1 for non-violent discourse to values exceeding 9 for explicit insults or threats. The model was evaluated against a reference dataset using confusion matrices and descriptive statistics, demonstrating high accuracy and robustness. Beyond classification, the framework enables temporal analysis of message-level violence, supporting the identification of escalation patterns in perpetrator–survivor dialogues. The proposed approach contributes to forensic psychology and digital criminology by offering a reliable tool for early detection, evidence collection, and the study of communicative dynamics in cases of gender-based violence.
