Media coverage of gender-based violence plays a critical role in shaping public understanding and policy, yet often perpetuates stereotypes and biases. We present a hybrid AI approach to analyze how French-language media represent gender-based violence. Combining rulebased Natural Language Processing (NLP) with Large Language Models (LLMs), the system applies expert-defined criteria across analytical categories, with each criterion assigned to the most effective method based on empirical performance. This strategy achieves 87.1% overall accuracy, surpassing previous models. GPT-4 led general performance (77.9%), while NLP delivered exceptional results in structural and language-sensitive categories. Our findings demonstrate that combining complementary AI
techniques enables near-human accuracy in evaluating media narratives and contributes to advancing web-based text mining for socially relevant media analysis.
