A multimodal AI approach to analysing gender in Maltese online news

Abstract

The representation of gender in online news serves as both a mirror of and an influence on societal norms, necessitating rigorous, data‐driven scrutiny. This study employs a multimodal AI framework—combining computer vision and natural language processing—to examine gender portrayal in Maltese online news across articles and images. Vision models (YOLOv11, YOLOv12, Roboflow 3.0) detect and classify individuals, yielding mAP@50 scores up to 61.2% (YOLOv12) and balanced precision–recall performance, while NLP techniques (Deepseek vs bespoke NER) quantify text mentions and sentiment, evaluated via MAE, MSE and RMSE. Key findings reveal that male subjects account for 67% of person‐mentions (female 23%, unknown 10%), and that Deepseek halves the MAE of a generic NER pipeline (0.35 vs 0.87 overall), evidencing the value of gender‐aware modelling. These insights are made actionable through a Flask‐based web application, which ingests article texts or images to deliver automated, real‐time analyses of for both. By integrating quantitative metrics with practical tooling, this work not only uncovers entrenched biases across news beats but also equips journalists, regulators and the public with transparent, scalable methods for monitoring and redressing gender imbalance in media coverage.