Moroccan News Dataset and Source Code for Arabic and French Topic Detection

This dataset contains news articles collected from 37 Moroccan press sources (21 Arabic and 16 French) via RSS feeds over a period of 8 months, from June 14, 2025 to February 14, 2026. The dataset comprises 729 Arabic files and 729 French files, corresponding to one file per system update (3 updates per day). After deduplication, the dataset contains 115145 unique Arabic articles and 55239 French articles. Each file is stored in Python Pickle format and contains a list of dictionaries, where each dictionary represents a news article with the following fields : title, content, source, url, and date. A Python script for reading and preprocessing the data is provided alongside the dataset. This dataset was created to support a study on automated news topic detection. Along side the dataset, we provide the source code of the of the proposed monitoring pipeline : data collection, preprocessing, transformer-based (BERT/AraBERT) contextual embedding, topic detection using Graph Community Analysis, and treemap and graph visualizations. To simplify deployment, the code comes with transformer models and NLP tools, allowing for immediate use on the provided data. Detailed instructions on how to set up the environment and run the topic detection pipeline are provided in the README file.