Technical Marketing Leader تحسين استراتيجيةالتسويق عبر جمع البيانات من الويب

Abstract

This study addresses the problem of sentiment analysis in texts written in the Algerian dialect, a linguistic variety characterized by complex features such as code-switching between Arabic, French, and English, in addition to the absence of standardized spelling and the scarcity of annotated datasets. The study relied on real data collected from YouTube using web scraping techniques, targeting users’ comments related to skincare products. Models representing the different approaches adopted in this study were evaluated, including traditional machine learning, deep learning, and transformer-based models. The results showed the superiority of the MARBERT model, which achieved an accuracy of 79% and an F1-Macro score of 0.72, followed by the SVM model with TF-IDF, which achieved an accuracy of 77%. Based on these results, an initial design of a marketing intelligence platform for Algerian enterprises was proposed, along with an economic feasibility study within the framework of a startup project.