In today’s online environment, large textual data are generated and created daily, which reflects customer opinions and intentions. Discovering the repurchase behavior of customers helps companies to build engagement, boost competitive advantages and to make gains. In our research we used the deep learning regression models to extract knowledge from this data and predict the repurchase score, by analyzing the customers’ comments wrote in Moroccan dialectal Arabic. Using natural language processing advanced techniques (NLP), the model could analyze the textual comments to evaluate the repurchase score. This method helps companies to transform unstructured text data into useful knowledge, it gives a clear understanding of customers’ behavior, which enables the companies to make better decisions, solve problems, boost customer engagement and improve their growth.