Opinion mining is a crucial research area in data science that focuses on the automatic analysis of textual data (such as reviews, customer comments, and feedback) to identify and evaluate sentiments expressed. In industrial contexts, it plays a vital role in enhancing customer feedback analysis, optimizing predictive maintenance, and improving process efficiency. In this research, we propose an innovative opinion mining method that combines Natural Language Processing (NLP), domain ontologies, and Machine Learning (ML) techniques to assist managers in making informed decisions about their products and services. Our method begins by constructing domain-specific ontologies, which play a key role in aspect detection within user comments. We then collect a raw dataset from social media platforms, such as YouTube and Facebook, focusing on four domains: Tunisian restaurants, smartphones, the 2019 Tunisian elections, and Tunisian TV programs. This dataset undergoes multiple pre-processing steps, including transliteration, stop-word removal, and vectorization, to prepare it for analysis. To classify sentiments, we employ two advanced deep learning models (i.e., LSTM and Bi-GRU), which are applied to analyze both the overall sentiment of each comment and the sentiment associated with specific aspects. This dual-layer analysis provides a detailed understanding of user opinions, identifying whether sentiments are positive, negative, or neutral at both the comment and aspect levels. This work not only advances the field of opinion mining through the integration of ontologies and deep learning but also offers valuable practical implications for decision-making in a wide range of industrial applications.