The healthcare sector has seen recent advancements in sentiment analysis, but there is a lack of research focusing on sentiment related to depression treatments. Additionally, there is a shortage of methods capable of analyzing the positive, neutral, or negative sentiment associated with specific aspects in patient-generated content, such as social media posts, forum discussions, or video comments. To address this disparity, we propose an ontology driven sentiment analysis framework that integrates a domain specific ontology with various models (machine learning and deep learning) to determine the most suitable classifier for sentiment detection. Our approach utilizes a sentiment ontology categorizing emotions into three categories: positive emotions such as gratitude and happiness, neutral emotions such as indifference and surprise, and negative emotions such as anxiety and despair. Several machine learning and deep learning models were evaluated to find the best-performed method, including Support Vector Machines (SVM) Random Forest (RF) Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and their combinations with Bag of Words (BoW) and BERT-based embeddings. The framework was validated on a manually labeled dataset collected from various platforms related to depression treatments. The experimental results demonstrate that Random Forest combined with Bag of Words (RF + BoW) achieved the best overall performance, followed closely by SVM + BoW. Among the deep learning models, LSTM combined with BERT-based embeddings outperformed others, while CNN combined with BoW showed competitive results. These findings underline that integrating machine learning and deep learning approaches togethers with structured ontologies significantly enhances sentiment analysis performance.