“Sentiment analysis has appeared as a powerful tool in the digital landscape, allowing us to determine individuals’ thoughts and emotions by analyzing their textual expressions. This capability has been effectively used across diverse domains such as business, governmental decisionmaking, and academic research. For instance, in the business industry, sentiment analysis plays a crucial role, providing valuable insights into customer perceptions of products. Furthermore, policymakers can utilize sentiment analysis to understand public sentiments regarding various policies, while researchers employ it to explore societal trends and attitudes. All in all, in the era of digital communication, sentiment analysis stands out as a valuable tool that has assisted decision-makers in making choices based on the collective voice of individuals across diverse
online platforms.
One of the fields where more applications of sentiment analysis can be found is Health. In this domain, sentiment analysis is an important tool for extracting valuable information from user opinions within healthcare systems. These opinions reflect various aspects, providing valuable information about mental health issues, emotions, or specific personality traits, among others. By analyzing emotions and personality traits in user opinions, this approach helps identify mental health issues such as depression or other conditions. It also highlights the sentiments experienced during medical treatments, offering an understanding of patients’ perspectives. Additionally, sentiment analysis can enable the evaluation of hospital quality based on patient feedback. Nonetheless, processing all these data is a challenging task, highlighting the importance of developing advanced techniques and tools to process and interpret this information automatically.
This thesis focuses on the development of new techniques, methods, and applications for improving some typical sentiment analysis tasks. The main research body of the thesis is composed of six chapters. The first proposes new resources for enhancing emotion recognition in Arabic. The second studies the capabilities of clinical and biomedical pre-trained language models for detecting emotions from patient reviews. The third proposes a framework for fine-tuning pretrained language models to detect personality traits. The fourth develops a flexible platform for detecting emotions from health narratives. The fifth proposes a stacked ensemble framework combining different classic machine learning and deep learning algorithms to predict personality traits, exploiting diverse lexical and semantic features as well as word embeddings. The last chapter proposes a method for computing the quality of healthcare services and ranking real hospitals based on this quality plus the patients’ opinions and their preferences.”