الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 155 |  |  |
| | | from | 155 | | |
Abstract
The growth of online media sites in recent years has led to the spread of content sharing like commercial advertisements, political news, celebrity news, and so on. Due to the widespread presence of social media, such as Facebook, Instagram, and Twitter, have been impacted by various reviews in our recent daily lifestyles, sentiment analysis has become an important field in pattern recognition and natural language processing (NLP). In this field, user feedback data on a particular issue is evaluated and analyzed. Therefore, detecting emotions within text is one of the important challenges of current NLP research. Emotions have been studied extensively in psychology and the behavioral sciences because they are an integral part of human nature. Feelings describe a mental state of distinct behaviors, feelings, thoughts, and experiences. Because of the easier access and rapid expansion of data through online media platforms, distinguishing between positive, and negative reviews or being fake and real data has become difficult. The massive volume of news transmitted over online media portals makes manual verification impractical, which has prompted the development and deployment of automated methods for identifying and detecting online reviews through sentiment analysis and machine learning approaches. The increased use of Artificial Intelligence (AI) has experienced significant growth in various smart applications, mainly employ machine learning (ML) algorithms, which have resulted in the construction of accurate decision-making models. This work provides a comprehensive overview of the prominent machine learning models applied in sentiment analysis. It explores the many classifications of sentiment analysis, as well as the limitations of prevailing deep learning architectures. The letter also reviews some of the contributions made previously to sentiment analysis with a focus on deep learning methodologies as well as the most popular data set. However, these algorithms still suffer from trustworthy interpretability and explainability of the models’ architectures and their outputs, known as black-box algorithms. This issue inspires the notion of eXplainable AI (XAI), which has been identified as a critical characteristic for the operational application of AI models. One of the main objectives of this study is to apply and compare common XAI methods for online reviews for explaining the model prediction. For sentiment analysis, we conducted a comparative study and experimental evaluation for the explainability of deep networks based on benchmark deep learning on sentiment 140 data.