Abstract
Sentiment analysis is a field of immense possibilities and application despite being an age-old topic. Various applications of machine learning and natural language processing keep contributing to this field with innovative techniques. Variants of neural networks with attention mechanism is a well-known tool in this field. However, very few of these techniques have been applied for Bengali sentences written with English letters which is a very common scenario for this era of social networking and online e-commerce sites. As for product reviews, it is of utmost importance as it helps the companies to suggest and review their products. In social networks, it may be used to analyze the emotion of the users. Moreover, in the online platforms of Bangladesh, most people use English letters to express their reviews in the Bengali language, which becomes a major issue for further analysis to guide them. In this paper, we propose a novel attention-based CNN model to solve the problem and analyze the performances of its variants along with the least features of NLP, thus making it able to work on different platforms.