A CNN-BiLSTM based deep learning model to sentiment analysis

Analyzing sentiment from texts constitutes a crucial domain within Natural Language Processing (NLP) and artificial intelligence, given its substantial importance across various social, commercial, and cultural aspects of life. This analysis, viewed collectively by society, facilitates a better understanding of individuals’ opinions and extracts valuable information, contributing to enhanced decision-making and diverse policymaking. The capability of sentiment analysis in automated systems allows for more accurate and prompt responses to feedback and user opinions. This powerful tool empowers organizations to improve customer satisfaction, manage feedback trends in social networks, and overall, streamline international communications. In this study, utilizing the reputable Semeval2017 Task 4 dataset, we delve into sentiment analysis on texts and suggest a hybrid approach comprising Embedding, CNN and BiLSTM. Additionally, we manage the imbalanced problem through data augmentation techniques. The effectiveness of the proposed approach is assessed utilizing various metrics, achieving an accuracy of 84%, showcasing acceptable improvement compared to similar prior works. This research not only contributes to enhancing model performance in sentiment analysis but also serves as an effective proposition in the field of artificial intelligence for analogous issues.