A Hybrid Deep Learning Approach for Sentiment and Emotion Analysis in Textual Data

Abstract

Sentiment analysis (SA) is an automated technique for detecting and understanding the emotions conveyed in text. In the recent ten years, SA has been increasingly popular in the field of natural language processing (NLP). When it comes to subjective judgments, SA can assist consumers transcend the haziness of human judgment and provide them with transparent, objective sentiment suggestions based on large datasets. Existing sentiment analysis systems could use some work to make them more accurate and robust. Marketing campaigns that rely on product reviews would benefit from a reliable way to forecast when opinions may vary. With the rise of several online platforms, social media analytics (SMA) has become an essential tool for organizations to understand customer sentiment and guide their advertising strategies. In addition, SA is used by researchers to address public sentiments on many matters. Using preexisting methods such as recurrent neural networks (RNNs) along with transformation models such as Bidirectional Encoder Representations (BERT), researchers may effectively assess the sentiment conveyed in literary works. These models are able to accurately determine the literary work's emotional tone because they learn intricate patterns and features particular to the setting. Sentiment analysis is an essential part of natural language processing since it helps us understand the feelings and thoughts conveyed in text. Word2Vec and Fast Text are two examples of word embeddings that are thoroughly examined in text for mapping that is closely related to real number vectors. However, deep learning and word embedding aren't perfect. Integrating word embedding with deep learning models is key to achieving high-performance sentiment recognition in natural language processing. Our BERT-BiGRU model, which employs NLP techniques, is part of a larger endeavor to enhance sentiment analysis. Several word embedding approaches are considered in the proposed model, which aggregates their attributes and uses them to classify texts based on their emotional tone. When compared to prior studies, the proposed model outperforms them in terms of sentiment classification.