Attention-Emotion-Embedding BiLSTM-GRU network based sentiment analysis

Abstract

Internet-based resource, which includes social  forums, review sites, blogs, and networks generates enormous heaps of data in the form of the views, users' feelings, thoughts, and disagreements on numerous things, brands, politics and social events. This data can be found in the form of "user generated content." The feelings of people who express themselves online have a significant impact not just on readers but also on those who sell products and on politicians. It is necessary to assess and well-structure the digital evidence that comes through facebook, and sentiment analysis has attracted a lot of attention for this use. Sentiment analysis, or text organization, is used to categorize the conveyed mentality or feelings in numerous ways, such as affirmative, favorable, unpleasant, thumbs up, hand gesture, negative, etc. Sentimental evaluation is also referred to as a "thumbs up" or "thumbs down" rating. Within the realm of natural language processing, the problem of insufficiently labelled data presents a hurdle for sentiment analysis (NLP). And as a solution to this problem, the techniques of sentiment analysis and deep learning have been combined. This was done given that deep learning models successful because of their ability to automatically learn new information. Therefore, this paper integrated the deep learning approach for domain independent sentiment analysis. The paper presents anattention-basedEmotion-Embedding BiLSTM-GRU Network for sentiment analysis. The paper presents comparative training accuracy and loss analysis with four baseline models. The network shows an accuracy of 93% which is higher as compared to baseline models and also achieved predictive accuracy compared to cutting-edge models.

URN:NBN:sciencein.jist.2023.v11.563