Enhancing Sentiment Analysis Through Deep Learning: Leveraging Emotionnet with Improved Mutual Information for Advanced Sentiment Classification
Year: 2024
Type: article
Abstract: Sentiment analysis is a dynamic implement for learning understandings from textual data across miscellaneous fields, with social media, customer evaluations, and market trends. This research proposal announces EmotionNet, an advanced deep learning background designed to improve the accuracy and efficiency of sentiment analysis. In the preprocessing stage, advanced text cleaning methods are employed, leveraging transformer-based models to standardize and eliminate noise from the text. This involves homogenizing non-standard text, such as contractions and slang, while eliminating irrelevant symbols. Tokenization and lemmatization further refine the text, contravening it into meaningful units and an improved analysis of minimizing words to their base forms. Word embeddings like Word2Vec capture semantic information. Feature abstraction exaggerates text analysis by extracting emotion and contextual embeddings, incorporating syntactic features like dependency parse trees, and proposing novel semantic features based on frequency-inverse document frequency (TF-IDF) and pointwise mutual information (PMI) to capture word co-occurrence patterns. Sentiment lexicons and traditional textual features are also integrated. To augment efficiency, Feature Selection employs advanced techniques, including an improved mutual information (IMI) approach to select the most informative features for sentiment analysis. The classification of sentiment is an innovative architecture of deep learning, combining of convolutional neural network (CNN), and attention-recurrent neural network (A-RNN). An architecture excels were capturing both local and contextual sentiment. Emotion Net represents a significant advancement in sentiment analysis, offering superior accuracy and contextual understanding without compromising efficiency. The proposed model is implemented using Python. The model's performance is notable, boasting an impressive accuracy rate of 95.7% in experiments and tests.
Source: Journal of Electrical Systems
Author: Bikshapathy Peruka
Institution: Jawaharlal Nehru Technological University, Hyderabad
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7216647542
