Valence-Enhanced Bidirectional Lstm Framework for Context-Aware Sentiment Analysis on Social Media Data

Abstract

Social media platforms like Twitter have now emerged as treasure houses of public opinion and provide deep insight into mass mood and behavioral patterns. This paper, therefore, proposes a systematic approach toward sentiment analysis using the Sentiment140 corpus, which contains over $\mathbf{1. 6}$ million labeled tweets in the form of positive or negative classes. A formalized Natural Language Understanding pipeline has been applied in this research, comprising data preprocessing, tokenization, and feature extraction via the Bag-of-Words model and CountVectorizer. Extended text normalization is done for the elimination of noise such as user mentions, hashtags, URLs, and punctuation to feed machine learning algorithms with goodquality input. Detection of the distribution and frequency of sentiment classes has been performed by exploratory data analysis and overcoming dataset imbalance by visualization and statistical testing. We proposed Valence-Enhanced Bidirectional LSTM framework for sentiment classification. This study contrasts typical classification algorithms based on their performance to detect sentiment polarity. Model performance was assessed through the standard metrics of accuracy, precision, recall, and F1-score with the aim of determining the most suitable algorithm to use for sentiment prediction at the tweet level. The proposed model offers a scalable and reproducible paradigm for opinion mining on social media with the development of computational linguistics and data-driven social research.