BIG DATA SENTIMENT ANALYSIS FOR PREDICTING FOUR CATEGORIES OF EMOTIONS USING BILSTM ALGORITHM

Abstract

Sentimental analysis is a way of understanding human text expressions based on the lexical content. It is an important in natural language processing (NLP) and machine learning that allows researchers and marketers to analyse public sentiment and understand customer sentiment from textual data. This work uses Bidirectional Long Short-Term Memory (BiLSTM) neural networks along with Adam optimizer to classify records of big data (31017 samples) into three types of sentiments and also the unknown such are non-expressive texts in the social media. With an entire dataset from tweets, the preprocessing steps includes text cleaning, tokenization, and lemmatization to improve the model’s performance. With the goal of achieving high category prediction, the architecture includes embedding layers, BiLSTM layers for capturing long-term memory of the given dataset, and a dropout for regularization. Metrics for evaluating the models include, recall, accuracy, precision, F1 Score and receiver operating characteristic (ROC) curve. These metrics collectively establish how reliable sentimental prediction is at a given point in the research process. The validity of the tool lies in its capability to provide emotion categorization into positive, neutral and negative with superior accuracy and to meet the stressful conditions of emotion evaluation from the whole samples. By incorporating these advanced techniques, our system provides an actual-time sentiment analysis, offering advanced functionality for social media management like Twitter, and each other area studying emotional intelligence. The model also consists of an extensive evaluation for overall performance and model enhancement, with a view to ensuring transparency and reliability in sentimental classification results.