ISSN: 2182-2069 (printed) / ISSN: 2182-2077 (online)
Hybrid Feature Extraction and Deep Learning Classifier Based Effective Classification for Twitter Sentiment Analysis
Twitter is a widely used social media platform recognized as a crucial source of information for gathering individuals’ opinions, attitudes, reactions and emotions. Therefore, the Twitter Sentiment Analysis (TSA) is developed to decide whether the textual tweets express positive or negative opinions. However, the abundance of slang phrases and poor spelling in short sentence formats make Twitter data difficult to analyze. This paper proposes Hybrid Feature Extraction (HFE) with a deep learning classifier to improve the classification. The HFE is a hybrid of the Bag of Word (BoW), and FastText Word Embedding (FTWE) approaches used to extract syntactic and semantic elements from tweets. Long Short-Term Memory (LSTM) and Softmax Regression Model (SRM) deep learning classifiers categorize tweets as positive or negative. The datasets used to analyze the proposed HFE-LSTM-SRM method are Twitter and Sentiment140 datasets. The HFE-LSTM-SRM is analyzed using Accuracy, Precision, Recall, F1 metrics and Average Computing Time. The existing methods like Spider-Monkey-Optimizer with K-Means Algorithm (SMOK) and Robustly Optimized Bidirectional Encoder Representations from Transformers with LSTM (ROBERT-LSTM) are used to evaluate the HFE-LSTM-SRM. The accuracy of HFE-LSTM-SRM for the Sentiment140 dataset is 98.87%, which is higher than that of ROBERT-LSTM.