Abstract:

Abstract Twitter is a widely used social media platform that is regarded as a crucial information source for gathering opinions, attitudes, reactions, and emotions from individuals. Therefore, the Twitter Sentiment Analysis (TSA) is developed for deciding the whether the textual tweets express a positive or negative opinion. The abundance of slang phrases and poor spellings in short sentence formats make it challenging to analyze Twitter data, nevertheless. In this paper, Hybrid Feature Extraction (HFE) is proposed along with the deep learning classifier to improve the classification. The HFE is the combination of Bag of Word (BoW) and FastText Word Embedding (FTWE) techniques that are used to extract the syntactic information and semantic information-related features from the tweets. The deep learning classifier namely Long Short-Term Memory (LSTM) with Softmax Regression Model (SRM) is used to classify the tweets as positive and negative. The datasets used to analyze the proposed HFE-LSTM-SRM method are Twitter and Sentiment140 datasets. The HFE-LSTM-SRM is analyzed by means of accuracy, precision, recall, F1-measure, and average computational time. The HFE-LSTM-SRM is evaluated using current techniques like Robustly Optimized Bidirectional Encoder Representations from Transformers (ROBERT-LSTM) and Spider-Monkey-Optimizer with K-Means Algorithm (SMOK). HFE-LSTM-SRM is more accurate than ROBERT-LSTM for the Sentiment140 dataset at 98.87%.

Keywords: