Tweet Classification on the Base of Sentiments Using Deep Learning

Abstract

Twitter data has been used to improve political campaigns, product quality, and sentiment analysis over the last several years. An essential and collaborative effort for many businesses is to classify tweets based on user sentiment. A machine learning classifier is proposed in this research in order to aid in sentiment analysis for these types of organizations. The classifier employs a soft voting process based on Logistic Regression (LR) to determine the final prediction. Based on the content and tone of the tweets, we divided them into three categories: “good,” “negative,” and “neutral.” Additionally, the accuracy and F1-scores were used to assess the performance of several machine learning classifiers. Classification accuracy was also examined in terms of feature extraction strategies, such as term frequencies, Inverse Document Frequencies (TF-IDF), and Words-To-Vectors (W2V). Furthermore, the performance of the Deep Long-Term Memory (DLTM) network was evaluated on the dataset. Compared to other classifiers, the presented classifier performs better. With TF-IDF feature extraction, the LR can attain an accuracy of 0.9616 and an F1-score of 0.7633. According to these findings, ensemble classifiers outperform non-ensemble classifiers. According to experiments, using TF-IDF as a feature extraction approach improves the performance of machine learning classifiers. The extraction of W2V features is less efficient than the extraction of TF-IDF features.