Towards harnessing based learning algorithms for tweets sentiment analysis

In the digital era of information, the number of web users and the volume of data being generated and exchanged over the internet has drastically increased. Sentiment analysis and opinion mining are open research fields having various everyday life applications, such as Twitter. Twitter is flooded with opinions, emotions, views and discussions among different communities across the world. Therefore, the tweet is related to various domains, such as stock market, product opinion and political elections. In this study, a classification problem is considered having a given set of tweets with various features and diverse opinions. The goal is to obtain expressions of opinion representing a target feature and to categorize these tweets as positive, neutral, or negative sentiment using different machine learning techniques. The dataset sentiment140 (STS-Test) is preprocessed and very commonly used for research purposes. The studied machine learning classifiers are random Forest, libSVM, IBk (kNN), decision table, and AdaBoostMl. Several evaluation measures are used for evaluating the performance of studied classifier on the selected dataset. These measures include accuracy, precision, recall, and F-measure. The confusion matrix is also obtained as a result of each classifier. The obtained results revealed that the RF classifier obtained the best accuracy results whereas the AdaBoostMl classifier obtained the worst accuracy results when tested for STS-Test dataset.