A novel framework based on bi-objective optimization and LAN²FIS for Twitter sentiment analysis

Nowadays, opinion mining and sentiment analysis are the hot topics of research in social networks. Twitter is a popular social media that has general way of expressing opinions and interacting with other users in the online. Since the topics in Twitter are very diverse, it is difficult to collect data in the domain of sentiment classification. In this paper, a novel framework has been proposed which preprocesses information to enrich the tweet. Once the tweets are processed, a various features are extracted from the tweets. In this paper, we propose hybrid machine learning algorithms called LAN²FIS (logistic adaptive network based on neuro-fuzzy inference system) to utilize this huge information. The major challenge encountered in machine learning method-based sentiment classification is the abundant amount of data available. This amount makes the system complex and difficult to train the learning algorithms in a feasible time and degrades the system model in terms of classification accuracy. Hence, feature selection becomes an essential task in proposing robust and efficient classification models while increasing the accuracy. This paper presents a bi-objective optimization (minimum redundancy and maximum relevancy) for feature selection, which finds the more efficient feature subsets can be obtained. We performed our experiments using public tweets. Our overall work is implemented in a parallel and distributed way using the Hadoop framework with the MongoDB database to solve the problem of the computation time of the analysis when the dataset of the tweets is very large. Finally, we evaluate the following the performance in terms of accuracy, precision, recall, F-measure and error rate. Experimental evaluations show that our H-MLA is more efficient and has higher accuracy than that of other classifiers.