ICAIMT: Twitter Sentiment Analysis Using Ensemble of Sentic Computing and a Novel Clustering Optimisation Algorithm Based on Combined Proximity Measures

Sentiment analysis based on Twitter data, known as Twitter sentiment analysis (TSA), has been widely explored in recent years. One of the key aspects of such analysis is the use of text representation techniques. These techniques are used to extract features from unstructured text. For more realist text representation, Sentic computing methods have been introduced in the literature to consider the underlying semantic relationships between words. This technique contrasts with the “Bag of words” that just considers occurrences of words. The ensemble of Sentic computing and supervised learning to perform TSA was investigated in the literature. However, these techniques require labelled data for training. This study investigated an unsupervised framework based on the ensemble (through delegation) of Sentic computing and clustering. A novel clustering optimisation algorithm is proposed using combined proximity measures to optimise Tweets’ clusters. The proposed framework eliminates the need for labelled data which are required in the case of supervised learning. Four Twitter datasets are used to evaluate the proposed framework and proposed novel clustering optimisation algorithm. The area under the curve has been considered for evaluation. The experiments elucidate that the proposed ensemble framework with integration of novel clustering optimisation algorithm could be a better choice for TSA when compared to a traditional supervised setup.