Document Type : Research Paper
Electrical and Computer Engineering Department, Tabriz University, Tabriz, Iran
Computer Engineering Department, Payame-Noor University, Tehran, Iran
In this article, we propose a hybrid semi-supervised boosting algorithm to sentiment analysis. Semi-supervised learning is a learning task from a limited amount of labeled data and plenty of unlabeled data which is the case in our used dataset. The proposed approach employs the classifier predictions along with the similarity information to assign label to unlabeled examples. We propose a hybrid model based on the agreement among different constructed classification model based on the boosting framework to assign a final label to unlabeled data. The proposed approach employs several different similarity measurements in its loss function to show the role of the similarity function. We further address the main preprocessing steps in the used dataset. Our experimental results on real-world microblog data from a commercial website show that the proposed approach can effectively exploit information from the unlabeled data and significantly improves the classification performance.