Using Machine Learning for Sentiment and Social Influence Analysis in Text

Students' academic achievement is largely driven by their social phenomena, which is shaped by social influence and opinion dynamics. In this paper, we employed a machine learning technique to detect social influence and sentiment in text-based students' life stories. The life stories were first pre-processed and clustered using k-means with euclidean distance. After that, we identified domestic, peer and school staff as the main influences on students' academic development. The various influences were used as class labels for supervised classification using SMO, MNB and J48 decision tree classifiers. In addition, the stories were manually labelled with positive and negative sentiments. We employed 10-folds cross-validation in classifying the sentiments and the social influences in the story corpus. The result show that peer influence is more salient on students' academic development followed by staff (15%) and domestic influences (12%). However, the remaining 54% of the stories contains unrelated social and other influences. Also, Students expressed more negative sentiment towards academic engagement than the positive sentiments. As per the classifier performance, SMO was found to be superior over MNB and J48 in the sentiment classification while MNB also performed slightly better than the SMO and J48 in the social influence analysis.