SWATCS65: Sentiment Classification Using an Ensemble of Class Projects

Abstract

This paper presents the SWATCS65 ensemble classifier used to identify the sentiment of tweets. The classifier was trained and tested using data provided by Semeval-2015, Task 10, subtask B with the goal to label the sentiment of an entire tweet. The ensemble was constructed from 26 classifiers, each written by a group of one to three undergraduate students in the Fall 2014 offering of a natural language processing course at Swarthmore College. Each of the classifiers was designed independently, though much of the early structure was provided by in-class lab assignments. There was high variability in the final performance of each of these classifiers, which were combined using a weighted voting scheme with weights correlated with performance using 5-fold cross-validation on the provided training data. The system performed very well, achieving an F1 score of 61.89.