Discipline: Mathematics and Statistics
Subcategory: Social Sciences/Psychology/Economics
Session: 2
Room: Exhibit Hall A
Micah Williams - North Carolina A&T State University
Co-Author(s): Jesus Brito, North Carolina A&T State University, Greensboro, NC
This paper addresses the topic of Binary Text Emotion Classification, within Text Sentiment Analysis, in movie (imbd.com), product (amazon.com) and destination (yelp.com) reviews. These reviews consist of multiple sentences, in which the polarity of the sentence is determined as positive or negative. In this paper, we take a Machine Learning based approach in order to modify the already written code to improve the given results including the accuracy, as the most important, as well as implement a number of different performance assessment measures. These modifications include implementing K-Fold Cross Validation, vectorizer changes, and a series of performance assessment measures including sensitivity, specificity, error rate and others. Along with implementing these modifications, we also fit the code to run with multiple different classifiers in order to determine which classifier method produced the most accurate and efficient results. The results yielded from our experiment showed that while, our modifications did not vastly outperform the originally written code, it did however produce a slightly greater accuracy with a .01 difference in favor of our method. Aside from the accuracy, our chosen method produced at a slower rate with a difference of 0.1381 seconds in favor of the originally written code. With these results considered, some future suggestions for future research and modification can be to: Test k-fold cross validation method against random splits, evaluate performance assessment measures for larger k values in k-fold cross validation, expand this approach to multiclass classification methods for Text Emotion Classification, and explore more vectorizing methods. This research was funded by the National Science Foundation (NSF) with the IRES REU program through North Carolina Agricultural and Technical State University Department of Mathematics.
Funder Acknowledgement(s): National Science Foundation (IRES REU Program)
Faculty Advisor: Dr. Guoqing Tang, tang@ncat.edu
Role: I did the vecotriziation changes as well as implementing the performance measures to test our code accuracy and other elements.