A Comparative Study: Best Machine Learning Algorithm for Social Media Sentiment Analysis
No Thumbnail Available
Date
2020
Journal Title
Journal ISSN
Volume Title
Publisher
Uva Wellassa University of Sri Lanka
Abstract
Sentiment analysis is a field of study that aims to derive the sentiment or the opinion of a
text using natural language processing techniques. Performing sentiment analysis on
Twitter data has a vast number of applications including predicting stock market prices,
product recommendations, etc. Sentiment analysis can be done in lexicon-based, machine
learning-based, or hybrid approaches. K Nearest Neighbor, Support Vector Machine,
Logistic Regression, Naïve Bayes, K Means Clustering, Decision Trees, and Random
Forest are the few most popular machine learning algorithms. This study aims to conduct
a comparative analysis among the usage of K Nearest Neighbor, Support Vector Machine,
Logistic Regression, and Multinomial Naïve Bayes machine learning algorithms
combined with sentword net lexicon to suggest which one provides the best accuracy in
sentiment classification of Twitter data. A data set of 1028 tweets was acquired using the
Twitter Standard Search API (Application Programming Interface) and Tweepy python
library. The name of a popular brand of mobile phones was used to search for tweets. 570
tweets remained after the duplication removal and cleaning process. Then the remaining
data was classified as positive, negative, or neutral using sentiword net lexicon and used
to train selected machine learning algorithms.80% of the data was used for training and
20% was used for testing. Word counts in the tweets were used as features. Multinomial
Naïve Bayes is proved to be the best machine learning algorithm with a model accuracy
of 74.56% and K Nearest Neighbor (k=3) is the worst-performing algorithm with an
accuracy of 54.38%. Logistic Regression and Support Vector Machine (linear kernel)
respectively had accuracies: 72.80% and 70.17%. The result of this research proves
Multinomial Naïve Bayes performs relatively better in Twitter sentiment analysis than K
Nearest Neighbor, Support Vector Machine, Logistic Regression. This is because two
basic assumptions for applying the Multinomial Naïve Bayes algorithm: feature
independency and multinomial distribution are well satisfied by the features selected for
this study. Also, Multinomial Naïve Bayes can perform well with high dimensional data
like tweet text. On the other hand, the poor performance of the K Nearest Neighbor is due
to the same reason. K Nearest Neighbor cannot handle a large number of features very
well.
Keywords: Sentiment analysis, Twitter, Hybrid approach, Machine learning algorithms,
Comparative analysis.
Description
Keywords
Computer Science, Social Media, Information Science, Computing and Information Management