Study of Sentiment Analysis Through a Machine Learning Perspective

Study of Sentiment Analysis Through a Machine Learning Perspective

When a person is scrolling through their social media feed, YouTube
feed, or simply on Google, they are shown a variety of websites,
posts, ads, and so on. If the information that is displayed is randomly
presented, it does not benefit the company presenting the data or the
user viewing the data. However if data is analyzed respectively to the
users to see what data users seem to like and what data users show
a disliking to, then both parties can benefit. The emotions of users
will be analyzed with respect to the posts or the information being
presented on the internet in an analysis known as Sentiment
analysis to display products or info that users like and will likely
purchase. Therefore, if companies rely on such analysis, appropriate
methods must be used to conduct such analyses. In this research,
different machine learning algorithms were analyzed and compared
to highlight the best ways to benefit internet users through sentiment
analysis. Multinomial Naive Bayes, Complement Naive Bayes,
Passive aggressive Classifier, Logistics Regression Classifier,
Support Vector Machine, and Decision Trees were the algorithms
that were analyzed in this study through three different ngrams.
Accuracy was the primary metric used for comparison, however
precision, F1, and recall were also used as comparison metrics. At
the culmination of the analysis, logistics regression with unigrams
was found to have the highest accuracy of 63.76% in sentiment
analysis

Narayanan, Sherlin Angel
Ali, Md L. Ph.D.

Department of Computer Science and Physics, Rider University

Poster

English