Sentiment Analysis on Youtube Comments to Predict Youtube Video Like Proportions

Sammanfattning: Social media websites are some of the world’s most popular websites and allow all users to have a voice and express opinions and emotions. Using sentiment analysis, these users’ opinions and emotions can be extracted and quantified. This study examines sentiment analysis on Youtube comments and how well the number of comments classified as positive, neutral and negative can be useful in predicting the like proportion of a Youtube video. Four different prediction formulas were tested that utilize neutral comments in different ways. Five different classifiers were examined with pretraining on Youtube comments, tweets, and a combination of tweets and comments. Some positive correlation between the predicted and actual like proportion was found. The best performing configuration was a logistic regression classifier trained only on Youtube comments with a prediction that attributes all comments classified as positive and neutral to likes. However, the errors of this type of prediction are so large that it likely has little real-world application. Possible method improvements include filtering out spam comments and include emoji sentiment.