Sentiment Analysis Experiment

I found the article about sentiment analysis.

http://www.hausmanmarketingletter.com/sad-state-sentiment-analysis/

and this statement:

"Recent experiments suggest sentiment analysis data is LESS accurate than a coin toss (accuracy 50%). That’s really scary if your brand makes strategic decisions based on sentiment analysis."

made me curious. I decided to quickly run some tests on my own.

Test dataset

I used IMDB reviews from Stanford library. https://ai.stanford.edu/~amaas/data/sentiment/

Tech stack:

• Java 11
• Spring Framework (just for dependency injection)
• Stanford Core NLP

Results:

Algorithm was right in ~50% cases, so the statement from the article was right 😉 Maybe I will try to improve this scores in the future.

How to load different data?

Check DataReader.java class. You can setup your own directories to positivie and negative reviews.

Contribution

If you think I made something wrong or you know how I can improve the results let me know or make a PR :)