06RanText & sentiment

Flipkart review sentiment

I lower-cased 2,304 product reviews, stripped punctuation and stop words and stemmed what was left. Then I scored each review with VADER, summed the positive, negative and neutral scores, and called the largest sum the overall mood.

Original tutorial by Aman Kharwal: Flipkart Reviews Sentiment Analysis (opens in a new tab)

What it printed in 2022

Numbers read from the saved notebook outputs. Cell numbers count every cell in the notebook, Markdown included, starting at 1.
Output	Value	Cell #
Summed VADER positive	923.55	24
Summed VADER negative	96.78	24
Summed VADER neutral	1,284	24
Overall verdict	Neutral	22

Cell # counts every cell from the top of the notebook, Markdown included, starting at 1.

'Neutral' won because VADER's neutral share is high for almost any short text. Stemming made it worse: 'awesome' became 'awesom', which VADER does not know, so it scored as fully neutral.

The data

Flipkart product reviews with star ratings from the tutorial author's GitHub data repository.

What I'd do differently

Run VADER on the raw text. It relies on capitals, punctuation and exact words, which the cleaning step removed.
Use the star ratings that come with the reviews to check whether VADER agrees with the customers.
Count reviews per label instead of summing scores.