The product review team analyzed e-commerce product reviews to find out what customers like and dislike the most about various products. They got tens of thousands of product reviews from Kaggle. Such a project would help a company to "find what customers like and dislike from product reviews and guide [its] product lines to increase customer satisfaction". Their project, like several others here, fell into sentiment analysis area. They extracted the most common n-grams from the reviews, and determined whether they were associated with negative or positive reviews.
Their analysis was somewhat distorted by the dataset being unbalanced: about 82% of the reviews were positive. Still they were able to find clear correlations between what n-grams (words that often occur together) are correlated with positive versus negative sentiments. Many of them seem to be logical. For example, negative n-grams were 'on the model', 'in the picture' and 'wanted to love this'. The first two are often a precursor to stating that the real thing didn't look anywhere near to what it looked on the model or in the picture. A more puzzling correlation was that 'The fabric is' indicated a negative review, while 'The material is' a positive one!
This was one of the projects presented at the Women in Tech Machine Learning Product hackathon that took place July 28 - 29, 2018 in Austin, TX at Capital Factory.