I was at a company that tried "sentiment analysis" 10 or 15 years ago. It was impossible to get right. The results were useless. It's funny how nothing seems to have changed. There's always some executive who's heard of "sentiment" and thinks they can spend a week shoving a bunch of text into a database, and then have it tell you whether the words are perceived as good or bad.<p>- I typed "Chevrolet" and it says "mpg" and "venerable" have negative sentiment, while "underachieving", "supposedly", "kilometers", and "cars" have positive sentiment. "Stunning" and "awkward" are both neutral.<p>- I tried "Mazda" and it was only slightly better. "Praise" is neutral, "regret" is positive, and "touring" is negative.<p>- I tried "Petzl" and only shows 16 words, all neutral, and that includes stop words like "had" and "and" and "etc".<p>Those are at least companies with unique names. For companies with less common names, you might be lucky to get keywords related to the right company at all.<p>I think there can be good uses for word clouds, but they are few and far between, and this isn't one. Just make 3 lists, side by side, and title them "positive", "neutral", and "negative". Instead of font size, put the more common words higher on their respective list. The only reason I can see to use a word cloud here is to hide how bad the analysis is.