[Kaedon]

A Twitter bot called @AmazonArtReview that tries to find funny 1-star reviews of media on Amazon. It's using some crude machine learning and just a fun side project.

Finding humor's tough! I've read some papers on finding humor in reviews, e.g. https://cs224d.stanford.edu/reports/OliveiraLuke.pdf. That's been helpful. It'd be great if I could use the Yelp review set for finding humor as they did but the humor doesn't translate between data sets. Mostly it's still manual curation, which is ok too.

https://www.twitter.com/AmazonArtReview

[WA]

What about the number of upvotes for 1-star reviews?

[Kaedon]

Yeah, that's a good suggestion. I should start storing and using that info. My hypothesis is that humor is inversely proportional to the number f helpful votes. For example, 0 of 57 helpful is probably funnier than 3 of 3.