191

1

Sentiment Analysis of Matt Levine's Money Stuff Newsletter

Since I don't read newspapers, I decide to use this Money Stuff newsletter by Matt Levine as the source of data:  http://www.bloombergview.com/articles/2016-01-08/chinese-markets-and-saudi-oil

I wanted to know how happy or sad certain stories are in that specific newsletter, which can be useful in exploring a person's bias (a person writing a negative article about banks implies that the person may not like banks). It's not perfect (the person may just be quoting a person who hates bank) but I think it's good enough.

Sentiment analysis itself can be rather subjective, and both humans and algorithms can disagree on whether an article is positive or negative. So I decided to use three different algorithms and plot their curve. All three algorithms are using Machine Learning, so you can pretty much think of them as three pundits giving their own opinions on Matt Levine.

Ruby's Sentimentalizer - https://github.com/malavbhavsar/sentimentalizer

Python NLTK - http://text-processing.com/demo/sentiment/

AlchemyAPI - http://www.alchemyapi.com/products/demo/alchemylanguage

 

8d821937

Key: -1 is unhappy, 0 is neutral, 1 is happy.

This is not a very interesting graph, and I plan on coming up with better visualizations in the future with the dataset...if I get any time. I am only turning this in because I may not get time in the future, and I might as well turn in something rather than nothing.

It's interesting to see the articles where the three different algorithms disagree, but probably more useful to see the articles where the algorithms do agree upon (Matt Levine seems to be negative when writing about China, for instance).

One thing that could be useful with sentiment analysis is understanding what specific words are seen as "positive" or "negative" within the article itself. For example, when using AlchemyAPI, I noticed that Matt Levine was negative towards the IEX, but positive towards IEX customers. This presents a more nunanced view of events than simply thinking Matt loves the IEX or is netural towards it. 

Another application of 'sentiment analysis' would be in anaylzing a single news story directly. Kurt Vonnegut Jr. came up with the idea of 'plot curves' (where you can analyze a story based on the trajectory of a story's mood), and recently, computer scientists have came up with a R package to calcuate plot curves. By drawing a "plot curve" of an individual story, you could see whether news authors themselves are inadventerly influencing the tone of an article by deciding where to place certain happy or sad events.

You can even see some unsophisicated plot curves of this newsletter if you are reading the graph from right to left, since the China article came first and the Bond Market story came last.

Comments

Please sign in or sign up to comment.