sentiment analysis with NLP

April 30 2017

Simple Sentiment Analysis with Clojure and the Standford’s NLP library.

The original java libraries and models of the standford-corenlp project can be found in:

[edu.stanford.nlp/stanford-corenlp "3.5.2"]
[edu.stanford.nlp/stanford-corenlp "3.5.2" :classifier "models"]

But using a Clojure wrapper make things slightly easier, and so the damionjunk/nlp dependency should be in your build file (Leiningen or Boot):

[damionjunk/nlp  "0.3.0"]

You can use the function sentiment-maps. When call, the sentiment-maps function returns a map of :sentiment :text where the sentiment key is assigned a value between 1 and 4, where 1 is very negative and 4 is very positive.. Let’s try it with a simple positive sentence.:

(require '[damionjunk.nlp.stanford :refer :all])

(sentiment-maps
"I am really having a wonderful day.")
; ({:sentiment 4, :text "I am really having a wonderful day."}) 

A neutral sentence works pretty nicely as well.

(sentiment-maps
"The French elections are on Television today")
; ({:sentiment 2, :text "The French elections are on Television today"}) 

The last one should be rather neutral, but for because of the double use of don’t and really the analysis gets a bit confused and gets this as a negative sentiment:

(sentiment-maps
  "I really don't want to think Marine will win the election")
; ({:sentiment 1, :text "I really don't want to think Marine will win the election"}) 

Built with Hugo

© Nicolas Modrzyk 2017 - hellonico @ gmail dot com