Sentiment Analysis is one of those things in Machine learning which is still getting improvement with the rise of Deep Learning based NLP solutions. There are many things like Sarcasm, Negations and similar items make Sentiment Analysis a rather tough nut to crack.
Deep learning as much as it’s effective, it’s also computationally expensive and if you are ready to trade off between Cost (expense) and Accuracy, then you this is the solution for building a negation-proof Sentiment Analysis solution (in R).
What’s Negation (Proof)?
lexicon based Sentiment Analysis solutions can’t handle Negations easily - which is that good is positive and not good is negative. Negation Proof solution is something that can handle such negations and classify their
For building a negation-proof sentiment analysis solution, we’re going to use the R package
sentimentr by Tyler Rinker.
According to the documentation of sentimentr,
So what does sentimentr do that other packages don’t?
sentimentr attempts to take into account valence shifters (i.e., negators, amplifiers (intensifiers), de-amplifiers (downtoners), and adversative conjunctions) while maintaining speed. Simply put, sentimentr is an augmented dictionary lookup.
For more information on Valence Shifters, Check out sentimentr’s documentation.
You can install the stable version of
sentimentr from CRAN:
install the development version from Github:
# install.packages("devtools") devtools::install_github("trinker/lexicon") devtools::install_github("trinker/sentimentr")
Loading the package
sentimentr and also
magrittr for pipe operator (
dplyr for data manipulation.
library(sentimentr) library(magrittr) library(dplyr)
Sample Text (with Negations)
Let’s define two sentences for us to check if
sentimentr is negation-proof.
text1 <- "I am a good girl. Today I am happy" text2 <- "I am not a good girl. Today I'm not happy"
Sentiment Analysis in Action
We’ll use the function
sentiment() to identify the approximate the sentiment (polarity) of text by sentence.
## element_id sentence_id word_count sentiment ## 1: 1 1 5 0.3354102 ## 2: 1 2 4 0.3750000
## element_id sentence_id word_count sentiment ## 1: 1 1 6 -0.3061862 ## 2: 1 2 4 -0.3750000
As we can see from the above outputs, using
sentimentr is doing a good job in rightly scoring the sentiment score for sentence with and without negations (valence shifters).
This is just a very simple (perhaps, Naive too) walkthrough of the
sentimentr package but it has got a lot more like
profanity() and much more. Tyler Rinker has got a very nice, comprehensive and super-helpful documentation and also benchmarks comparing
sentimentr with other similar packages.