Sentiment Analysis
with sentimentR
The Problem
Suppose you are launching an exciting new meme…
meme ranked third best on Reddit
The Problem
Your only source of feedback is thousands of comments.
How do you turn this into workable data?
| “This is hilarious!” |
| “Made me laugh out loud” |
| “Not bad” |
| “Meh” |
| “I don’t get it” |
| … |
| “Could be better” |
The Solution
Sentiment Analysis
Natural language processing enables the vectorization of arbitrary text.
| “This is hilarious!” | Good |
| “Made me laugh out loud” | Good |
| “Not bad” | Neutral |
| “Meh” | Bad |
| “I don’t get it” | Bad |
| … | |
| “Could be better” | Neutral |
The Solution
Outcome
- Objective
- Systemic
- Automated
Now you do not need to read every comment to know how your meme is performing!
Sentiment Analysis
Vectorization
- Computers do not process in good, neutral, and bad.
- Sentiment Analysis assigns a decimal to each statement.
| “This is hilarious!” | Good | 0.90 |
| “Made me laugh out loud” | Good | 0.50 |
| “Not bad” | Neutral | 0.21 |
| “Meh” | Bad | -0.65 |
| “I don’t get it” | Bad | -0.55 |
| … | ||
| “Could be better” | Neutral | -0.17 |
Sentiment Analysis
Vectorization
What does this mean for me?
Now, you can convert qualitative data to quantitative data!
- Objective data based on established standards
- Consistency throughout implementation
- Little room for researcher bias
Sentiment Analysis
Expanding beyond sentiment
At its core, sentiment analysis uses large dictionaries of words and their weights.
| “excellent” | 1.0 |
| “awful” | -1.0 |
\[ Sentiment(T) = \sum_{i=1}^{T_{length}}weight(T_i)\cdot valence(T_{i_{valence}}) \]
Sentiment Analysis ### Valencies
- Some words indicate different connotations:
Very, Slightly, Absolutely, Not, Never - SentimentR knows to process these words in a logical matter
Sentiment Analysis
Limitations
- Words that are misspelled are dropped
- Confusing syntax and double negatives cause confusion
It was not the case that I had an amazing time at the event bc it was teribl.
Sentiment: 0.1147079 (slightly positive)
Sentiment Analysis
Expanding beyond sentiment
Why is this useful?
You can use sentiment analysis to understand nearly any idea, not just positive or negative.
Applications
How is this useful?
- Many companies use sentiment analysis to process customer feedback.
- Customer surveys
- Product reviews
- Etc.
- Political campaigns
- Finance and Market
Applications
Getting started with code
sentimentr is very easy to use.
library(sentimentr)
texts <- c("I love this meme!", "This is okay.", "Not funny at all.")
sentiment(texts)Don’t be fooled by simple implmentation – the algorithim behind this package is more sophisticated than several other packages.
Sentiment Analysis
Learn more about sentiment analysis at
https://cran.r-project.org/web/packages/sentimentr/sentimentr.pdf
Presenter
Noah Christensen
noah_christensen1@baylor.edu