README.md

In chris31415926535/tidyvader: Tidyvader: Fast, Clear, and Tidy Rule-Based Sentiment Analysis with VADER

tidyvader

A fast, clear, and tidy implementation of the rule-based sentiment analysis algorithm VADER (Valence Aware Dictionary and Sentiment Reasoner).

• It’s fast because the code has been re-written from scratch and uses C++ for the core algorithm: in benchmarking it’s over 500 times faster than the competing R package.
• It’s clear because it tries to make the rule-based algorithm and the dictionaries simple to read so that users can inspect and judge them.
• It’s tidy because it tries to follow tidy design principles and works well with the %>% pipe.

Under Development

Please note that this package (and this documentation) is under active development. At present it’s pretty well tested and functional, but there are known limitations and there may yet be bugs. This is a development package not yet on CRAN and things may change. Expect more/better documentation and development as soon as time allows.

Installation

You can install the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("chris31415926535/tidyvader")

What is VADER?

VADER’s authors describe it on their GitHub page as “a lexicon and rule-based sentiment analysis tool that is specifically attuned to sentiments expressed in social media” that was originally written in Python (link).

Let’s break this definition down:

• A sentiment analysis tool takes plain English text and decides whether it expresses positive or negative feelings.
• A lexicon in this context is a dictionary that assigns words positive or negative scores.
• A set of rules modifies these scores based on other words in the sentence. For example, one rule says that if a positive-score word has a negation in front of it, like “not happy,” that word should flip to have a negative score.
• And text in social media is different from, for example, text in the New York Times, so VADER’s lexicon and rules are tailored for online communications.

This is notably different from two other common approaches to sentiment analysis:

• A “bag-of-words” approach like AFINN has a lexicon but no rules, so it can’t tell the difference between “I’m happy” and “I’m not happy” (much less “I’m not not unhappy”).
• A neural-net or other unsupervised approach is essentially a “black box,” and it’s hard (or impossible) to understand and evaluate the way it generates outputs. It may have something approaching rules and a lexicon, but you can’t review them to judge how credible they are.

VADER has advantages over both of these approaches. First, it’s more nuanced than a pure bag-of-words approach and so it should be more accurate. Second, it’s more surveyable than an unsupervised approach and so users can make informed decisions about when and how it’s appropriate to use it.

Why tidyvader?

Example

This example shows how to send sentences in a dataframe through vader(). It also shows how punctuation, capitalization, modifiers, and negations all work together to affect a sentence’s compound score.

library(tidyvader)
library(tibble)
library(magrittr)
library(knitr)

# set up a tibble with some sentences
texts <- tibble(sentences = c("I feel happy today.",
                              "I feel happy today!",
                              "I feel HAPPY today!",
                              "I feel NOT HAPPY today!",
                              "I feel REALLY NOT HAPPY today!"))

# pipe the data to tidyvader::vader() and specify the column with text 
texts %>%
  tidyvader::vader(sentences) %>%
  knitr::kable()

| sentences | compound | pos | neu | neg | | :------------------------------ | -------: | -----: | -----: | -----: | | I feel happy today. | 0.5719 | 0.5522 | 0.4478 | 0.0000 | | I feel happy today! | 0.6114 | 0.5709 | 0.4291 | 0.0000 | | I feel HAPPY today! | 0.6932 | 0.6117 | 0.3883 | 0.0000 | | I feel NOT HAPPY today! | -0.5903 | 0.0000 | 0.5107 | 0.4893 | | I feel REALLY NOT HAPPY today! | -0.6761 | 0.0000 | 0.5234 | 0.4766 |

If you want to score a single sentence in a length-1 character vector you can use vader_chr(). This is good for quickly checking things, but it’s much slower than vader() so I don’t recommend it for analysis at scale. The results will come in a one-row tibble, like so:

tidyvader::vader_chr("I feel HAPPY today!") %>%
  knitr::kable()

| compound | pos | neu | neg | | -------: | -----: | -----: | --: | | 0.6932 | 0.6117 | 0.3883 | 0 |

You can also easily pull the VADER dictionaries and some test sentences in a nested tibble using get_vader_dictionaries(). It’s easy to take a look through RStudio’s viewer, and you can also pull them out and inspect them as regular tibbles.

library(dplyr)

vader_dicts <- tidyvader::get_vader_dictionaries()

vader_sentiments <- vader_dicts %>%
  filter(name == "dict_sent_sorted") %>%
  pull(dictionary) %>% `[[` (1)

vader_sentiments[2968:2973,] %>%
  knitr::kable()

| word | sentiment | | :---------- | --------: | | friendship | 1.9 | | friendships | 1.6 | | fright | -1.6 | | frighted | -1.4 | | frighten | -1.4 | | frightened | -1.9 |

Known Limitations

• Unicode emojis aren’t supported yet 😢
• Several special cases in base VADER aren’t implemented yet: e.g. some multi-word phrases like “never so” and “without a doubt.”
• Expressions and turns of phrase in base VADER like “bad ass” not yet supported.
• Pos/Neu/Neg scores don’t match the Python implementation when ascii emojis and caps differences are present, although the compound score matches.s
• There’s no support for custom dictionaries.
• VADER’s lexicon is 6 years old so doesn’t reflect some currant usage (DEAD!).
• VADER’s lexicon includes some terms that will limit its applicability in some contexts and communities. Not all words are used in the same way by all people in all settings.

Resources and References

• VADER’s Python GitHub page: https://github.com/cjhutto/vaderSentiment

• Citation for conference proceedings introducing VADER:

  • Hutto, C.J. & Gilbert, E.E. (2014). VADER: A Parsimonious Rule-based Model for Sentiment Analysis of Social Media Text. Eighth International Conference on Weblogs and Social Media (ICWSM-14). Ann Arbor, MI, June 2014.

chris31415926535/tidyvader documentation built on June 9, 2025, 1:50 p.m.