wordorientation is an R package that measures how strongly pairs of words attract or repel each other in a text corpus.
For every pair of words, it computes the phi coefficient (a correlation measure for co-occurrence data), tests it for statistical significance, and classifies the pair as:
- attraction — the words co-occur more often than chance would predict
- repulsion — the words co-occur less often than chance would predict
- neutral — no significant relationship
Unlike general-purpose correlation tools, it's built specifically for text: it handles tokenization and stopword removal automatically, applies proper significance testing rather than just reporting a raw correlation, and produces a ready-to-plot attraction/repulsion network.
Installation
install.packages("wordorientation")
Basic usage
library(wordorientation)
result <- analyze_word_orientation(
my_data,
text_col = "text",
doc_col = "id",
min_count = 2
)
head(result$scored)
plot_orientation_network(result$scored)
Each row of your data is treated as one "document" — the unit within which two words are said to co-occur.
One thing worth knowing: the package needs a large corpus (survey responses, social media posts, reviews, or similar) to have enough statistical power to find real patterns. On a single short text, it will correctly report few or no significant pairs rather than manufacturing noise into a misleading result.
Full source and documentation: https://github.com/yosleycarrero2025/wordorientation
Feedback and issues welcome!
