BRASS - Categorical variable association estimator

BRASS is a Bayesian method for estimating the degree of association between categorical variables. It is described in the paper "Applying a new category association estimator to sentiment analysis on the Web" (Xavier et al., 2023), where it was applied to annotations of emotions identified on tweets in Portuguese.

Structure of the project:

.
├── analysis                <- Analysis made for the paper
|   └── plots               <- Where plots generated by the notebooks are saved
├── data                    <- Where data used in the analysis is stored
|   ├── chains              <- Where MCMC chains for sentiment analysis are saved
|   ├── cleaned             <- Auxiliary data for translating emotions
|   ├── processed           <- Final annotations for the tweets (majority voting)
|   └── sim_chains          <- Where MCMC chains for synthetic data is saved
├── examples                <- Jupyter notebooks showing how to use BRASS
├── LICENSE                 <- The license for this work
├── README.md               <- This document
├── brass.py                <- A Python module that implements BRASS as a class
├── min_requirements.txt    <- Python packages required by brass.py
└── requirements.txt        <- Python packages required to run everything

Installation

The Python packages required to run BRASS and all the analysis done in the paper are specified in requirements.txt. If you just want to use the BRASS module, you can run the following command:

  pip install -r min_requirements.txt

Then, just copy the brass.py file to a local folder where it can be found.

Association between emotions

The analysis presented in the paper, that tests the method on synthetic data and applies it to emotions detected on tweets in portuguese, is available in the ./analysis folder.

The data resulting from the complete annotation process of 4,613 tweets is available at ./data/processed. This is the data used in the analysis presented in the paper.

Citing this work

If you use the data, code, or the analysis in this repository, please cite:

@unpublished{Xavier2023, author = "Henrique S. Xavier and Diogo Cortiz and Mateus Silvestrin and Ana Lu'isa Freitas and Let'icia Yumi Nakao Morello and Fernanda Naomi Pantale~ao and Gabriel Gaudencio do R\^ego", title = "Applying a new category association estimator to sentiment analysis on the Web", archivePrefix = {arXiv}, eprint = {2311.05330}, primaryClass = {stat.AP}, month = "11", year = "2023" }

Contact

For more information, contact Henrique S. Xavier (https://github.com/hsxavier).