alpineapache / ai_biases_final_project

Decoding Biases in Artificial Intelligence: Testing Python Sentiment Analysis Libraries for Racial and Gender Biases

by Nicolaus CHRISTENSON, Matteo NEBBIAI & Orestas STRAUKA

The Paper

The final version of the paper can be found here.

Requirements

In order to run the code yourself, make sure you follow these steps:

• install Python 3.6 on your machine.
• run pip install requirements.txt in your console.
• run python index.py in your console.

The Files

• index.py = the main script that has to be called and invokes all of the relevant functions.
• equity_evaluation_corpus.csv = the file containing the test sentences from the Equity Evaluation Corpus with adjusted column names and some redactions/additions.
• text_blob_library.py = the script that returns the sentiment analysis score of the TextBlob library for each sentence.
• vader_sentiment_library.py = the script that returns the sentiment analysis score of the VADERSentiment library for each sentence.
• results.csv = the file containing the data from equity_evaluation_corpus.csv and the scores of each library for each sentence.
• analysis.py = the file that runs statistical analysis on results.csv along racial and gender lines.
• gender_bias_results.csv = the file containing the results of analysis.py on gender bias in the libraries.
• racial_bias_results.csv = the file containing the results of analysis.py on racial bias in the libraries.
• mean_differences.png = the file containing the plot of the differences between the mean predicted scores of clusters of sentence pairs.
• requirements.txt = the file containing all of the dependencies.