For data preprocessing, sentiment analysis, and evaluation, we used Jupyter Notebook with Python. Alongside Python, the packages "pandas" and "re" need to be installed for data preprocessing. The packages "pandas", "torch", "transformers", and "pathlib" are needed for sentiment analysis. We utilized (and recommend) "Anaconda" to handle Python dependencies.
For hypothesis testing, model development, and additional evaluation, we used RStudio with R. Alongside R, the packages "tidyverse" and "caret" need to be installed for hypothesis testing.
We all worked with Mac OS.
DS-4002-Project-1 Folder Outline
│
├── DATA
│ ├── cleaned
│ │ ├── bases
│ │ │ ├── Cabin
│ │ │ │ ├── .ipynb_checkpoints
│ │ │ │ │ └── cabin_base_cabin_Business Class-checkpoint.csv
│ │ │ │ ├── cabin_base_cabin_Business Class.csv
│ │ │ │ ├── cabin_base_cabin_Economy.csv
│ │ │ │ ├── cabin_base_cabin_First Class.csv
│ │ │ │ ├── cabin_base_cabin_Premium Economy.csv
│ │ │ │ └── cabin_base_cabin_Unknown.csv
│ │ │ ├── Traveler
│ │ │ │ ├── traveler_base_traveler_Business.csv
│ │ │ │ ├── traveler_base_traveler_Couple Lesisure.csv
│ │ │ │ ├── traveler_base_traveler_FamilyLeisure.csv
│ │ │ │ ├── traveler_base_traveler_Solo Leisure.csv
│ │ │ │ └── traveler_base_traveler_Unknown.csv
│ │ │ └── .DS_Store
│ │ ├── data/.ipynb_checkpoints
│ │ │ ├── airline-checkpoint.csv
│ │ │ └── airport-checkpoint.csv
│ │ ├── .DS_Store
│ │ ├── airline_all_cols.csv
│ │ ├── airline_cleaned.csv
│ │ └── airline_dummies_only.csv
│ │
│ ├── original
│ │ ├── airline.csv
│ │ ├── airport.csv
│ │ ├── lounge.csv
│ │ └── seat.csv
│ │
│ ├── with_mismatch
│ │ └── airline_with_mismatch.csv
│ │
│ ├── with_sentiments
│ │ └── airline_cleaned_with_sentiment.csv
│ │
│ └── Data Appendix - DS 4002 Group 1 Project 1.pdf
│
├── OUTPUT
│ ├── cfnmatrix_results.png
│ ├── eda_plots_with_purposes.png
│ ├── logregmodel_results.png
│ ├── modelfit_results.png
│ └── oddsratio_results.png
│
├── SCRIPTS
│ ├── cleaning_data.ipynb
│ ├── hypothesis_testing.Rmd
│ └── sentiment_analysis.ipynb
│
├── LICENSE.md
│
├── README.md
│
└── DS4002 Project 1 Slides.pdf
To reproduce our results, first download the data in the DATA/original directory, titled 'airline.csv.' Make sure to install the packages required for data preprocessing as listed above in this README. An alternative recommendation is to install Anaconda and enter a conda virtual environment. Afterwards, run the the 'cleaning_data.ipynb' script in the SCRIPTS directory. This will produced cleaned data and dummy variable files. Ensure that the cleaned files are in the DATA/cleaned directory if they are not already.
Next, complete sentiment analysis by firstly installing the packages required for sentiment analysis as mentioned above in this README or being in a conda environment before running the code. Ensure that your directory's structure for your existing files so far matches the structure of our directory (which should mainly be data right now). Afterwards, download the 'sentiment_analysis.ipynb' notebook from the SCRIPTS folder and run the entire file, once again ensuring the structure is the same. It may take >40 min to finish running. The conclusion of the notebook running should result in a 'with_sentiments' folder in the DATA directory with a new .csv file with sentiment analysis scores attached to each review, and a 'with_mismatch' folder in the DATA directory with a new .csv file with the mismatch indicators attached to each review.
Finally, get results for hypothesis testing by making sure to install and load in the two R packages as mentioned above in this README. Download the 'hypothesis_testing.Rmd' script from the SCRIPTS folder and run the entire file (after making sure to set the working directory to the same one as where the cleaned data files are located). Output produced in R by using the hypothesis testing script should mimic the output as presented in the OUTPUT folder exactly.
[1] Pongsathon Pookduang, Rapeepat Klangbunrueang, Wirapong Chansanam, and Tassanee Lunrasri, “Advancing Sentiment Analysis: Evaluating RoBERTa against Traditional and Deep Learning Models,” Engineering Technology & Applied Science Research, vol. 15, no. 1, pp. 20167–20174, Feb. 2025, doi: https://doi.org/10.48084/etasr.9703.
[2] quankiquanki, “GitHub - quankiquanki/skytrax-reviews-dataset: An air travel dataset consisting of user reviews from Skytrax (www.airlinequality.com),” GitHub, 2025. https://github.com/quankiquanki/skytrax-reviews-dataset