Github README Natural Language Processing/ Web-Scraping

Josh Burch, Mack McGlenn, Skyler Schnee

Codeup: O'neil Cohort

Project Overview

The goal will be to build a model that can predict the main programming language of a repository, given the text of the README file.

Objectives

• Document code, process (data acquistion, preparation, exploratory data analysis and statistical testing, modeling, and model evaluation), findings, and key takeaways in a Jupyter notebook final report.

• Create modules (acquire.py, prepare.py) that make our process repeateable and our report (notebook) easier to read and follow.

Deliverables

• A GitHub repository containing our work
  • a well-documented jupyter notebook that contains our analysis
• a README file that contains a description of our project, as well as instructions on how to run it
• 2-5 google slides suitable for a general audience: - our slides should summarize our findings in exploration and the results of our modeling - include well-labeled visualizations in our slides - our google slide deck should be linked in the README of our repository

Project Planning

Outline:

• Acquire & Prepare Data
• Exploration
• Modeling
• Conclusion

1. Acquire & Prepare

Acquire Actions

• Data acquired from web-scraping 30 pages of README.md files from popular github repositories on 15 May 2023.
  • Filters: Java, Javascript, Python, most starred

Prepare Actions: - Convert text to all lower case for normalcy. - Remove any accented characters, non-ASCII characters. - Remove special characters. - Stem or lemmatize the words. - Remove stopwords. - Store the clean text and the original text for use in future notebooks.

• Remove "giveaways"

  • Remove 'python', 'py', 'java', 'javascript', 'js', 'script'
  • Removed because these words showed to be dead giveaways of the language we are predicting.

• Calculate README word counts

  • Define word counts for analysis, exploration, and determining a proper statistical test.

Functions called from:

• acquire.py
• prepare.py
• functions.py

2. Exploration/ Correlation Analysis

Here are the steps we took during the exploration phase:

• performed statistical test
• created bigrams
• create word clouds
• visualize top 20 words with frequencies

3. Modeling

Here are the steps we took during the modeling phase:

• define the basline (33% accuracy)
• created 4 ML models
• judged model performace on train and validate subset accuracy scores
• ran our best model on test subset

In conclusion, we are successful in our efforts to out-perform baseline accuracy predicting the programing language by analyzing the README of a github repository between python, java, and javascript. Our K-Nearest Neighbor Classifier was our best performing model with solid accuracy scores on train and validate sets, with the most most relatively optimal score delta.

Conclusion & Reccomendations

• To understand the limitations we purposely placed on ourselves in the effort of using alternate methods to increase our model performance, re-review the "giveaway" words we removed from the dataset. Words like "python" and "js" among others proved to be giveaways of the programming language used.
• To make futher advancements with this project, we would like to scrape more github README content and re-run through this notebook to further solidify or improve our resluts. We would also like to experiment with additional models and parameters.

Steps to Reproduce

• create env.py file
  • needs username varible with github username
  • needs github token for api use
• clone repo and run the notebook