Workshop: Natural Language Processing for Data Science

UC Davis DataLab
Spring 2024

Instructor: Tyler Shoemaker <tshoemaker@ucdavis.edu>, Carl Stahmer <cstahmer@ucdavis.edu>

Maintainer: Tyler Shoemaker <tshoemaker@ucdavis.edu>

• Reader
• Event Page

This week-long workshop series covers the basics of text mining and natural language processing (NLP) with Python. We will focus primarily on unstructured text data, discussing how to format and clean text to enable the discovering of significant patterns in collections of documents. Sessions will introduce participants to core terminology in text mining/NLP and will walk through methods that range from tokenization and dependency parsing to text classification, topic modeling, and word embeddings. Basic familiarity with Python is required. We welcome students, postdocs, faculty, and staff from a variety of research domains, ranging from health informatics to the humanities.

Note: this series concludes with a special session on large language models, "The Basics of Large Language Models."

By the end of this series, you will be able to:

• Clean and structure textual data for analysis Recognize and explain how these cleaning processes impact research findings
• Explain key concepts and terminology in text mining/NLP, including tokenization, dependency parsing, word embedding
• Use special data structures such as document-term matrices to efficiently analyze multiple texts
• Use statistical measures (pointwise mutual information, tf-idf) to identify significant patterns in text
• Classify texts on the basis of their features
• Produce statistical models of topics from/about a collection of texts
• Produce models of word meanings from a corpus

Contributing

The course reader is a live webpage, hosted through GitHub, where you can enter curriculum content and post it to a public-facing site for learners.

To make alterations to the reader:

1. Check in with the reader's current maintainer and notify them about your intended changes. Maintainers might ask you to open an issue, use pull requests, tag your commits with versions, etc.

2. Run git pull, or if it's your first time contributing, see Setup.

3. Edit an existing chapter file or create a new one. Chapter files may be either Markdown files (.md) or Jupyter Notebook files (.ipynb). Either is fine, but you must remain consistent across the reader (i.e. don't mix and match filetypes). Put all chapter filess in the chapters/ directory. Enter your text, code, and other information directly into the file. Make sure your file:

   • Follows the naming scheme ##_topic-of-chapter.md/ipynb (the only exception is index.md/ipynb, which contains the reader's front page).
   • Begins with a first-level header (like # This). This will be the title of your chapter. Subsequent section headers should be second-level headers (like ## This) or below.

   Put any supporting resources in data/ or img/.

4. Run the command jupyter-book build . in a shell at the top level of the repo to regenerate the HTML files in the _build/.

5. When you're finished, git add:

   • Any files you edited directly
   • Any supporting media you added to img/

   Then git commit and git push. This updates the main branch of the repo, which contains source materials for the web page (but not the web page itself).

6. Run the following command in a shell at the top level of the repo to update the gh-pages branch:

   ghp-import -n -p -f _build/html
   

   This uses the ghp-import Python package, which you will need to install first (pip install ghp-import). The live web page will update automatically after 1-10 minutes.

Setup

Python Packages

We recommend using conda to manage Python dependencies. The env.yaml file in this repo contains a list of packages necessary to build the reader. You can create a new conda environment with all of the packages listed in that file with this shell command:

conda env create --file env.yaml