███████╗    ██████╗ ██╗███████╗ ██████╗ ██████╗ ██╗   ██╗███████╗██████╗ ██╗   ██╗
██╔════╝    ██╔══██╗██║██╔════╝██╔════╝██╔═══██╗██║   ██║██╔════╝██╔══██╗╚██╗ ██╔╝
█████╗      ██║  ██║██║███████╗██║     ██║   ██║██║   ██║█████╗  ██████╔╝ ╚████╔╝
██╔══╝      ██║  ██║██║╚════██║██║     ██║   ██║╚██╗ ██╔╝██╔══╝  ██╔══██╗  ╚██╔╝  
███████╗    ██████╔╝██║███████║╚██████╗╚██████╔╝ ╚████╔╝ ███████╗██║  ██║   ██║   
╚══════╝    ╚═════╝ ╚═╝╚══════╝ ╚═════╝ ╚═════╝   ╚═══╝  ╚══════╝╚═╝  ╚═╝   ╚═╝

Tree Discovery

Machine Learning Algorithms for E-Discovery

Some background of the project Discovery in a civil lawsuit is a process where each party turns over documents that are relevant to the case. We improve this process with an utility that helps law professionals classify these documents more efficiently.

We process emails through TF-IDF and LSA and use the output to train our proprietary implementation of random forest.

Notable Parts of the Directory Structure

• Documentation
• flask_server
• data_api.py: Everything to do with interfacing between our ML backend and the beautiful front end.
• lib
• forest.py: Entry point into the ML code
• web_new
• src: All of the front end code is in here
• email_filter.py
• README.md: This file!
• makefile
• scenario1Full.py: Run a full scenario using this file.

Dependencies

At Project Root:

• Install Everything: make all_dep
• Install Python libs: make py_dep
• Install NPM Modules: make web_dep

How to run

Web:

1. Then run make web
2. command+t to make a new terminal tab and run make -B web in the new tab at project root and
3. Navigate to http://localhost:8080/

Running Scenario 1 from TREC 2011:

python3 scenario1Full.py

The Data

We used a combination of Kaggle's dataset and Trec's

Authors

Ju Yun Kim Dan Mayer Lazar Zamurovic Adam Tigar Micah Nacht Xingfan Xia

Advised by Eric Alexander, Carleton College Computer Science Professor