Recommender System for US Court Cases

This repository contains the prototype of a recommender system that can do content-based citation recommendations based on the majority opinion of US court cases.

Folder Structure

• data: Contains all the code for
  • Data preprocessing
  • Creation of the datastructures required for training, validation and testing
  • Data analysis
• document_embedding: Contains all the code for the hierarchical recurrant neural net (HRNN) document embedder
• pretrained_word_embedding: contains all the code to generate the needed datastructures from the pre-trained GloVe word embedding
• ranking_models: Contains the code for the ItemPopularity and EmbedTextNCF ranking models
• training: Contains the code for the training of the recommender systems

Running the Code

In order to run the entire training you must do the following steps

1. Create the word-embedding datastructures from GloVe as described in the folder pretrained_word_embedding
2. Create the training, validaiton and test data structures as described in the folder data (this requires access to a proprietary dataset from LawEcon)