README.md

---
title: Big Data Application
emoji: 🌖
colorFrom: red
colorTo: indigo
sdk: docker
pinned: false
license: mit
---

# Lab 03

## What I did

1. I used the first 20 document from gamino/wiki_medical_terms to create the PropertyGraphIndex, using 2 extractors: `ImplicitPathExtractor` and `DynamicLLMPathExtractor`. The result is a graph with the chunked text from the original documents and relationship between subject such as diseases and symptom.
2. I then enhance the graph using LightGCN by using the node's embedding as the initial embedding of LightGCN. The model is then trained for 300 epochs. The trained model's embedding is used as the new embedding for each node.