IMDb Knowledge Graph RAG

Made using Neo4j, LangChain, Gemini API and IMDb Top 250 movies dataset

Click and zoom in to see nodes and relationships.

• Neo4j database stores the network graph.
  • There are 3 types of nodes - Movie, Genre, Person
  • Person can be - Actor, Director, Writer or a combination of these
  • Relationships are -
    • Person - [ WROTE / ACTED_IN / DIRECTED ] -> Movie
    • Movie - [ BELONGS_TO ] -> Genre
• LangChain and Gemini are used for the pipeline, which
  • Processes the natural language prompt
  • Generates a Cypher query
  • Queries the Neo4j database with generated query and gets back the result in JSON
  • Parses the JSON and responds back in natural language

Steps to run -

• Clone the repo and navigate to the diretory.
• Download the dataset, rename and move it to the /data directory as imdb.csv.
• Create a virtual environment with Python version 3.10.14, install the requirements from requirements.txt. For Conda,

  $ conda create -c conda-forge --name <env> --file requirements.txt
  

• Recommended - Create new Neo4j database. (for Community edition)
• Start the Neo4j server.
• Fill in Neo4j credentials and Gemini API key in .env_template and rename to .env.
• First create the network graph by running the Jupyter Notebook ./src/Knowledge_Graph.ipynb.
• Run the Jupyter Notebook ./src/Graph_RAG.ipynb.
• Run the Streamlit web app by running

  streamlit run ./src/App.py
  

Credits

• Chidambara Raju G - IMDb Top 250 Movies Dataset