Graph-Based Movie Chatbot

Overview

Built an intelligent QA chatbot capable of answering natural language questions about movies, actors, directors, and genres by querying a movie knowledge graph. The system also provides personalized movie recommendations based on user preferences.

System Architecture

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│  User Question  │────▶│  NER & Entity    │────▶│  Query Builder  │
│                 │     │  Linking         │     │                 │
└─────────────────┘     └──────────────────┘     └────────┬────────┘
                              │                           │
                              │ BERT + Fuzzy              │
                              │ Matching                  ▼
                              │                   ┌───────────────┐
                              │                   │    Hybrid     │
                              │                   │   Execution   │
                              │                   └───────┬───────┘
                              │                           │
                        ┌─────▼─────┐              ┌──────▼──────┐
                        │  LLM NER  │              │   SPARQL    │
                        │  Fallback │              │   + kNN     │
                        └───────────┘              └──────┬──────┘
                                                         │
                                                         ▼
                                                 ┌───────────────┐
                                                 │    Answer     │
                                                 │  Generation   │
                                                 └───────────────┘

Key Features

Entity Recognition & Linking: Combined BERT-based NER with fuzzy string matching to accurately identify and link movie entities, actor names, and genres from user queries.
LLM Fallback: Integrated LLM-based NER for handling complex or ambiguous queries where traditional methods fail.
Hybrid Query Execution: Implemented both SPARQL queries for structured retrieval and embedding-based vector search for semantic similarity matching.
Smart Recommendations: Designed a hybrid recommendation approach combining graph-based feature overlap (shared actors, directors, genres) with embedding-based kNN retrieval for diverse suggestions.

Example Queries

"Who directed Inception?"
"What movies has Leonardo DiCaprio acted in?"
"Recommend me movies similar to The Dark Knight"
"List all sci-fi movies from 2020"
"What's the rating of Parasite?"

Tech Stack

Python SPARQL BERT LLM RDF FAISS FuzzyWuzzy Neo4j