Balancing Relevance and Diversity with MMR Search - Qdrant

This post introduces and demonstrates the implementation of Maximum Marginal Relevance (MMR) search within Qdrant. It explains how MMR balances query relevance with result diversity, providing a practical example using the DeepFashion dataset and CLIP embeddings. The post details the setup, data processing, embedding generation, Qdrant collection configuration, and provides Python code for both standard similarity search and MMR search, highlighting the difference in results for a 'black jacket' query. It also shows how MMR can be combined with advanced filtering.