Similarity search vs MMR retrieval in LangChain - Hands-On Comparison

Similarity Search vs MMR Retrieval with LangChain

📖 Scenario: You are building a simple document search tool using LangChain. You want to compare two ways to find relevant documents: similarity search and Maximal Marginal Relevance (MMR) retrieval. Similarity search finds documents closest to your query. MMR retrieval finds documents that are both relevant and diverse.

🎯 Goal: Build a LangChain script that loads a small set of documents, sets up a vector store, and performs both similarity search and MMR retrieval. See how the results differ.

📋 What You'll Learn

Create a list of 3 documents with exact texts

Create a LangChain vector store from the documents

Perform a similarity search for the query 'apple fruit'

Perform an MMR retrieval for the query 'apple fruit' with k=2 and fetch_k=3

💡 Why This Matters

🌍 Real World

This project shows how to build a simple document search tool that can find relevant information quickly and also provide diverse results to avoid repetition.

💼 Career

Understanding similarity search and MMR retrieval is important for building smart search engines, recommendation systems, and chatbots that use document retrieval.

1

Create the documents list

Create a list called documents containing these exact strings: 'Apple is a fruit.', 'Banana is yellow.', and 'Carrots are vegetables.'

LangChain

# Create the documents list
# Your code here

Need a hint?

Use a Python list with the exact strings given.

2

Set up the vector store

Import FAISS and OpenAIEmbeddings from LangChain. Create a variable called embeddings as OpenAIEmbeddings(). Then create a vector_store by calling FAISS.from_texts(documents, embeddings).

LangChain

documents = ['Apple is a fruit.', 'Banana is yellow.', 'Carrots are vegetables.']
# Import FAISS and OpenAIEmbeddings
# Create embeddings and vector_store
# Your code here

Need a hint?

Use the exact class names and method calls as shown.

3

Perform similarity search

Use vector_store.similarity_search with the query 'apple fruit' and k=2. Store the result in a variable called similar_docs.

LangChain

from langchain.vectorstores import FAISS
from langchain.embeddings import OpenAIEmbeddings

documents = ['Apple is a fruit.', 'Banana is yellow.', 'Carrots are vegetables.']

embeddings = OpenAIEmbeddings()
vector_store = FAISS.from_texts(documents, embeddings)
# Perform similarity search for 'apple fruit' with k=2
# Your code here

Need a hint?

Call similarity_search on vector_store with the exact query and k value.

4

Perform MMR retrieval

Use vector_store.max_marginal_relevance_search with the query 'apple fruit', k=2, and fetch_k=3. Store the result in a variable called mmr_docs.

LangChain

from langchain.vectorstores import FAISS
from langchain.embeddings import OpenAIEmbeddings

documents = ['Apple is a fruit.', 'Banana is yellow.', 'Carrots are vegetables.']

embeddings = OpenAIEmbeddings()
vector_store = FAISS.from_texts(documents, embeddings)

similar_docs = vector_store.similarity_search('apple fruit', k=2)
# Perform MMR retrieval for 'apple fruit' with k=2 and fetch_k=3
# Your code here

Need a hint?

Call max_marginal_relevance_search on vector_store with the exact parameters.