Similarity search vs MMR retrieval in LangChain - Practice Questions

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Similarity Search in Langchain

What is the main purpose of similarity search in Langchain?

A To randomly select documents from the database.

B To find documents most similar to a query based on vector embeddings.

C To sort documents alphabetically by title.

D To remove duplicate documents from the dataset.

💡 Hint

Think about how Langchain uses vectors to find related content.

Similarity search uses vector embeddings to find documents closest in meaning to the query.

❓ component_behavior

intermediate

Behavior of MMR Retrieval in Langchain

What unique behavior does Maximal Marginal Relevance (MMR) retrieval add compared to simple similarity search?

A It only returns the single most relevant document.

B It sorts documents by their creation date.

C It balances relevance with diversity to avoid redundant results.

D It filters out documents with low word counts.

💡 Hint

MMR tries to mix relevance with variety in results.

MMR retrieval selects documents that are relevant but also diverse to reduce repetition.

📝 Syntax

advanced

Correct Usage of MMR Retrieval in Langchain Code

Which code snippet correctly initializes an MMR retriever with a vector store in Langchain?

LangChain

from langchain.vectorstores import FAISS
from langchain.retrievers import MMRRetriever

vectorstore = FAISS.load_local('my_faiss_index')

# Choose the correct MMR retriever initialization:

A retriever = MMRRetriever(vectorstore=vectorstore, k=5, fetch_k=10, lambda_mult=0.5)

B retriever = MMRRetriever(vectorstore=vectorstore, top_k=5, fetch_k=10, lambda_mult=0.5)

C retriever = MMRRetriever(store=vectorstore, k=5, fetch_k=10, lambda_mult=0.5)

D retriever = MMRRetriever(vectorstore=vectorstore, k=5, fetch=10, lambda_mult=0.5)

💡 Hint

Check the parameter names carefully in the MMRRetriever constructor.

The correct parameters are vectorstore, k, fetch_k, and lambda_mult.

❓ state_output

advanced

Output Difference Between Similarity Search and MMR Retrieval

Given a query, what difference would you expect in the list of documents returned by similarity search vs MMR retrieval?

A Similarity search returns documents sorted by length; MMR returns documents sorted by date.

B Similarity search returns only one document; MMR returns all documents.

C Similarity search returns random documents; MMR returns documents with the highest word count.

D Similarity search returns the top most similar documents, possibly with redundancy; MMR returns relevant but more diverse documents.

💡 Hint

Think about how MMR tries to reduce repeated information.

Similarity search focuses on closeness to query, MMR adds diversity to avoid repeated content.

🔧 Debug

expert

Debugging Unexpected Results in MMR Retrieval

You use MMR retrieval but notice the results are very similar and not diverse. Which mistake in your code is most likely causing this?

LangChain

retriever = MMRRetriever(vectorstore=vectorstore, k=5, fetch_k=5, lambda_mult=0.0)
results = retriever.get_relevant_documents(query)

A Setting lambda_mult to 0 disables diversity, causing results to be similar.

B Setting fetch_k equal to k causes an error and no results.

C Not setting a random seed causes repeated results.

D Using get_relevant_documents instead of get_documents causes wrong output.

💡 Hint

Check the effect of the lambda_mult parameter on diversity.

Lambda_mult controls the trade-off between relevance and diversity; 0 means no diversity.