Skip to main content

Chapter 2: Hybrid Search

Search TypeStrengthWeakness
Keyword (BM25)Exact matches, rare termsMisses synonyms
SemanticUnderstanding meaningMay miss exact terms
HybridBest of bothMore complex

BM25 Overview​

BM25 is a keyword-based ranking algorithm:

from rank_bm25 import BM25Okapi

# Tokenize documents
tokenized_corpus = [doc.split() for doc in documents]
bm25 = BM25Okapi(tokenized_corpus)

# Search
query_tokens = query.split()
scores = bm25.get_scores(query_tokens)

With LangChain​

from langchain.retrievers import EnsembleRetriever
from langchain.retrievers import BM25Retriever

# BM25 retriever
bm25_retriever = BM25Retriever.from_documents(documents)
bm25_retriever.k = 5

# Vector retriever
vector_retriever = vectorstore.as_retriever(search_kwargs={"k": 5})

# Combine
ensemble_retriever = EnsembleRetriever(
    retrievers=[bm25_retriever, vector_retriever],
    weights=[0.4, 0.6]  # BM25 weight, Vector weight
)

# Search
results = ensemble_retriever.get_relevant_documents(query)

Reciprocal Rank Fusion (RRF)​

Combine rankings from multiple retrievers:

def reciprocal_rank_fusion(rankings, k=60):
    """
    Combine multiple rankings using RRF.
    k: constant to prevent high ranks from dominating
    """
    fused_scores = {}
    
    for ranking in rankings:
        for rank, doc in enumerate(ranking):
            doc_id = doc.metadata.get('id', hash(doc.page_content))
            if doc_id not in fused_scores:
                fused_scores[doc_id] = {'doc': doc, 'score': 0}
            fused_scores[doc_id]['score'] += 1 / (k + rank + 1)
    
    # Sort by fused score
    sorted_docs = sorted(
        fused_scores.values(), 
        key=lambda x: x['score'], 
        reverse=True
    )
    
    return [item['doc'] for item in sorted_docs]

✅ Use hybrid when:

  • Queries contain specific terms/names
  • Domain has specialized vocabulary
  • Users expect exact matches

❌ Stick with semantic when:

  • Natural language queries
  • Concept-based search
  • Multilingual content

Next Steps​

Continue to Chapter 3: Reranking.