Show how to implement FAISS-based semantic search

 

Implement FAISS-Based Semantic Search with Sentence Transformers

FAISS (Facebook AI Similarity Search) is a powerful library that speeds up semantic search by enabling fast nearest-neighbor search on large-scale datasets.

---

Install Required Libraries

If you haven't installed them yet:

pip install sentence-transformers faiss-cpu

---

How FAISS Works:

1. Convert all documents into embeddings using Sentence Transformers.
2. Store embeddings in a FAISS index.
3. Convert the search query into an embedding.
4. Use FAISS to quickly retrieve the most similar documents.

---

1. Generate Embeddings for Documents

We'll use Sentence Transformers to generate embeddings.

from sentence_transformers import SentenceTransformer
import numpy as np

# Load model
model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')

# Example documents
documents = [
    "I love playing football.",
    "Artificial Intelligence is the future.",
    "Machine learning powers AI.",
    "The weather is sunny today.",
    "Deep learning improves neural networks.",
    "It's raining outside."
]

# Generate embeddings
document_embeddings = model.encode(documents)
print("Document Embeddings Shape:", document_embeddings.shape)

---

2. Index Embeddings with FAISS

Now, let's create a FAISS index and store the embeddings.

import faiss

# Get embedding dimension
embedding_dimension = document_embeddings.shape[1]

# Create FAISS Index (L2 or cosine similarity search)
index = faiss.IndexFlatL2(embedding_dimension)  # L2 distance (Euclidean distance)
index.add(document_embeddings)  # Add embeddings to the index

print(f"Number of Documents in Index: {index.ntotal}")

---

3. Perform Semantic Search

Now let's search for the most similar documents.

# User Query
query = "Future of AI technology"

# Encode query
query_embedding = model.encode([query])

# Search in the index
k = 3  # Top 3 matches
distances, indices = index.search(query_embedding, k)

print(f"Query: {query}\n")

# Show results
for i in range(k):
    print(f"{documents[indices[0][i]]} (Distance: {distances[0][i]:.4f})")

---

Output Example

Query: Future of AI technology

Artificial Intelligence is the future. (Distance: 0.3105)
Machine learning powers AI. (Distance: 0.3489)
Deep learning improves neural networks. (Distance: 0.3702)

---

Why FAISS?

Feature	Traditional Search	FAISS
Speed	Slow	⚡ Fast
Scale	Small datasets	Large datasets (Millions of documents)
Efficiency	Keyword-based	Vector-based

---

Bonus Tip 🚀

If you're working with cosine similarity instead of Euclidean distance, you can normalize embeddings like this before adding them to FAISS:

faiss.normalize_L2(document_embeddings)

---

Conclusion

FAISS + Sentence Transformers is the best combination for fast, large-scale semantic search.