Cosine similarity
Cosine similarity measures the directional alignment of two nonzero vectors as the cosine of the angle between them, equal to their dot product divided by the product of their magnitudes.
Natural Language Processing in Action, Second Edition
Cosine similarity measures the angular similarity between two nonzero vectors by dividing their dot product by the product of their magnitudes.
- How it is used
- Uses cosine similarity to rank text matches and compare embeddings across sentences.
- How deep it goes
- Derives the normalized dot product, interprets its range, and implements pairwise comparisons.
- Examples
- Compares an algorithmic-bias question with corpus sentences and plots similarities between embedded passages.
- What you can do after
- Build it from scratch in Python and use normalized embeddings to calculate a pairwise similarity matrix.
Read how Natural Language Processing in Action, Second Edition explains cosine similarity →
Rearchitecting LLMs
Cosine similarity measures the cosine of the angle between two vectors, reflecting their directional alignment independently of magnitude.
- How it is used
- Ranks transformer blocks for task-specific depth pruning by comparing input and output activations.
- How deep it goes
- Explains directional interpretation and implements masked token-level scoring, aggregation, and block selection.
- Examples
- Compares three sentence embeddings and ranks Qwen3-0.6B blocks using WinoGrande and WikiText calibration.
- What you can do after
- Apply it via PyTorch to calculate similarities and select candidate transformer blocks for pruning.
AI-Powered Search
Cosine similarity measures the angular similarity between two vectors by dividing their dot product by the product of their magnitudes.
- How it is used
- Uses cosine similarity to rank query-document matches and compare semantic embeddings.
- How deep it goes
- Explains magnitude normalization and implements query-document scoring and pairwise comparison of normalized embeddings.
- Examples
- Compares apple-juice query and document vectors, then ranks similarities among four sentence embeddings.
- What you can do after
- Apply it via NumPy and sentence-transformers to score query-document vectors and compare normalized embeddings.
Build AI-Enhanced Web Apps
Cosine similarity is a metric that measures the cosine of the angle between two vectors, with larger values indicating greater directional similarity.
- How it is used
- Uses cosine similarity to rank knowledge-base matches and check generated responses against expected outputs.
- How deep it goes
- Defines the vector-angle score and demonstrates comparison loops and threshold assertions without implementing the cosine function.
- Examples
- Compares an IT support password query with stored questions and tests a mocked response about AI’s influence against expected wording.
- What you can do after
- Apply it via string-comparison for a threshold test, or use a cosineSimilarity function to rank stored embeddings.
Read how Build AI-Enhanced Web Apps explains cosine similarity →
Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI
Cosine similarity measures the similarity between nonzero vectors using the cosine of the angle between them.
- How it is used
- Uses cosine similarity to cluster high-dimensional data for diversity sampling and compare training with unlabeled data for representative sampling.
- How deep it goes
- Defines the normalized dot product and describes iterative clustering, while relying on PyTorch’s native function rather than detailing its implementation.
- Examples
- Illustrates angular similarity with stars and compares clusters of news headlines with potentially misleading pixel-based image clusters.
- What you can do after
- Apply it via PyTorch to cluster data and select items for annotation.
Math and Architectures of Deep Learning
A dot product computed on unit-length vectors so that the score is insensitive to vector length.
- How it is used
- Uses normalized vector direction to compare TF-IDF document representations during retrieval.
- How deep it goes
- Explains normalization, angular geometry, and retrieval limitations, then connects the limitation to latent semantic analysis.
- Examples
- Uses TF-IDF document vectors and the d5–d6 gun-versus-violence mismatch to expose direct-overlap limits.
- What you can do after
- Apply cosine similarity to compare weighted document vectors and recognize when direct term matching misses latent semantic relations.
Read how Math and Architectures of Deep Learning explains cosine similarity →
Graph Algorithms for Data Science
Cosine similarity is a measure of the angular resemblance between two vectors, computed from the cosine of the angle between them.
- How it is used
- Uses vector representations to infer nearest-neighbor similarity networks for user segmentation.
- How deep it goes
- Defines angular similarity, score interpretation, normalization implications, and thresholded graph construction without presenting the underlying formula.
- Examples
- Demonstrates user vectors built from activity, degree, graphlet, closeness, and betweenness features.
- What you can do after
- Apply it via Neo4j GDS gds.knn to construct a scored similarity graph for downstream community detection.
Read how Graph Algorithms for Data Science explains cosine similarity →
Generative AI in Action
Cosine similarity measures the cosine of the angle between two vectors and indicates their directional similarity.
- How it is used
- Supports semantic retrieval by comparing query and document embeddings in a retrieval-augmented generation workflow.
- How deep it goes
- Defines directional similarity and its range, compares alternative metrics, and configures cosine distance in a Redis index without a mathematical derivation.
- Examples
- Uses a Longhorn query to retrieve three similar posts and a Redis HNSW schema configured with COSINE.
- What you can do after
- Apply it via Redis to configure an embedding index and retrieve similar posts.
Read how Generative AI in Action explains cosine similarity →
Fight Fraud with Machine Learning
Cosine similarity is a metric that quantifies the directional similarity between two vectors as the cosine of the angle between them.
- How it is used
- Supports duplicate checks and selfie-ID identity verification by comparing face embeddings.
- How deep it goes
- Defines directional vector similarity and works through inference, thresholding, and Siamese-network training with sigmoid and binary cross-entropy.
- Examples
- Demonstrates Einstein selfie pairs, an Einstein–McCarthy mismatch, ID comparisons, and duplicated or pasted-photo fraud cases.
- What you can do after
- Apply it via PyTorch to compare MTCNN-cropped, InceptionResnetV1-generated face embeddings for duplicate and selfie-ID checks.
Read how Fight Fraud with Machine Learning explains cosine similarity →
Build Python Web Apps with Streamlit
Cosine similarity is a metric that compares vectors by measuring the cosine of the angle between them.
- How it is used
- Supports embedding-based retrieval by ranking vectors according to the angle between them.
- How deep it goes
- Defines angle-based similarity and applies it to vector indexes, but does not derive the mathematical formula.
- Examples
- Uses plotted movie vectors and support-article embeddings as concrete similarity-search examples.
- What you can do after
- Operate/configure a Pinecone index using cosine as its vector similarity metric.
Read how Build Python Web Apps with Streamlit explains cosine similarity →
Build an Advanced RAG Application (From Scratch)
Cosine similarity measures the cosine of the angle between two vectors to quantify their directional relatedness.
- How it is used
- Ranks document embeddings against query embeddings within semantic search.
- How deep it goes
- Explains vector angles, score interpretation, cosine distance, and retrieval code, but does not derive the formula mathematically.
- Examples
- Uses pet-related vector diagrams, capital-country embeddings, and a six-sentence machine-learning retrieval example.
- What you can do after
- Apply it via Sentence Transformers and SciPy to rank documents by semantic relatedness.
Read how Build an Advanced RAG Application (From Scratch) explains cosine similarity →
LLM Evaluation and Alignment, The Foundational Ideas
Cosine similarity measures the orientational similarity of two non-zero vectors by calculating the cosine of the angle between them, with values from -1 to 1.
- How it is used
- Evaluates semantic overlap between candidate and reference texts when exact lexical matching misses paraphrases.
- How deep it goes
- Defines orientation-based scoring, demonstrates embedding construction, and stops short of solving contextual meaning and word order.
- Examples
- Uses attorney–lawyer paraphrases, weather–freezing paraphrases, and committee-approval sentences to illustrate semantic closeness.
- What you can do after
- Apply it via PyTorch to compare embedding vectors for semantically related passages.
Read how LLM Evaluation and Alignment, The Foundational Ideas explains cosine similarity →
AI Agents in Action, Second Edition
Cosine similarity measures the similarity of two nonzero vectors by the cosine of the angle between them, independently of their magnitudes.
- How it is used
- Supports vector similarity search for retrieving documents without making vector magnitude the main signal.
- How deep it goes
- Defines angular similarity, compares distance scales and alternatives, and demonstrates calculation and retrieval; it does not derive the formula.
- Examples
- Compares sentence vectors and asks for rankings on “blue” and “breakfast” using TF-IDF and embeddings.
- What you can do after
- Apply it via scikit-learn to compare document vectors and inspect similarity scores.
Read how AI Agents in Action, Second Edition explains cosine similarity →
AI Agents in Action
Cosine similarity is a measure of the angular similarity between two vectors, commonly used to compare the meanings represented by embeddings.
- How it is used
- Supports document comparison, semantic retrieval, and selection of a consistent generated answer.
- How deep it goes
- Defines angular similarity and distance, then demonstrates document similarity matrices and answer selection using embeddings.
- Examples
- Compares a set of documents and selects a generated answer closest to the mean embedding.
- What you can do after
- Apply it via scikit-learn to compare documents and interpret cosine distance scores in retrieval.
LLM Customization and Fine-Tuning
A measure of the angle-based similarity between two vectors, often used to compare weighted text representations.
- How it is used
- Supports semantic retrieval, distribution-level drift detection, and screening for training-test overlap.
- How deep it goes
- Defines normalized dot products and works through centroid drift and pairwise overlap screening without a long mathematical derivation.
- Examples
- Compares query and chunk embeddings, reference and production centroids, and teacher-generated training examples against held-out tests.
- What you can do after
- Apply it via scikit-learn to screen training examples for excessive TF-IDF overlap with a held-out test set.
Read how LLM Customization and Fine-Tuning explains cosine similarity →
AI Recommender Systems
Cosine similarity is a measure of the angular similarity between two vectors, typically ranging from zero to one for nonnegative vectors.
- How it is used
- Supports item retrieval by ranking vectors that point in similar directions across behavioral, metadata, and embedding-based recommendation pipelines.
- How deep it goes
- Explains the angular formula, sparse and content-based implementations, top-k retrieval, metadata limitations, and normalized-vector optimization.
- Examples
- Uses movie watch histories, title-and-genre TF-IDF vectors, Drama-only matches, and embedding comparisons as concrete examples.
- What you can do after
- Build it from scratch with sparse matrices or content vectors, rank top-k candidates, and substitute dot products for normalized embeddings.
Read how AI Recommender Systems explains cosine similarity →
Machine Learning Platform Engineering
Cosine similarity measures the angular alignment of two vectors independently of their magnitudes.
- How it is used
- Uses cosine similarity to retrieve semantically relevant chunks in text-focused FAISS applications.
- How deep it goes
- Defines the semantic-direction rationale and operational trade-offs but stops short of mathematical derivation or implementation details.
- Examples
- Demonstrates similarity search over indexed documentation using the query “How do I rotate encryption keys?” and three retrieved results.
- What you can do after
- Evaluate retrieval quality by selecting a distance metric and tuning result counts and similarity thresholds.
Read how Machine Learning Platform Engineering explains cosine similarity →
Retrieval Augmented Generation, The Foundational Ideas
Cosine similarity is a metric that measures the angular similarity between two vectors.
- How it is used
- Ranks semantically related documents when exact keyword overlap is insufficient.
- How deep it goes
- Explains normalization, dot products, geometric values, and retrieval failures, but stops short of detailed mathematical derivation.
- Examples
- Compares automobile-maintenance retrieval scores and illustrates normalized vectors with differing lengths.
- What you can do after
- Evaluate semantic-ranking behavior and diagnose errors caused by magnitude removal, averaging, antonyms, or lost sentence structure.
Read how Retrieval Augmented Generation, The Foundational Ideas explains cosine similarity →
Domain-Specific Small Language Models
A similarity measure that compares vectors by the cosine of the angle between them.
- How it is used
- Ranks query-document embeddings for RAG retrieval and helps retrieve relevant past interactions in conversational memory.
- How deep it goes
- Defines its role in measuring embedding relationships, but provides no formula or implementation details for the similarity calculation.
- Examples
- Uses query-document embeddings in RAG and past-interaction search in Memoripy.
- What you can do after
- Apply it in embedding-based retrieval to select relevant documents or past interactions.
Read how Domain-Specific Small Language Models explains cosine similarity →
Graph Neural Networks in Action
Cosine similarity is a measure of similarity between vectors based on the cosine of the angle between them.
- How it is used
- Builds item-similarity workflows from graph-model embeddings for recommendation and product-bundle generation.
- How deep it goes
- Defines cosine similarity operationally through embedding comparison, matrix construction, and ranked nearest-neighbor retrieval.
- Examples
- Demonstrates a Marceline product bundle produced by ranking similarities for node 123.
- What you can do after
- Apply it via SciKit Learn and NumPy to rank products nearest to a chosen graph node.
Read how Graph Neural Networks in Action explains cosine similarity →
Build an AI Agent (From Scratch)
Cosine similarity is a measure of the angular similarity between two vectors, commonly computed as their dot product divided by the product of their magnitudes.
- How it is used
- Ranks semantically relevant documents or chunks during vector search and retrieval-augmented generation.
- How deep it goes
- Defines angular similarity and demonstrates implementation with embeddings, but does not derive the underlying vector formula.
- Examples
- Compares cat, kitten, and dog sentences and ranks documents about artificial intelligence, machine learning, Python, and pets.
- What you can do after
- Apply it via scikit-learn to rank embedded queries against stored document or chunk vectors.
Read how Build an AI Agent (From Scratch) explains cosine similarity →
Enterprise RAG
Cosine similarity measures the angular similarity between two vectors by computing the cosine of the angle between them.
- How it is used
- Configures vector retrieval to rank embedding vectors by directional semantic similarity.
- How deep it goes
- Defines angular similarity and applies it to index configuration, stopping short of mathematical derivations or implementation code.
- Examples
- Uses a 1,536-dimensional Azure AI Search vector field and compares Exhaustive KNN with HNSW.
- What you can do after
- Operate/configure Azure AI Search vector retrieval with cosine similarity and an appropriate nearest-neighbor strategy.
MongoDB 8.0 in Action, Third Edition
Cosine similarity is a measure of the cosine of the angle between two vectors, indicating their directional similarity.
- How it is used
- Supports semantic search, recommendation systems, question answering, fraud detection, and retrieval-augmented chatbot workflows.
- How deep it goes
- Explains directional comparison and index configuration, then stops short of deriving the cosine formula or implementing the search algorithm.
- Examples
- Uses semantic document retrieval and MongoDB Atlas index definitions with 1,536-dimensional embedding fields.
- What you can do after
- Operate/configure it through Atlas CLI and mongosh, creating and verifying a cosine-based vector search index.
Read how MongoDB 8.0 in Action, Third Edition explains cosine similarity →
Transformers in Action
Cosine similarity is a metric that measures the directional similarity between two vectors using the cosine of the angle between them.
- How it is used
- Supports semantic comparison, vector retrieval, chatbot context resolution, and response-quality estimation.
- How deep it goes
- Defines cosine similarity as an embedding comparison measure and demonstrates computation, but does not derive its mathematical formula.
- Examples
- Uses contextual word meanings, sentence embeddings, query-document pairs, and a printed similarity matrix.
- What you can do after
- Apply it via SentenceTransformer and util.cos_sim to compare encoded queries with documents.
Read how Transformers in Action explains cosine similarity →
Building Reliable AI Systems
Cosine similarity is a measure of the angular similarity between two vectors, calculated from the cosine of the angle between them.
- How it is used
- Ranks text embeddings by semantic proximity in search and retrieval, including health-care policy lookup.
- How deep it goes
- Defines angular comparison of vector direction, notes scale invariance, and contrasts dot product’s sensitivity to magnitude.
- Examples
- Uses paraphrased therapy questions and a Pinecone index configured with the cosine metric.
- What you can do after
- Apply it via Pinecone by selecting cosine as the vector index metric for text embeddings.
Read how Building Reliable AI Systems explains cosine similarity →
Also covered in
- Essential GraphRAG
- Introduction to Generative AI, Second Edition
- Knowledge Graphs and LLMs in Action
- Build a Text-to-Image Generator (from Scratch)
- Data Without Labels
- Data Pipelines with Apache Airflow, Second Edition
- Grokking Algorithms, Second Edition
- Building LLM Applications with DSPy
- Financial AI in Practice
- Machine Learning Algorithms in Depth
- Designing AI SystemsIn early access (MEAP)
- Context EngineeringIn early access (MEAP)
What to know first
- embeddingAn embedding is a numerical vector representation of an item in a feature space, encoding selected properties or relationships for computational tasks such as comparison or retrieval.
- vector spaceA vector space is a set equipped with vector addition and scalar multiplication over a field that satisfy the vector-space axioms.
- similarity searchSimilarity search is a retrieval technique that finds or ranks items by their similarity to a query, often by comparing feature or vector representations under a similarity or distance measure.
- language modelA language model is a computational model that represents patterns in linguistic sequences to assign probabilities, predict, generate, or evaluate language.
Related concepts
- cosine distanceequivalent to · Cosine distance is a measure of directional dissimilarity between nonzero vectors, commonly defined as one minus their cosine similarity.
- Euclidean distancecontrasts with · Euclidean distance is the straight-line distance between two points, calculated as the square root of the sum of squared differences between corresponding coordinates.
- drift scoreused by · A drift score is a quantitative measure of the difference between data distributions being compared for drift.
- K-means clusteringused by · K-means clustering is an algorithm that partitions data into a specified number of clusters by iteratively assigning observations to the nearest centroid and updating each centroid to the mean of its assigned observations.
- similarity searchused by · Similarity search is a retrieval technique that finds or ranks items by their similarity to a query, often by comparing feature or vector representations under a similarity or distance measure.
- embeddingmeasures · An embedding is a numerical vector representation of an item in a feature space, encoding selected properties or relationships for computational tasks such as comparison or retrieval.
Questions about cosine similarity
What is cosine similarity?
Cosine similarity measures the directional alignment of two nonzero vectors as the cosine of the angle between them, equal to their dot product divided by the product of their magnitudes.
Which Manning books explain cosine similarity?
Natural Language Processing in Action, Second Edition, Rearchitecting LLMs, AI-Powered Search, Build AI-Enhanced Web Apps, Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI, Math and Architectures of Deep Learning, Graph Algorithms for Data Science, Generative AI in Action and 17 more.
What should I know before learning cosine similarity?
Manning books that teach cosine similarity usually introduce embedding, vector space, similarity search and language model first.
What is the difference between cosine similarity and Euclidean distance?
Cosine similarity measures the directional alignment of two nonzero vectors as the cosine of the angle between them, equal to their dot product divided by the product of their magnitudes. Euclidean distance is the straight-line distance between two points, calculated as the square root of the sum of squared differences between corresponding coordinates.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.
























