Team
math

Cosine Similarity Calculator

Cosine similarity quantifies how aligned two vectors are by computing the cosine of the angle between them. Essential in machine learning, natural language processing, and recommendation systems, it ranges from −1 (opposite directions) to 1 (identical direction). Unlike Euclidean distance, it ignores vector magnitude, focusing purely on orientation. This calculator computes similarity, distance, and the angle for vectors up to 10 dimensions.

Last updated:

Creators Jasmine J. Mah, BSc
7,696 people find this calculator helpful

How to Use This Calculator

Start by selecting your vector dimension—anywhere from 2 to 10 components. Enter each element of vector a and vector b in the corresponding input fields. If your vectors have fewer dimensions than selected, simply pad them with zeros. The calculator instantly displays:

  • Cosine similarity (ranges from −1 to 1)
  • Cosine distance (1 minus similarity)
  • The angle between vectors in degrees

A detailed breakdown of intermediate calculations—dot products, magnitudes, and the final similarity—appears below the results so you can verify each step.

Cosine Similarity Formula

Cosine similarity between two N-dimensional vectors a and b is the dot product divided by the product of their magnitudes. This approach works when you don't know the angle directly.

S_C = (a · b) / (‖a‖ × ‖b‖)

where:

a · b = a₁b₁ + a₂b₂ + ... + aₙbₙ

‖a‖ = √(a₁² + a₂² + ... + aₙ²)

‖b‖ = √(b₁² + b₂² + ... + bₙ²)

  • S_C — Cosine similarity (ranges from −1 to 1)
  • a · b — Dot product of vectors a and b
  • ‖a‖ — Magnitude (Euclidean norm) of vector a
  • ‖b‖ — Magnitude (Euclidean norm) of vector b
  • n — Number of dimensions in both vectors

Understanding Cosine Similarity Values

The cosine similarity metric captures directional alignment without regard to scale. A value of 1 indicates vectors pointing in exactly the same direction. A value of 0 means orthogonal vectors (perpendicular at 90°). A value of −1 means vectors point in completely opposite directions.

This property makes cosine similarity invaluable in text analysis—two documents with identical word distributions but different lengths will have high similarity—and in recommendation engines, where direction in feature space matters more than magnitude.

Cosine Distance Explained

Cosine distance is simply the inverse of similarity:

D_C = 1 − S_C

It quantifies dissimilarity on a scale from 0 (identical) to 2 (opposite). However, cosine distance is not a true metric in the mathematical sense because it violates the triangle inequality—the path from vector a to c via vector b may exceed the direct distance from a to c. For clustering or distance-based algorithms, use Euclidean distance if a proper metric is required.

Key Considerations When Using Cosine Similarity

Avoid common pitfalls when interpreting or computing cosine similarity.

  1. Ignores magnitude entirely — Vectors [1, 1] and [10, 10] have identical cosine similarity (both point the same direction). If vector size matters to your analysis, supplement with magnitude checks or use Euclidean distance instead.
  2. Zero vectors cause division errors — A zero vector (all elements zero) has undefined magnitude, making cosine similarity undefined. Always validate input vectors are non-zero before computation.
  3. Negative values signal opposing directions — Don't assume negative cosine similarity means poor match. In domains like sentiment analysis or directional data, negative similarity is meaningful and expected when vectors point opposite ways.
  4. Scale vectors consistently for text — In NLP, normalizing term frequency vectors before computing cosine similarity prevents high-frequency words from dominating. Use TF-IDF or sublinear scaling for robust text comparisons.

Frequently Asked Questions

What practical applications use cosine similarity?

Cosine similarity is fundamental in information retrieval—search engines rank documents by similarity to a query vector. Recommender systems use it to find movies, products, or users with aligned preferences. Natural language processing relies on it for sentiment classification, clustering similar documents, and measuring semantic closeness between word embeddings. It's also used in image recognition to compare feature vectors from neural networks.

Can cosine similarity be negative, and what does it mean?

Yes. When the angle between vectors exceeds 90°, the cosine becomes negative. A cosine similarity of −0.5 indicates vectors point somewhat away from each other (120° angle). A value of −1 represents perfect opposition (180° angle). Negative values don't indicate error—they simply mean vectors have opposing directional components. In applications like sentiment analysis, negative similarity can be a valid finding.

Why use cosine similarity instead of Euclidean distance?

Cosine similarity measures direction only, ignoring magnitude. This is ideal when vector length is uninformative—for instance, two long documents with identical word ratios should be similar even if one is twice as long. Euclidean distance penalizes magnitude differences, making it unsuitable for normalized or relative comparisons. Choose cosine similarity for orientation-based problems; use Euclidean distance when absolute position in space matters.

How do I calculate cosine similarity in Python?

Use NumPy for efficient computation. Import the dot function for the dot product and norm from numpy.linalg for vector magnitude: <code>from numpy import dot; from numpy.linalg import norm; similarity = dot(a, b) / (norm(a) * norm(b))</code>. For text data, scikit-learn's cosine_similarity function handles sparse matrices and multiple vectors at once, avoiding manual loops and improving performance on large datasets.

What range of values can cosine similarity have?

Cosine similarity ranges from −1 to 1. A value of 1 means vectors are perfectly aligned (same direction). A value of 0 means orthogonal vectors (perpendicular, no directional correlation). A value of −1 means opposite directions (antiparallel). Values between these extremes indicate partial alignment. The metric never exceeds these bounds because the cosine function itself is bounded between −1 and 1.

Does vector order matter in cosine similarity?

No. Cosine similarity is symmetric: the similarity of vector a to vector b equals the similarity of vector b to vector a. The dot product and magnitudes are computed identically regardless of order. This symmetry is why it's well-suited to symmetrical relationships like document similarity or image comparison, where there's no inherent direction of comparison.

More math calculators (see all)

Square Diagonal Calculator Clock Angle Calculator Prime Number Calculator Equilateral Triangle Calculator Complex Number to Rectangular Form Calculator Perimeter of a Triangle with Fractions Calculator Cube Calculator Matrix Calculator