Cosine Similarity

Semantic search converts text into high-dimensional vectors and finds the most relevant conclusions by computing cosine similarity, allowing queries phrased in different ways to match semantically equivalent results.

余弦相似度是衡量两个向量方向接近程度的指标，值域在-1到1之间，越接近1表示语义越相似

Cosine similarity has a value range of [-1, 1]; in semantic retrieval, attention is typically focused only on the positively correlated range of 0 to 1, with values closer to 1 indicating greater semantic similarity.

Cosine similarity only cares about vector direction and not length, with values strictly falling in the range [-1, 1].

When feature vectors are L2-normalized, cosine similarity is equivalent to the dot product operation.

Text is encoded by an Embedding model into floating-point vectors of 768 to 3072 dimensions, stored in a vector database, and similarity retrieval is completed at millisecond speed via approximate nearest neighbor algorithms (HNSW, IVF, etc.).

Cosine similarity measures the cosine of the angle between two vectors, with the formula cos(θ) = (A · B) / (||A|| × ||B||).

Entailment relations are inherently directional and fine-grained, making them better suited to concept-level alignment judgments than global measures such as cosine similarity.

A RAG system first splits documents and converts them into vector embeddings, then retrieves via measures such as cosine similarity; similarity measures semantic closeness, not factual correctness.

After projecting three competitor copy samples into 1536-dimensional space using text-embedding-3-large, cosine similarity typically falls between 0.91 and 0.94.

Contrastive activation probing uses cosine similarity to measure the directional deviation of intermediate-layer activations between the base model and the fine-tuned model under the same probe prompts, with lower similarity indicating greater drift.

In the Little Kiln Support Copilot tests, relevant historical cases scored above 0.63 in cosine similarity and noise cases below 0.58, so the author raised the retrieval threshold from 0.5 to 0.6.

ItemCF's cosine similarity formula is sim(i,j) = C[i][j] / sqrt(N[i] * N[j]), and the denominator normalization naturally downweights popular items.

Vectors are normalized when building the index, so that cosine similarity at runtime reduces to a single dot product.

Vector search maps queries and documents into dense vectors in a high-dimensional space, measuring semantic closeness via cosine similarity or dot product.

On unit vectors, the squared L2 distance ranks identically to cosine similarity, because |a - b|² = 2 - 2·cos(a, b).

At retrieval time, the user's question is likewise encoded into a vector, and the system uses measures such as cosine similarity to find the top-K chunks closest to the question vector.

In the experiments, the average cosine similarity between old facts and their replacements is 0.816, versus a corpus baseline of 0.372, with one pair reaching 1.000; after removing metadata filtering, the stale version becomes the nearest neighbor in 6 of 9 cases, with a median rank of 1.

Cosine similarity measures how close two vectors are in direction, with a value range of -1 to 1; the closer the value is to 1, the more similar the semantics, and because it is insensitive to vector magnitude and stable to compute, it has become the most mainstream choice in RAG scenarios.

Embedding models are deep neural networks that map text into a fixed-dimensional dense vector space (such as 768 dimensions or 1536 dimensions); cosine similarity is commonly used to measure semantic closeness, with a value range from -1 to 1.