What Is the Difference Between Vector Mode and Text Mode?
Vector mode takes two lists of numbers and compares them directly as points in space. Text mode lowercases two texts, splits them into words, counts how often each word occurs, and compares those word counts as vectors. Use vector mode for embeddings and measurements; use text mode for a quick word-overlap check.
Vector Mode
You type two comma-separated lists of numbers, for example 1, 2, 3 and 4, 5, 6. The tool computes the dot product and both lengths, then divides:
cosine(A, B) = (A . B) / (||A|| * ||B||)
Both lists must have the same length. The numbers keep their meaning: a value of 12 stays a value of 12, and a difference in position is a difference in the vector. This is the mode to use when the numbers already are features, such as a sentence embedding, a user profile, or a row of sensor readings.
Text Mode
Text mode turns text into numbers first. The tool lowercases both texts, splits them on any run of non-word characters, and counts each word. This is raw term frequency: no inverse document frequency, no stopword list, no stemming, and no word-order information.
It then builds one vocabulary from the words of both texts, sorted alphabetically, and turns each text into a vector of counts in that shared order. Those two count vectors go through the same cosine formula as vector mode.
Worked Example
Text A is "the cat sat on the mat" and text B is "the dog sat on the log". The vocabulary is:
cat, dog, log, mat, on, sat, the
vecA = [1, 0, 0, 1, 1, 1, 2]
vecB = [0, 1, 1, 0, 1, 1, 2]
dot = 6 ||A|| = 2.8284 ||B|| = 2.8284 cosine = 0.7500
0.75 looks high for two sentences about different animals. The reason is "the": it appears twice in each text, so it alone contributes 4 of the 6 dot product, and it adds length to both vectors. Remove that one word and the same two sentences score 0.5000. Frequent function words move the score more than the words that carry meaning.
Which Mode to Use
| Question | Mode |
|---|---|
| Do two embedding vectors point the same way? | Vector |
| Do two short texts share vocabulary? | Text |
| Do the numbers have units or a fixed meaning per position? | Vector |
| Do you need a number you can defend in a report? | Vector (text mode is a rough screen) |
Limits of Text Mode
- Raw counts let one repeated word dominate the vector. A text that says "matrix" ten times is mostly a vector about "matrix".
- No stopword removal: "the", "a" and "of" count as much as technical terms.
- Word order is lost. "dog bites man" and "man bites dog" produce the same vector and the same score.
- No stemming, so "multiply" and "multiplying" are different dimensions.
For anything beyond a quick screen, use vector mode with embeddings from a model. The cosine similarity calculator shows the term-frequency matrix for text mode so you can see exactly which counts produced the score.