Почему моделям нужны числа, а не слова

A model cannot do math on the word cat. Every algorithm under the hood only understands numbers, so text must be converted first.

Your goal is to turn each document into a row of numbers, a vector, that captures what the text contains.

You cannot say cat is greater than dog. Words are categories, not quantities, so raw text has no numeric meaning to compute on.

This conversion from words into number arrays is called vectorization. It is the bridge between language and machine learning.

The easiest vector is just how many times each word appears. This count-based idea is the heart of bag-of-words. 🛍️

It is a bag because order is thrown away. You keep which words appear and how often, but not the sequence they came in.

Imagine two reviews. We can score each by counting good and bad to get a simple numeric representation of its tone.

Every document becomes a vector of the same length, one slot per known word, so a model can compare rows directly.

If a word never appears in a document, its slot is simply zero. Most slots end up zero, which makes these vectors sparse.

Once text is numeric, every classic tool works: distance, similarity, and classifiers all operate on these vectors.

Counts ignore grammar and word order, so bag-of-words is a rough but surprisingly strong baseline for many tasks.

Why must text be converted before modeling?

You saw why models need vectors, met bag-of-words, and learned that word counts turn documents into numbers a model can read. 🎉