word2vec เรียนรู้ความหมายอย่างไร

How can a program learn word meaning without any dictionary? The famous word2vec method answers this with one clever trick.

The core idea is the distributional hypothesis: a word is defined by the words around it. Similar context means similar meaning.

word2vec slides a small window over text and learns to predict each word from its neighbors. This is the skip-gram training setup.

Given a center word, the model guesses nearby words. Here the window shows what neighbors it tries to predict.

Skip-gram predicts context from a word, while CBOW predicts a word from its context. Both learn vectors; they just look in opposite directions.

The prediction task is just an excuse. What we keep are the hidden weights, which become each word vector after training.

Each wrong guess nudges the vectors a little. Over millions of examples these tiny updates shape meaning into the numbers.

Words appearing in similar contexts get pulled toward each other in space. That is how king and queen end up close together.

Training on every vocabulary word is slow, so word2vec uses negative sampling, learning from a few wrong words instead of all of them.

The genius is that raw text is its own teacher. word2vec is self-supervised, needing no human labels to learn meaning. 🤖

For the first time, simple math on vectors reflected real word relationships. word2vec made embeddings practical for everyone.

What task does word2vec actually train on?

word2vec learns meaning by predicting context, then keeps the weights as vectors. Similar contexts pull words close, all without labels. ✅