How to use the Word Embedding Visualizer
- 1. Add sentencesUse our sample or paste your own, one sentence per line.
- 2. Set the windowChoose how many words either side count as context and how often a word must appear.
- 3. Build vectorsSee the 2-D map, pick a word and read its closest neighbours.
The idea behind embeddings
"You shall know a word by the company it keeps." Words used in similar contexts tend to have similar meanings, and embeddings turn that idea into numbers. This tool does it the classic, transparent way. It counts how often each word appears near each other word, converts the counts to positive pointwise mutual information (PPMI) so that common words like "the" don't dominate, and gives each word a vector made of those scores. For the chart, it squeezes the vectors to two dimensions using their top two singular directions; similarities use the full vectors.
Worked example
The sample has 18 short sentences about royalty, programming languages and pets. With a window of 2, a minimum count of 2 and common words ignored, it keeps 24 words. The closest words to king are palace and queen (both 1.000), because they appear in exactly the same contexts. Python sits nearest to java and javascript (1.000 each), then computer (0.675). Cat lands near garden (0.892), dog (0.829) and home (0.829). On the chart the three topics form separate groups.
Limits
- With a few sentences, results are noisy; add more examples of each topic and they sharpen.
- Modern embeddings (word2vec, GloVe, transformer models) learn from billions of words and capture far more meaning, but they rest on the same idea.
- Up to 80 of the most frequent words are kept so the chart stays readable.
Frequently asked questions
Is this how ChatGPT's embeddings work?
Not exactly. Transformer embeddings are learned by neural networks and depend on the whole sentence. This shows the older count-based method, which makes the core idea easy to see.
Is my text uploaded?
No. All the maths runs in your browser.
Is the word embedding visualizer free?
Yes. You can open the word embedding visualizer and fill in everything without signing up. When you build word vectors for the first time, we ask once for your name and WhatsApp number; after that it, and every other Uncodemy tool, opens straight away.