Thursday, 15 May 2025

AI Algorithms: Word2Vec

 See the FAQs here

🔍 What is Word2Vec?

Word2Vec is a neural network-based algorithm that learns vector representations (embeddings) of words from a large corpus of text, capturing their semantic and syntactic meaning.


🎯 Goal

To place similar words (in meaning and context) close together in vector space.


🧠 Key Idea

Words that appear in similar contexts have similar meanings — “You shall know a word by the company it keeps.”


🛠️ How It Works

Word2Vec has two main model architectures:

  1. CBOW (Continuous Bag of Words):
    Predicts the current word based on its context (surrounding words).
    Input: Context words → Output: Target word

  2. Skip-Gram:
    Predicts surrounding context words from the current word.
    Input: Target word → Output: Context words
    (Works better with small datasets and infrequent words.)


🧮 Training

  • Uses a shallow neural network with one hidden layer.

  • The model learns to predict probabilities using softmax or approximations like negative sampling or hierarchical softmax.

  • During training, the weight matrix between the input and hidden layer becomes the word embedding matrix.


📦 Input and Output

  • Input: Large corpus of raw text.

  • Output: A vector for each word (e.g., 100-300 dimensions) where semantic relationships are captured (e.g., vector("king") - vector("man") + vector("woman") ≈ vector("queen")).


Why It’s Useful

  • Captures word similarity, analogy, and relationships.

  • Forms the backbone of many NLP models (pre-BERT era).

  • Fast and scalable.

No comments:

Post a Comment

🧠 You Only Laugh Once: Creativity and Humor in Deep Learning Community

It all started with a simple truth: Attention Is All You Need . Or at least, that’s what the transformers keep whispering at every AI confer...