BAIL657C - Generative AI-Lab program 3

 

Program 3

Aim:

Train a custom Word2Vec model on a small dataset. Train embeddings on a domain-specific corpus (e.g., legal, medical) and analyze how embeddings capture domain-specific semantics.

Theory:

Word2Vec is a neural network–based technique used to convert words into dense vector representations (embeddings). These embeddings capture the semantic meaning and relationships between words based on the contexts in which they appear. When Word2Vec is trained on a domain-specific corpus (such as medical or legal texts), the learned embeddings reflect domain-specific terminology and relationships.

Program:

 

This section imports the necessary libraries:

• NLTK – Used for text preprocessing and tokenization.
• Word2Vec (gensim) – Used to train the word embedding model.
• PCA (scikit-learn) – Used to reduce high-dimensional vectors into two dimensions for visualization.
• Matplotlib – Used to plot the word embeddings

 

A small medical corpus is created manually.
The sentences contain medical terms such as:

• diabetes
• glucose
• insulin
• blood sugar
• treatment

Since the dataset belongs to the medical domain, the trained embeddings will learn relationships between medical terms.

 

 

This program demonstrates how to:

1. Prepare a domain-specific text corpus
2. Tokenize the dataset
3. Train a custom Word2Vec embedding model
4. Analyze semantic relationships between words
5. Compute word similarity
6. Visualize embeddings using PCA

The results show that Word2Vec successfully learns meaningful relationships between medical terms, proving that domain-specific corpora produce specialized word embeddings.