Vector_Embedding.py

https://github.com/rhyslwells/ML_Tools/blob/main/Explorations/Build/NLP/Vector_Embedding.py

Explanation of the Script

1. Vocabulary and Embedding Layer:
   • Terms are mapped to indices using a dictionary.
   • The embedding layer learns continuous vector representations for these terms.
2. Cosine Similarity:
   • The cosine similarity function measures how similar two terms are in the embedding space. Higher values indicate closer relationships.
3. Visualization:
   • Embeddings are plotted in a 2D space to show semantic relationships. Terms with similar meanings (e.g., “king” and “queen”) are expected to cluster together.
4. t-SNE for Dimensionality Reduction:
   • If the embedding dimension is higher than 2, t-SNE can reduce it to 2D for visualization while preserving semantic relationships.

Outputs

1. Cosine Similarities:
   • Pairwise similarity scores between terms to quantify their semantic closeness.
2. Visualization:
   • A scatter plot showing the positions of terms in the embedding space.