The Machine Learning “Advent Calendar” Day 22: Embeddings in Excel

Text embeddings are a core concept in deep learning for representing non-numerical text data as numerical vectors, contrasting with the manual feature engineering of classic machine learning. Two primary methods for creating embeddings are explored: a supervised approach for a specific task like sentiment analysis, and an unsupervised approach like Word2Vec which learns from word co-occurrence patterns. The article uses Excel to provide a practical demonstration of how these embeddings work, from building a simple sentiment classifier to exploring word similarities. This highlights how supervised methods capture task-specific meaning, while unsupervised methods capture distributional similarity based on context.