Using Embeddings to Understand the Variance and Evolution of Data Science...
summary
Dynamic exponential family embeddings applied to a million-plus job descriptions over three years use gensim and TensorFlow with heavy custom preprocessing and SpaCy components, producing time-aligned vectors that track skill shifts. Key claims: exponential family embeddings handle categorical and temporal data without stitching, models reveal a roughly 40% drop in MBA demand and a marked decline in PhD requirements for data roles, rising Python/Tableau use, falling Hadoop/Perl, and the custom pipeline successfully surfaces these concrete, quantifiable changes.
AI-written summary · about
chapters (12)
- 0:00Skill mapping with embeddings
- 3:51Job description matching limits
- 5:16Feature extraction vs topic modeling
- 8:53Word embeddings power
- 12:45Pre-trained embeddings pros/cons
- 16:37Pre-trained embeddings challenges
- 19:51Custom language models pipelines
- 23:43Visualizing embeddings t-SNE
- 25:16Tracking skill evolution
- 28:07Evolving data science skills
- 34:18Exponential family embeddings
- 37:52Summary
related talks
31mEmbeddings... Embeddings Everywhere
33mUsing the Amazing World
27mData with Purpose: The Power of Embeddings
35mUsing Embedding Layers to Manage High Cardinality Categorical Data
01h 43mWorkshop: Amazing Embeddings — How Vector Representations Are Applied
34mWord Embeddings for fun and profit in Gensim
16mMachine Learning on Source Code
28mEmbeddings