Amazon NLP & search
Semantic search at Amazon
Moving an internal tool beyond keyword search, for an employee audience of more than 38,000.
PythonLSITF-IDFSVDDash
Semantic search at Amazon
Search query
Text representation
Ensemble ranking
Relevant results
From a business question to a working system.Workflow overview
- employee audience
- 38,000+
- ensemble ranking
- 2 models
The problem
An internal Amazon tool relied on basic keyword search. Improving retrieval meant representing meaning more effectively and ranking results more usefully.
What I built
Developed semantic search using Latent Semantic Indexing and TF-IDF, with SVD for dimensionality reduction.
Tuned model hyperparameters and combined two models in a bagging approach to rank results.
Also built Python Dash applications that made machine learning and NLP tools accessible to non-technical users.
The outcome
Improved search accuracy and efficiency for an internal tool serving more than 38,000 employees.
A summary of the work described on my public profile.
View on LinkedIn