What’s in a Name? – Fuzzy Matching Techniques for Proper Nouns
summary
Demonstration applies edit‑distance libraries, nickname banks and generative retrieval to real event data, comparing six libraries and finding rapidfuzz wins; a synthetic set of ~100 fake students exercises misspellings and nicknames, then an evaluation on 128 real attendees yields 124 of 128 matched with zero false matches. Failures cluster on Southeast Asian name reorderings and some nicknames. A mapping tool and a RAG chatbot using Azure OpenAI GPT5 runs slower but recovers a few cases; a hybrid pipeline proves practical for tracking EpiPen holders and longitudinal participant journeys.
AI-written summary · about
chapters (17)
- 0:00Welcome
- 1:46Girls Programming Network data dilemma
- 3:31School name matching challenges
- 5:19Initialisms and nicknames
- 7:03Generative AI for school names
- 8:07Fuzzy matching human names
- 10:54Fuzzy matching libraries
- 13:47Test results
- 15:33Real student data matching
- 17:00Missing last names
- 18:24Unmatched cases
- 19:08Generative AI with RAG
- 20:57AI vs algorithmic results
- 21:41Future longitudinal tracking
- 23:45Scaling GPN data
- 24:06Code and closing
- 24:28Q&A
related talks
40mBalancing Privacy & Utility: Efficient PII Detection and Replacement in Textual Data
49mNamed-Entity Recognition and Name Resolution Using Similarity Queries
30mFuzzy Matching at Scale
11mMutating text toward better nonsense?!
33mFuzzy Search Algorithms How and When to Use Them
21mFinding Spam in Short Text Fields
33mBulk Labelling Techniques
44mVector Search Isn’t Enough