Mahal Systems Magellan / DeepMatcher entity matching datasets
Fits record linkage or deduplication agents that need labeled pairs to test on across domains. Borrow its three flavors (structured, textual, dirty) and its 3:1:1 splits as a template for building a matching test set.
What it asks the agent to do
- decide whether two records from different tables refer to the same entity
- match citations, products, restaurants, songs or companies across sources
- handle dirty records where attribute values are moved into the title field
- match text-only records such as company pages and homepages
What to borrow from it
case templatebaselinetaxonomy
Caveats
Classic ML datasets (13 sets from 450 to 112,632 labeled pairs), not an agent benchmark: no tools, no environment, F1 on pairs; most sets date from 2015-2018 and are close to saturated for fine-tuned models; no license is stated on the datasets page.
Details
| Workloads | data analysis |
|---|---|
| Interaction | single turn |
| Environment | static |
| End state checked | No |
| Tools | none |
| Grading | programmatic check |
| Size | Not published |
| Data | real public |
| Personal data | none |
| License | unknown |
| Harness | none |
| Status | frozen |