This dataset has an accompanying paper "Introducing a novel dataset for product matching: A new challenge for matching systems" that is accepted at The 3rd International Conference on Computers and Automation (CompAuto 2023) and will be published in IEEE Xplore.
The structure of the dataset is as follows: Each data point consists of a pair products and a binary label that indicates if these two product refer to the same real-world entity.
It consists of four subsets that differ in size and class distribution:
| Dataset |Data points | Negative | Positive | Imbalance Ratio |
|---|---:|---:|---:|---:|
| Full | 960,532| 665,831 | 294,701 | 2.3 |
| L | 243,954| 199,749 | 44,205 | 4.5 |
| M |66,556 | 59,925 | 6,631 | 9.0 |
| S | 18,973 |17,978 | 995 | 18.1 |
The test set consists of 5,000 manually checked data points and is shared across all four subsets.
