A cleaned version of the Hebrew Sentiment data set published by Amram, A., Ben-David, A., and Tsarfaty, R. (2018).{1}
Original dataset can be found under the omilab Github.
This dataset was prepared and used for the paper AlephBERT:A Hebrew Large Pre-Trained Language Model to Start-off your Hebrew NLP Application With
Since we discovered that there was a data leakage (shared material between test and train) we provide a cleaned version without duplications and with removal of text that is almost identical. (for example 2 samples that say "I love you!" and "I love you!!" considerd identical), with a new split of train-dev-test, instead of the original train-test. We provide in this repository both the new clean data and both the full deduping code.
| Data set | Original Set sizes | Deduped Set sizes - Token | Deduped Set sizes - Morph | Comment |
|---|---|---|---|---|
| Train | 9220 | 5926 | 5932 | |
| Dev | 1026 | 846 | 847 | Originally didn't exist and was part of the train |
| Test | 2561 | 1695 | 1696 |
| Data set | Leakage percentage |
|---|---|
| Train | 24.6% |
| Dev | 3.8% |
| Test | 7.53% |
| Data set | Leakage |
|---|---|
| Test - Train | 1120 (12%) |
| Test + Dev - Train | 1685 (16.22%) |
{1} Representations and Architectures in Neural Sentiment Analysis for Morphologically Rich Languages: A Case Study from Modern Hebrew. In: Proceedings of The 27th International Conference on Computational Linguistics (COLING 2018) Santa Fe, NM, (pp. 2242-2252).