| Labeled datasets for the Standard Based Impact Classification method extracted from approx. 230 CSR GRI reports (150 International companies, 2017-2021 period). | |
| - labels_ipnit_paragraphs 22k records | |
| - labels_ipnit_sentences 75k records | |
| - labels_iporit_paragraphs 57k recors | |
| - labels_iporit_sentences 193k records | |
| Automatic labeling | |
| *ipnit* stands for "index-page **AND** in-text" criteria of label identification. *iporit* stands for "index-page **OR** in-text" criteria of label identification. | |
| index-page means the algorithm searches for index page within the pdf file, and extracts page numbers from there. in-text means the algorithm searching for the label using regex on each page of the report. | |
| *ipnit* represents more strict condition for considering the text as labeled, as it needs both of above methods to return true value. *iporit* is a more relaxed version, considering text as labeled when either of the two returns true value. | |
| No sifnificant increase in the accuracy of the prediction model was observed when using *ipnit* compared to *iporit*. | |