References to deforestation were identified based on sources published by the Global Canopy Programme, Global Forest Watch, and WWF. For more information about our approach, please see our methodology.
The taxonomy of deforestation was compiled based on on sources published by the Global Canopy Program, Global Forest Watch, and WWF. These sources were chosen because they are internationally recognised authorities on this topic. The taxonomy was compiled by identifying key and common terminology used across sources from these authorities relating to deforestation.
Forest-risk commodities were identified using the definitions provided by the Global Canopy Program and Global Forest Watch. ‘Paper’ as a single term returns false positives, so that entry was amended using ‘paper industry/manufacturing’ and ‘make/manufacture paper’ to avoid false positives.
The taxonomy was adapted to match the 3 hierarchies required for the tool and additions were made to capture acronyms, synonyms, and other grammatical expressions of the concepts within the taxonomy. Learn more about our taxonomy methodology.
- Removed the term paper from the category Deforestation
- Created the term paper industry for the category Deforestation
- Created the term paper manufacturing for the category Deforestation
- Created the term make paper for the category Deforestation
- Created the term manufacture paper for the category Deforestation
- Created the term forest degradation from the category Deforestation
- Created the term illegal logging for the category Deforestation
- Removed the term tropical deforestation from the category Deforestation
- Created the term afforestation in the category Afforestation
- Created the term tree planting in the category Afforestation
- Created the term planting trees in the category Afforestation
- Created the term revegetation in the category Revegetation
- Created the term woodlands in the category Woodlands
- Created the term REDD and REDD+ in the category Deforestation
- Created the term forestry in the category Forests
We measured the performance of this concept by manually labelling a test set and comparing the labels to annotations made by this method. We will continue to update this section as we improve our approach for detecting this concept.
Over time we will use these statistics to evaluate which concepts are better to detect using machine learning models vs linguistic rules.
| Precision | Recall | F1 | Support |
|---|---|---|---|
| 0.63 | 0.61 | 0.62 | 39 |