Here you can contribute to the development of an Artificial Intelligence tool to detect offensive language in descriptions of donations. The quality of this tool depends upon having a large number of examples of both unoffensive descriptions and offensive descriptions of donations. Currently the model is built using over 30,000 examples, but many more are needed.
Contributions on this website are strictly anonymous. This allows contributors to freely provide text as offensive as they think a "bad actor" may maliciously provide.
Here you can see how the AI tool scores your contributions and how that scoring changes as the model is improved with more contributions.
Contributions will be reviewed by the developers before inclusion in the dataset from which the AI model is built. If it might not be self evident why the text is offensive, you can provide an explanation to help ensure inclusion of your examples.
Submit text and sentiment
Enter your contributed text or description of a donation in the "Donation Description" field (limited to 100 characters). If you feel the text is offensive, check the "Offensive" box. If necessary, use the "Explanation" field to tell why the text is offensive. Record the "Session Code" if you will want to return to this site later to see how the AI tool scores your contributions. (You will need that code to do so)
Multiple contributions will improve the quality of the model, although it is best if the repeat contributions are not identical. For example, the non-offensive text contributions of "pillow", "pillows", "large square pillows", "pillows & pillowcases", "Pillow" and "fluffy pillow" will all help to ensure that 'pillow', in the absence of associated offensive text, is scored with a unoffensive sentiment.
English is the language for which apps will be developed using this tool. However, the tool is intended to identify text that is offensive to individuals with English as first/only language or to individuals with English as a second language. As such, offensive terms from all cultures are welcome contributions. The AI tool is intended for scoring charitable donations; language such as "for sale" that indicates a commercial intention should be considered offensive.
Scored Contributions
If your contributions have already been scored, you can enter the Session Code from your entries to generate a list of your entries and their associated scores. The scores will range from 0 to 100, with 100 a score for text certain to offend some individuals.
Allow at least an hour for the the scoring of your contributions. Scores will change with time, and rerunning the scoring a month later will likely result in at least subtle differences in some scores as the tool is rebuilt with more data. The model will generally be rebuilt every 1 - 4 weeks depending upon the volume of contributions received.