Extract involved crash humans and their injuries automatically from article text

[digitaldutch]

Currently, the humans and their injuries involved in the crash are determined by volunteers copying the article full text selecting this data manually. Helpers click the Add button and fill in a url to a media page. The roaddanger spider then tries to read the meta tags (JSON-LD, Twitter/X, Open graph, etc).

Any missing data is copied manually. From the title and full text all important data is extracted like:

• All involved humans and their mode of transportation. All transport options can be found on the data export page.
• Their injuries: Dead, injured, unharmed of unknown
• If the human is a child (below 18)
• If the human was intoxicated
• If the human drove away of fled
• If it was a one sided crash (no other humans involved)

Entry screen:

Entering data will be faster if these steps can be automated using an AI language model that reads the text and then automatically selects all involved humans and their characteristics. As roaddanger.org is multilingual, it would be nice if this feature supports multiple languages.

All current crash data (full texts and all meta data like involved humans) can be downloaded in JSON format from this page. This data can be used to train or test the language models.