Annotated tweet data of mixed Wolof-French for detecting obnoxious messages

Automatic detection of obnoxious (abusive) messages on social networks is complex, especially for low-resource languages and in the case of mixed code, such as Wolof-French. This phenomenon is common in Senegalese tweets, but there is a lack of annotated data to facilitate this task. To fill this gap, we created AWOFRO, the first annotated corpus of 3510 tweets in mixed code. We analysed this corpus and validated the annotations using measures such as Cohen's Kappa.