Challenges in Modifying Existing Scales for Detecting Harassment in Individual Tweets

Abstract

In an effort to create new sociotechnical tools to combat online harassment, we developed a scale to detect and measure verbal violence within individual tweets. Unfortunately, we found that the scale, based on scales effective at detecting harassment offline, was unreliable for tweets. Here, we begin with information about the development and validation of our scale, then discuss the scale’s shortcomings for detecting harassment in tweets, and explore what we can learn from this scale’s failures. We explore how rarity, context, and individual coder’s differences create challenges for detecting verbal violence in individual tweets. We also examine differences in on- and offline harassment that limit the utility of existing harassment measures for online contexts. We close with a discussion of potential avenues for future work in automated harassment detection.