Understanding User Perception and Trust when Results from a Dating Abuse Detection Application Are Displayed.

Second Look is a mobile phone application that uses machine learning algorithms to detect digital dating abuse from text messages. An online survey was conducted where participants (N = 202) were provided with three different visualizations of the detection screen (labeled text only, percentage of abusive text messages only, and labeled and percentage of abusive textmessages) to understand a) What is the threshold of abusive text messages that would motivate the user to consider themselves in an abusive relationship? (30%, 50%, or 70% abusive text messages) b) What is the most effective way to visualize the results of the detection classifier that would invoke user trust and encourage them to receive necessary help? c) Which visualizations nudge users to trust machine learningbased classification results? We found the Text only condition to show significant differences across all three research questions and will use these results for future iterations of SecondLook.