Using machine learning to assess rape reports: Sentiment analysis detection of officers' “signaling” about victims' credibility

Purpose: The first of two articles from a larger study whose aim was to teach a computer to detect innuendo (or signaling) about a victim's credibility in incident reports of rape. This study explored the degree of sentiment and subjectivity in the reports and whether these predicted case progression and outcomes. Methods: We employed machine learning, specifically sentiment analysis to assess sentiment (opinion) and subjectivity of textual content. The sample consists of 5638 incident reports of rape with a sexual assault kit from a U.S., urban Midwestern jurisdiction. Results: Sentiment was detected, tended to skew near neutral/slightly negative and more subjective, and predicted case progression and outcomes, but was not quite what was expected. Conclusions: We taught a computer to detect signaling via tone that predicted case progression and outcomes. Findings indicate that the cases recommended for prosecution were longer and had positive sentiment and positive subjectivity. Cases not recommended for prosecution were shorter with more neutral statements of “fact” or observations. Implications and recommendations for improved, less biased report writing are provided.