Designing a Chatbot for Survivors of Sexual Violence: Exploratory Study for Hybrid Approach Combining Rule-based Chatbot and ML-based Chatbot

Abstract

Chatbots are developed using two approaches: rule-based and ML-based. The former has a scripted tree structure consisting of questions and answers, while the latter can understand the intent of a user's question and generate answers from machine learning. Despite these benefits, each method has drawbacks: rule-based chatbots cannot answer slightly different questions with the same meaning in the patterns, and ML-based chatbots have difficulty performing continuous conversations while keeping the conversation context. We propose the concept of a hybrid chatbot that combines a rule-based chatbot and an ML-based chatbot to overcome the limitations of each. We conducted an exploratory study to investigate how a hybrid chatbot could support a survivor of sexual violence. A total of 349 questions asked from the stance of survivors were collected and analyzed. We found that the most frequently asked questions were about the punishment for sexual violence, the police report, and the list of support centers. Also, 30% of the questions omitted contextual information about sexual violence, and 10% of questions consisted of only keywords, not complete sentences. We suggest two design implications for the hybrid chatbot based on the survey's findings: 1) dialog flow for helping survivors understand the abuse experience sufficiently to report the sexual violence to police, and 2) contextualization strategies for connecting dialog flow with questions from the survivors.