"Safer Chatbots offers 3 different mechanisms, depending on the level of technical sophistication of your chatbot, as outlined in the Safer Chatbots explainer video. This guide covers the second highest level of sophistication, providing high-risk keyword based recognition and response. Keyword-based recognition is, of course, an imperfect solution, lacking the ability to detect more nuanced disclosures of harm, but we believe it is better than no detection at all. The approach in this guide has been piloted by UNICEF and WHO chatbots in 3 countries - U-Report Tanzania and Uganda, and HealthBuddy Bulgaria, as well as in external chatbots such as Big Sis by Girl Effect, and Vibe Check by the Prince’s Trust International.
This guide was developed for any chatbot creator who uses RapidPro or Textit to communicate with their users via SMS, Facebook Messenger, Telegram or WhatsApp (although the approach could easily be adapted for other chatbot platforms). Your chatbot does not need a Natural Language Processing (NLP) engine, although you can still benefit from these guidelines if you have. You do not need to be an engineer or AI expert to use this guide, but a decent understanding of RapidPro/Textit by at least one team member is important." (Who is this guide for? Page 3)