Handling user frustration and hate speech queries

Rasa Open Source
1

What are the best ways to handle the frustration and hate speech in user queries. Here are some of the ways i can think of . Please expand on them or/and add more ideas to it.

  1. train an intent with frustrated queries. This intent can have examples like “you are so stupd”, “dumb bot” and all sorts of general frustration queries from user.
  2. use a sentiment classifier as a custom component that detects negative sentiment.

for option 1 is there a good dataset of hate or frustration queries in the context of chatbots. I have seen many emotions datasets, but they are mostly derived from public tweets or reddit feeds, which may not accurately represent queries that user ask to a chatbot.

1 Like
2

I’m personally skeptical of the idea of inferring sentiment. I’m not convinced it’s possible to build a model that reliably predicts emotions from only text, and it seems quite easy to unintentionally encode prejudices and biases in the model.

If you take the frustrated intent approach, I’d recommend collecting these examples from real users, following conversation-driven development.

1 Like