How To Do Fuzzy String Matching In Rasa

In this article, I will share how to create a custom component in rasa to make entity extraction more robust to typos. More specifically, we will use the fuzzywuzzy library to do fuzzy string matching to autocorrect an entity based on its similarity score. The code to reproduce the bot described in this article can be found here. Suppose the bot is expected to extract entities representing a country from an utterance and normalize them so some canonical form. This can be done with rasa's synonyms feature: Therefore, an utterance like "I am from the united states" will be processed by the NLU pipeline as: However, if the user made a typo e.g.