Data-driven chatbot design uses a large database of human conversations to find an input or phrase and reuse responses that people have given before. It replaces much of the tedious pattern-matching work with statistical matching, but exact character sequences remain fragile when users introduce typos, new wording, or unfamiliar expressions. section 1.5.3
Its practical strengths and weaknesses follow from how language is represented:
- Reuse observed responses for familiar utterances instead of hand-coding every phrase. section 1.5.3
- Handle variation by comparing meaning rather than relying only on exact character matches. section 1.5.3
- Treat character-based distance cautiously: good and okay differ greatly as strings, while bad and bag differ slightly despite different meanings. section 1.5.3
Greeting recognition
A chatbot can search a conversation database for a greeting such as “Good morning” and reuse a response previously paired with it. A typo or novel phrasing may defeat exact lookup, motivating better methods for extracting meaning from text. section 1.5.3
