"How do users speak?"
To help the language model accurately interpret user input, it's essential to build and train it based on real data reflecting how users talk to the virtual assistant.
- Search all actions linked to a particular service or product. For example, "What do users want to do with their account?"
- Search all forms of an object, including phrases and derived terms, to answer questions such as "What types of cards do users request?"
To find the answer to a question like the one about card types, enter the verbs want and need in the Action field using the "OR" regex (want|need), and enter card in the Object field. This returns all combinations with these words. The Expanded Object column will then display relevant variations such as new debit card, card for a partner, ATM card, etc.
To see a wide range of user inputs related to a specific topic, use the Conversations view and filter by Topic. For example, the topic "Appointments – banker – consultations" will help you discover all the different ways users request a meeting with their banker, including synonyms that may not yet be covered by your language model (e.g., schedule, set up, book, meeting, appointment, sit-down, etc.).