Meta-Learning for Fast Dialog System Habituation to New Users
Abstract: When a new user comes on board a mobile dialog app, the system needs to quickly learn this user’s preferences in order to provide customized service. Learning speed depends on miscellaneous parameters such as probability smoothing parameters. When a system already has a large number of other users, then we show how meta-learning algorithms can leverage that corpus to find parameter values which improve the learning speed on new users.
1 Motivation
You arrive at the office in the morning and say to your new assistant:
“Get me Joe on the phone, please.” The assistant looks at your address book: “Is that Joe Smith or Joe Brown?” You ...
Get Natural Language Processing and Cognitive Science now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.