Communications of the ACM
Facebook Is Trying to Teach Chatbots How to Chit-Chat
Credit: Getty Images
Researchers at Facebook's FAIR lab have developed Persona-Chat, a dataset of more than 160,000 lines of dialogue taken from workers on Amazon's Mechanical Turk marketplace, to train chatbots to be more lifelike when they interact with humans. To organize the data and give the chatbots more personality, the Mechanical Turk workers were asked to design their own character to guide the dialogue. This involved coming up with five basic biographical statements and using them as topics of conversation.
The data was used to train neural networks used for existing chatbots, and the results were assessed by a different group of Mechanical Turk workers. In each case, they conducted a conversation with the persona-drive bot, and compared it with other chatbots and humans.
Although the persona bot did not score as highly on metrics such as "fluency" and "consistency" as humans, it did outperform the chatbot trained on movie dialogue alone.
From The Verge
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found