Researchers receive Google grant to improve conversational recommender systems
Ever wish Alexa, Siri or Google could just read your mind and tell you which restaurant to pick?
A team of researchers got a $1 million grant over three years from Google to accomplish just that, and UCSB’s William Wang is on this team.
Dr. Wang is the director of UCSB’s Natural Language Processing group and the Center for Responsible Machine Learning. He’s also the Duncan and Suzanne Mellichamp Chair in Artificial Intelligence and Designs and an assistant professor in the Department of Computer Science at UCSB.
He specializes in natural language processing and knowledge representation, and he’s joining UC Santa Cruz’s computer science professor Lise Getoor, who researches machine learning and reasoning under uncertainty.
In this project, he and his team will focus on how to improve the reasoning capabilities of dialogue agents to better understand user requests by using context and other information.
“The question is definitely, ‘How can we understand the intrinsic relationship among all these entities so that we can understand better the user’s preference and generate better answers?’ ” Mr. Wang told the News-Press. “This requires machines to be able to think like humans.”
This concept is measured by the Turing test — a method of inquiry in artificial intelligence for determining whether a computer is capable of thinking like a human being.
Dr. Wang and other computer scientists use this test in the form of crowdsourcing platforms, and they’ve had co-workers participate in conversations with both these kinds of systems and actual humans.
“I ask my co-workers to rate which one they think is the machine at the end of the day,” Dr. Wang said.
He succeeds when his A.I. system confuses his co-workers into believing they were conversing with a real human.
“A few years back, we did one storytelling generation evaluation. We were able to get about 50% of times that we can successfully make people believe the generated text is from humans,” Dr. Wang said.
However, the capabilities with these A.I. systems are limited. With online recommender systems, Dr. Wang said an example of what he and his team are trying to combat is when a user purchases a gift off of Amazon, and then they receive advertisements on social media for that product after already buying it.
One of the reasons for that could be the lack of user feedback, according to Dr. Wang.
“I think right now, the format is pretty passive for users on social media platforms,” he said. “There’s no feedback from the user and not much interaction between the recommended system and the user.”
He explained that a platform like Netflix has automatic user feedback. When a user selects a movie to watch, they’re providing a positive feedback signal to the system. However, with something like restaurant suggestions, unless the user rates the system or provides feedback in some way, it doesn’t have room to improve or learn any more about the user.
“Having the dialog with your virtual system is going to be more and more popular because in the future you could imagine that people can keep more specific constraints on what kinds of restaurants they’re looking for,” Dr. Wang said. “(For example) I’m looking for one with a patio that allows my dog to sit with me on the patio, and one for a party of 10 but also one that has vegetarian food.
“It’s very difficult to convey all that if you’re typing or just clicking on the web … But imagine if you have a sort of system that can talk to you like a real person,” he said. “That can save you a lot of time. You don’t have to call all these restaurants or spend a half an hour on Yelp if your assistant can talk to you and help you understand your preference.”
So while even a machine that thinks like a human may never be able to figure out what you want for dinner, Dr. Wang and his team may be able to figure out a way for your device to get you as close as possible to that perfect Italian dish or juicy burger and fries you’re craving.