Introducing STEER: A Steering Detection Model for Voice Assistants
Voice assistants like Siri or Alexa are super helpful, but sometimes they don’t quite understand what you’re asking. That’s where steering comes in. Steering is when you ask a follow-up question to clarify what you’re talking about.
At *Company Name, they’ve created a new model called STEER, that can detect when a follow-up question is a user trying to steer the conversation. This is really important because it helps the voice assistant understand what the user is really asking for.
Creating STEER wasn’t easy. They had to figure out how to train the model without a lot of data to start with. But they came up with some clever rules to help them sample data and train the model. And guess what? It worked! STEER can accurately detect steering intent with over 95% accuracy.
But they didn’t stop there. They also developed STEER+, which uses a fancy thing called a semantic parse tree to give the model even more context. This helps reduce errors, especially when dealing with things like names or other tricky words.
And they didn’t just stop at making the model. They also did a study to show how much better the user experience is when voice assistants can handle steering. Turns out, it’s a big improvement!
So, thanks to *Company Name, voice assistants just got a lot smarter with STEER and STEER+. That means they can understand you better and help you out more. Cool, right?