Intelligent robot assistants like Siri have changed the way we interact with our mobile devices, and can now make recommendations for us. But the technology has yet to fully adapt to our personal preferences.
University of Washington graduate student Jared Bauer is developing an application that has several of the same functions as Siri, such as listing nearby restaurants. But unlike Siri, it studies text messages to get to know its users.
Bauer calls the program the Tech Relating Intelligent Agent (TRIA). It’s a mobile app that analyzes your chat conversations to build a profile of interests to help recommend restaurants, bars, cafes, etc.
For example, when the user searches for a restaurant or bar, the program uses the information in the his or her profile to filter the results and make recommendations. This allows the system to improve over time as the user interacts with members of his or her social circle.
Here’s how it works. The program employs an approach called term frequency-inverse document frequency to select phrases from chat conversations that it deems important to the user. The application then sends these phrases to the Facebook API, where it attempts to find matching phrases on restaurant or cafe pages. The program can determine if it’s a valid phrase, rather than just a jumble of words, and then adds it to the user’s profile.
The Facebook API also provides a set of “user-curated images and valid metadata,” which TRIA adds to users’ profiles. When they search for recommendations, the metadata is used to filter search results.
TRIA also uses topic modeling to determine the conversation topic of a chat conversation. For example, if you are texting a friend about where to eat, you can opt to share your chat history with TRIA, which can recognize that you want to find a restaurant. It will then determine your location and send your request to the Yelp API, which the program uses to recommend a series of restaurants based on your profile and previously-determined interests.
“By leveraging our relationship with the members of our social circle and [the] socially contextualized information from Facebook, we believe that TRIA will allow us to begin to understand how to design systems that use our personal data to delight us in unexpected ways,” Bauer said.
Bauer and Maria Bezaitis, both senior researchers at Intel’s Interaction and Experience Research Lab (IXR), have worked together to create the app, and have plans to expand the capabilities of the program past just hospitality.
Bauer first got on board with the project last summer as an intern at IXR, which is currently funding the application development. In the beginning, the goal was to explore multimodal interaction for intelligent agents. But after the initial user research, Bauer said that he and his advisor, Glen Anderson, decided to take different approach.
Their research revealed several interesting findings about how technology users want to interact with their devices:
- Keep it together: The participants didn’t want to have to bounce between SMS, text and email just to organize a movie outing with two friends.
- Know my personality: Participants wanted suggestions that were geared toward them, specifically.
- Share information when it’s useful: Humans are great at knowing how much and when to provide information. In contrast, a search engine will always produce the same results to a search, regardless of your need.
These three themes helped Bauer and his colleagues design TRIA, which will soon be part of a field study to help the team better understand the implications of the application.
“We also wanted to design an agent that acted more like a new friend, rather than a digital assistant,” Bauer said. “We decided on the idea of a new friend, because it’s a metaphor that people can relate to right away. Just like a new friend might, the application tries to improve its knowledge of you to suggest activities that you are more likely to enjoy.”
Previously on GeekWire: App of the Week: Google’s updated voice search kicks Siri’s butt on iPhone, iPad