Language is an intricate component of daily life that many people do not think about twice. But for Prof. Claire Cardie, computer science, language is at the center of her research in natural language processing.
NLP is a subfield of artificial intelligence that deals with producing systems that understand and produce language, with the ultimate goal of producing language as well as a human being. While scientists have yet to reach this ultimate objective, there are many systems today that assess language to provide valuable information such as finding fake reviews.
In collaboration with Prof. Jeffrey Hancock, communications, and Myle Ott grad, Cardie examined single words and word pairs associated with true and false reviews of a hotel. She found that the truthful reviews often contain concrete terms about specific details of the hotel, such as “bathroom” and “hallway.”
False reviewers cannot rely on this information in their reviews, so they often discuss why they were at the hotel, rather than information about the hotel itself.
“We can train a system that is able to determine which kinds of words are associated with true reviews and which kinds of words are associated with fake reviews,” Cardie said. “Words like ‘husband’ and ‘business’ were common in fake reviews. Deceptive people have to focus on something aside from the hotel.
Her research garnered much media attention, including a feature piece by The New York Times.
Speech recognition technology is also being applied in the business world, where it can be used to summarize meetings.
“It would be nice if there was a way to share the outcome of a meeting with upper-level managers or people who missed the meeting,” Cardie said.
Working with Lu Wang grad, Cardie is studying methods of providing focused meeting summaries.
“They are focused in the sense that each summary is picking up one dimension of the meeting outcome,” Cardie said. “You can get summaries of action items, decisions made and problems raised during the meeting.”
This research would also allow scientists to examine an opinion piece, such as an editorial, and quickly retrieve important information in a concise way.
“We want to identify where [in the text] someone is expressing an opinion, whether the opinion is positive or negative, who is providing the opinion, and what the opinion is about,” Cardie said.
Systems like these are particularly instrumental in web searches, as they would allow people to type questions into a search engine and automatically find the data they were searching for, as opposed to searching for a series of key words.
“With a lot of good engineering, some of this research is already being commercialized,” Cardie said.
Cardie is the co-founder of the start-up company Appinions, which built on this research.
She is also involved with a hub program at Cornell’s NYC Tech campus. She led a group in developing a curriculum for the hub in collective media which was recently approved and will be ready for students by the next academic year.
Original Author: Nicolas Ramos