Program Assists the Fashion-Challenged

Feb. 15, 2007 — We can't all be fashionistas. Most of us look in the closet everyday with the same question hanging over our heads: "What am I gonna wear?"

The answer could soon get easier.

Researchers have developed a computer program that applies artificial intelligence techniques (since real intelligence in these matters is sometimes useless) to help the fashion-baffled.

The program could work for getting dressed, as well as to assist online shoppers in search of a range of products, including shoes or sporting equipment.

"Our recommender is based on the user telling us a scenario of how he or she would like to use the product. For example, 'I want to buy a dress to go to my boss's birthday party,'" said Henry Lieberman, a research scientist at the Massachusetts Institute of Technology in Cambridge, MA.

Lieberman leads the project with graduate students Edward Yu-Te Shen and Francis Lam.

According to Lieberman, his team's scenario-based recommender is different from conventional programs already in use by online stores, such as Amazon.com. These programs suggest additional purchases based on the buyer's previous purchases or on items bought by other customers who have similar taste.

But these programs work best for customers with a substantial buying history and don't work at all if the customer doesn't know exactly what he wants.

Lieberman's program, on the other hand, lets a person pose a scenario, and then spits out a recommendation.

The style-confused person uses a website database application to enter the scenario immediately or he can supply descriptions and photos about himself and the clothes he already owns. He can use brand names labels (Nike, Armani, Calvin Klein, etc.) and annotate any piece with comments ("These jeans make me look sexy").

And because it's a website application, he can give friends in his social network permission to view the clothes and provide commentary.

The application uses an artificial intelligence program based on 800,000 common sense sentences (contributed by a network of volunteers) that describe a broad range of subjects concerning people and everyday life. The so-called knowledge database contains information that, for example, a wedding may take place in a church or that a boss may be an important figure.

The program uses the knowledge to categorize clothing and situations. For example, a silk T-shirt might be "luxurious," a suit "formal," a jacket "funky," a watch "elegant," a pair of sunglasses "trendy," and a sweater "sporty."

If the person disagrees with a descriptor, he can use a scroll bar to adjust an adjective from "trendy" to "sporty," for example.

After providing a set of descriptions about already owned clothes and accessories, the person types in the scenario ("I'm going to a dinner with my girlfriend's parents").

The program maps the goal to the characteristics of the appropriate clothing and, like a personal Paris Hilton, picks out a red-carpet outfit. The guy can accept or reject ideas, which helps the program refine future recommendations.

"It's very innovative in the way that it takes into consideration the context in which the recommendation should be provided," said Pearl Pu, research scientist and director of Human Computer Interaction Group at the Swiss Institute of Technology in Lausanne.

The main challenges, she said, are being sure that the common sense database has sufficient knowledge to cover all scenarios. Conflicts could arise if the program tries to solve multiple scenarios at once.

User acceptance could also be a hurdle.

"If you involve the user too much and ask too many questions, they go away," she said.

The researchers are looking for companies interested in developing this system further, and estimate that within a year it could be practical for wide consumer use.