Update: Organisciak, Teevan, Dumais, Miller, and Kalai received the Notable Paper Award at HCOMP 2014 for the following research paper.
The New Scientist recently featured research on crowdsourcing conducted by doctoral candidate Peter Organisciak. The article, “Online crowd can guess what you want to watch or buy,” highlighted research conducted by Organisciak and a team of researchers from Microsoft Research and MIT including Jaime Teevan (Microsoft Research); Susan Dumais (Microsoft Research); Robert Miller (MIT CSAIL); Adam Kalai (Microsoft Research).
Currently, most online prediction models use algorithms that need a lot of data to make suggestions to a user. In certain instances, however, there isn’t enough data to make those algorithms work. Organisciak and the team turned to crowdsourcing by online workers to make consumer recommendations and found that these “human recommendation engines” were very successful in predicting a user’s tastes.
“When asking for people's input, the goal is often to reach a consensus or popular opinion. But with this work we explored online crowds for a different type of problem, where the right answer depends on who is asking,” said Organisciak. “In personal contexts, such as apartment hunting or travel planning, it is difficult to get help online because your preferences are highly particular.”
Organisciak and the team gauged the opinions of online workers who were very similar to their user—a group of one-time personal assistants who had similar tastes they deemed “taste matchers.” They also solicited opinions from workers who only had a small amount of information with which to make a recommendation, known as “taste grokkers.”
“We found that some people are remarkably adept at understanding your needs regardless of how similar they are. In fact, these taste grokkers can produce recommendations as good or even better than a well-matched worker,” said Organisciak.
Consumers have become used to relying on the suggestions made by large online companies such as Amazon and because personalization is such a key component of a successful online business, Organisciak and the team are hopeful that crowdsourcing can bring success to smaller online companies as well.
“Older systems have a competitive edge in recommendation because they hold so much data. Our approach can give a boost to new businesses trying to compete, helping the next Amazon or Netflix as they are starting out,” Organisciak said.
The team's paper, "A Crowd of Your Own: Crowdsourcing for On-Demand Personalization,” will be presented at the Conference on Human Computation & Crowdsourcing (HCOMP 2014) in early November.