Royal Jelly

Royal Jelly, led by researchers Sihem Amer-Yahia and Cong Yu with contributions from the Yahoo! Vibes team within the Strategic Data Solutions group and academic collaborator professor Laks Lakshmanan of the University of British Columbia, provides a more efficient and flexible way of making online recommendations and connections by studying the patterns of users on social network sites. The idea evolved after indicators emerged that people use various implicit networks in addition to declared friendship networks to find things that interest them. For example, a user may look to one group of individuals for hiking trail recommendations, and another group for video game recommendations. “There are so many more people who share interests with you than you’ll find in your explicitly friendship-based social networks,” says Amer-Yahia.

Social content sites such as Flickr and Yahoo! Travel offer users a unique opportunity to organize and share information within communities that have common interests. Another example is del.icio.us – a site that enables its users to tag their favorite URLs, create a network of friends, and subscribe to their friends’ feeds to learn about their most recently tagged URLs. The richness of information with these sites presents an enormous opportunity, as well as challenges, for the design of semantically enriched recommender systems. “Royal Jelly leverages these networks to find topics of interests to users,” says Amer-Yahia. “It also tries to provide explanations to each item it recommends to improve the effectiveness of the recommendation,” adds Yu.

One major feature of Royal Jelly is that it allows recommendation strategies, especially collaborative filtering strategies based on user network definitions, to be produced in a declarative, and therefore more flexible, fashion. For example, if a specific topic of interest (such as travel) is desired in Flickr, Royal Jelly can easily adjust the recommendation strategy to create a travel-oriented user network to make more targeted recommendations. “By contrast, the network definition is rigid in traditional recommender systems,” says Yu.

Royal Jelly is expected to make an impact on many social network sites within Yahoo!. One proof-of-concept implementation is del.icio.us. Through a simple tab-based interface, users of del.icio.us can get recommendations in the form of a list on other bookmarks, users outside their network, and other interesting topics as represented by tags. The lists are organized through a rating hierarchy – each item is rated through a formula that ranks the desirability of that item within the network, with the highest rated item appearing at the top of the list. What appears on the list is determined by analyzing the user’s profile and past behavior, such as their bookmarks, tags, and friends. Using this data, Royal Jelly can predict what might interest the user. For example, if a user has heavily bookmarked sites related to camping, Royal Jelly will make recommendations on camping sites not currently bookmarked by the user, and other users that have similar bookmarks and tags, demonstrating a shared interest in camping.

The idea of making recommendations on not only the content, but also the connections such as people and topics is powerful. “Users are no longer limited to getting recommendations on items such as bookmarked URLs, “says Amer-Yahia. “Instead, they can now explore people and topics as well.”