本周日即将进行的
推荐系统论坛，我们成功邀请到了 Netflix Prize 冠军队成员
Yehuda Koren 先生参会，进行主题演讲。Koren 先生在推荐系统领域有着卓越成就，其在 Netflix Prize 参赛过程中公布的一系列论文与成果，对推动推荐系统领域的发展做出了重要的贡献。
国内最大的IT社区 CSDN，将于会议当天对 Koren 先生进行一次专访，特此向广大朋友征集问题。机会难得，请大家踊跃贡献给力的好问题。
豆瓣王守崑同学先抛了两块砖。
1. As the rapid growth of SNS websites worldwide, friends-recommendations have established a solid influences on users’s choices comparing to algorithms-recommendatins. Say, on media products such as books, movie, and music, people trust their friends a lot more rather than machines. What do you think about this phenomenon, and are there any evidence or measures to quantitively evaluate it?
2. People visit more often on facebook, twitter and other social platforms, and in these websites information is private, transient, and contextual, which is very hard for Google to index, ranking, and present relevant results in response to search queries. Are there any chance for recommender system to be involved and provide better performance to both search engines and social platforms?
中英文提问均可，请直接在本文后面留言提问。期待你的问题！
附，Koren 先生简介及本次 Topic 内容介绍。
Yehuda Koren is a Senior Researcher at Yahoo! Research, Haifa. Prior to this, he was a member of AT&T Labs-Research. He obtained his Ph.D. in Computer Science from The Weizmann Institute in 2003. His thesis topic was algorithms for drawing, clustering and visualizing large graphs. He was awarded best paper award at INFOVIS 2005 for the work on directed graph layout through constrained energy minimization. More recently he won KDD 2009 best research paper award for his work on collaborative filtering with temporal dynamics. Yehuda is a member of the teams that won the grand prize and the two progress prizes in the Netflix Prize competition.
Web-Scale Recommendation Systems
Web retailers and content providers offer a huge selection of products, with unprecedented opportunities to meet a variety of special needs and tastes. This opened an opportunity for recommender systems, which provide personalized product recommendations that suit a user’s taste. The collaborative filtering approach to recommender systems predicts user preferences for products or services by learning past user-item relationships. In this talk I will survey some of the recent advances made in the field.
Successful collaborative filtering systems need to address multiple challenges. The designer needs to model the very elusive human taste appropriately, and deal with the fact that humans constantly change and redefine their preferences. Furthermore, balancing between a detailed modeling of heavy users and an adequate modeling of newcomers within a single model makes the task all the more challenging. Finally, systems often need to combine various kinds of signals, such as different kinds of feedback originating from its users together with external third-party data describing the offered items.
The methods described in the talk were central to winning the Netflix Prize competition. In addition, some of the methods are demonstrated on the Yahoo! Music service, while using one of the largest publicly available datasets, containing over 250 million ratings, more than a million users and more than half a million items. This dataset, which genuinely reflects the wisdom of a large crowd and its cumulative taste in music, will be the focus of the KDD-Cup’2011 contest, commencing on March 15, 2011.
