Query processing by, for, and of the masses: putting the user in the loop

点击翻译按钮获取中文摘要

Today's web search is confined to users submitting a keyword search query to a generic search engine and sifting through a large linearly ordered list of answers. While this is satisfactory for many purposes, it ignores the significant trends, spurred by the advent of Web 2.0, toward increased integration of content information that users search for and their social profiles (activities and connections), leading to social content sites. First, sites like del.icio.us and flickr started as social content sites: users can form social ties and tag content items and share them with their friends. Second, more and more, content-only sites like Amazon are now allowing users to tag/rate items and share them with their friends. Similarly, social sites like facebook are allowing users to share content with their friends. Third, there is a definite trend toward virtual social content sites: e.g., most news sites let users share news articles with their friends in, say MySpace. In all these social content sites -- virtual and real -- there is rich structured information on both content items and users as well on user-provided content. Treating social content sites the same way as any other site misses out on leveraging the significant content and social structure present in these sites. We envision a future where domain-specific social content sites will increasingly offer an attractive alternative for users compared to generic search engines. In this program, we will investigate the combination of techniques drawn from information retrieval style search, database style querying, recommender systems, social networks, graph modeling and querying, data analysis and data mining over rich social and content structure, and multi-dimensional querying and OLAP, to significantly enrich and advance user experience in interacting with collections of social content sites. Users will be able to seamlessly shift between heterogeneous groups of answers: these groups may consist of content items at various levels of granularity, or "similar" items which "recommended" those items, or "similar" or expert users who recommended the items and their tags/ratings, or other "topics" that may interest the user, leading to a powerful paradigm for information discovery.