Reference

Context The huge amounts of data produced every day by Web 2.0 applications, call for efficient online aggregation and filtering techniques. News aggregators (Google News, Yahoo!News) and Social Media (Twitter, Facebook) provide users with the possibility to filter and personalize their information streams. Focusing on the personalization of information, we developed the MeowsReader news aggregator. With MeowsReader, users can define their interests by issuing queries to be continuously informed about relevant news items as these are published on the web. Based on the textual similarity, as well as on the news item’s importance computed by using user-independent factors (e.g. source authority), only the best matching results (top-k results) reach the subscribed users. Continuous top-k text queries over text streams Users define continuous text queries. Information streams are collected and filtered through these queries. Users can continuously consult the top-k query results, according to a time aware query-item scoring function 11Top-k News Recommendation, Filtering and Personalization Top-k query filtering solution Combining query dependent cosine similarity Squery and query independent item importance Sitem we define the query-item score as: S(q, i) = α Sitem(i) + β Squery(q, i) Using a two-dimensional representation for the queries and knowledge on score upper bounds and expected average values we define a number of optimal early stopping conditions to prune the search space [1]. Meows Reader architecture and demonstration An RSS aggregator collects news items from several sources. Items are matched over the stored continuous queries (Top-k filtering). The Hot topic extractor computes a set of topics with high importance over recent items. Users can choose hot topics or define their own queries through th