Ranking in information streams

Abstract

Information streams allow social network users to receive and interact with the latest messages from friends and followers. But as our social graphs grow and mature it becomes increasingly difficult to deal with the information overload that these realtime streams introduce. Some social networks, like Facebook, use proprietary interestingness metrics to rank messages in an effort to improve stream relevance and drive engagement. In this paper we evaluate learning to rank approaches to rank content based on a variety of features taken from live-user data.

Re-ranking the Twitter search result

“… Hashtag refers to the topics or keywords marked by a "#" symbol in a tweet, and it can be used to categorize messages by Twitter users. For Example, "hashtags":[{"indices": [32,36],"text":"lol"}].…”

“…"http:\/\/twitter.com\/yunorno\/status\/114080493036773378\/photo\/1",  URLs refer to the URLs included in the text of a tweet or within the textual fields of a User object. For example, "urls":[{"indices": [32,52],"url":"http:\/\/t.co\/IOwBrTZR", "display_url":"youtube.com\/watch?v=oHg5SJ\u2026", "expanded_url":"http:\/\/www.youtube.com\/watch?v=oHg5SJYRHA0"}].…”

A user-centered and group-based approach for social data filtering and sharing

“…Researchers also started to be interested in social stream (re)ranking. Several re-ranking methods have been introduced for different SNSs, for example Twitter [136], LinkedIn [11,71], and Facebook [28]. Like EdgeRank, these methods do not focus on the analysis of the textual content of incoming information, but rely on a wide set of specific features involving not only the information, but also the source and the target users.…”