Understanding User Behavior From Online Traces

Abstract

People nowadays share large amounts of data online, explicitly or implicitly. Analysis of such data can detect useful behavior patterns of varying natures and scales, from mass immigration between continents to trendy venues in a city in turn. Detecting these patterns can be used for improving online services. However, capturing behavior patterns may be challenging, since such patterns are often of a specialized essence, no benchmark or labeled data exist, and it is not even clear how to formulate them to enable computation. Moreover, it is often unclear how recognition of these patterns can be translated into concrete service improvement. We analyzed major datasets of three common types of online traces: microbloging, social networking, and web search. We detected online behavior patterns and utilized them toward novel services and improvement of traditional services. In this paper we describe our studies and findings, and offer a vision for future development.