Online conversations are deeper than we believe

NEWS
Feb 9, 2011

By Deepa Kurup Originally published February 8th on The Hindu Yahoo researcher Ravi Kumar proposes simple mathematical models to capture patterns in these interactions and understand user behaviour Many are quick to dismiss conversations in the online world as all too superficial, or even short-lived. Others will argue that micro-blogs and social networking platforms are all about people shouting out their thoughts or informing people about the venue of their last party. Are conversations online really that shallow? Certainly not, proves this young researcher. Ravi Kumar, principal research scientist at Yahoo Labs has spent over a year studying not only the nature of online conversations but also analysing how these build on different platforms. In his attempt to identify a common model to explain and predict these conversations, Kumar finds that online conversations tend to be far longer that what existing research models predict. Trawling data and conversations as datasets from different platforms — Usenet groups, Yahoo Groups and Twitter — he found that existing models far underestimate the depth and length of these “conversation trees”. “The research proposes a simple mathematical model for the generation of basic conversation structures and then refine this model to take into account the identity of each member of the conversation. Our finding is robust across different settings.” Key factors He explains that further analysis of content of these conversations found that ‘recency' (or timeliness) of a conversation topic, or the popularity, are two key factors that drive these interactions. Conversations or heated debates on news, or discussions on a popular celebrity or political figure or ideology, tend to be richer. Anecdotal evidence suggests that ‘recency' plays a bigger role for topics such as sports and news whereas popularity plays a bigger role for topics such as politics, he says. And it was a lot of data that he analysed to understand, and then confirm, these patterns, explains Kumar. Around 15,000 groups and 20 million messages later – using algorithms developed for this purpose using Hadoop and other tools – Kumar's research paper proposes simple mathematical models to capture patterns in these interactions. So, for conversations to last, and for people to be drawn in to staying online longer and make them contribute, Kumar suggests that content providers offer the right mix of timeliness and popularity. Getting it right But why evolve these models? How can understanding the nuances of online interactions help companies or businesses interested in the online space? Kumar points out that his ultimate goal is to understand user behaviour. By doing so, several applications can be developed around these models, for example User Generated Content that many websites are already capitalising on. Companies or Internet technologists can predict and understand what can potentially go viral or what are the combinations of factors that can drive more users and get them hooked on to a platform, Kumar explains.

More information at:

  • http://www.thehindu.com/sci-tech/technology/article1167526.ece