US researchers develop software that identifies key players in online communities

Washington -- Scientists from the Cornell University, New York and Microsoft Research in Washington State, both US, have developed a software that identifies the most informative people in an online community, based on their posting patterns.

Researchers, who worked out to spot key players within discussions by analysing the connections between thousands of messages on several topics, have said their work could help website designers automatically reward, or highlight, the most valuable members of a community, or improve methods for searching through a conversation for the most relevant information.

Previous research has shown that certain people underpin the usefulness of a group or discussion by providing brief but straightforward and useful answers.

"You have a socially recognised role of some people as experts in some way in a community. That role was what we were trying to measure. The indicators we found had to do with the structure of their interaction with others," said Howard Welser, a sociologist at Cornell University, who led the work.

Welser and colleagues examined a month‘s worth of posts from three Usenet newsgroups:

Comp, soft-sys and matlab, about a mathematical computer programming language; Microsoft, public, windows, server, and general, about a certain type of computer server; and rec and kites about making and flying kites.

The researchers first rated the content of a total of 5,700 messages from about 450 active users and then calculated how often each user replied to messages or were replied to, how often each person started a discussion, and how many posts they contributed to an individual thread.

Then, by going through each message and rating their usefulness, they were able to spot patterns in the behaviour of different posters.

Welser‘s group found that the most informative individuals - dubbed "answer people" - were also relatively taciturn, rarely participating in discussions heavily. They also tended to shy away from the "discussion artists" who dominate most threads.

Instead, these people mostly posted one or two messages to a lot of different discussion threads, and tended to respond to users who did not post a lot. They also tended to avoid long discussions, jumping in when someone had a specific question, providing a useful answer and then bowing out from further talk.

According to Welser, since the findings use quantitative data about posting behaviour, they could prove useful for developing automated systems that assigns high reputation to certain people within a discussion.

Or, they could make it easier for a search engine to find messages that are most likely to be useful, based on the user, he said.

Scott Golder, at the Information Dynamics Lab at HP Labs in Palo Alto, California, US, said the work showed how people took on roles in an online environment, and how those roles influence the nature of the community they take part in.

"These guys do really terrific work," he said.

The findings appear in the Journal of Social Structure. (ANI)