KAIST

Rumors sporadically spread with people with fewer followers in the center

Researched over 100 rumors in the US from 2006 to 2009

Is it possible to filter information on SNS such as Twitter and Facebook?

A research team led by Professor Mee-Young Cha from the Department of Cultural Technology Graduate School at KAIST, Professor Kyo-Min Jung of Seoul National University, Doctor Wei Chen and Yajun Wang of Microsoft Asia, has developed a technology that can accurately filter out information on Twitter to 90% accuracy.

The research not only deduced a new mathematical model, network structure, and linguistic characteristics on rumors from SNS data, but is also expected to enhance the effort to make secure technology to regulate Internet rumors. 

The team analysed the characteristics of rumors in over 100 widespread cases in the US from 2006 to 2009 on Twitter. The team gathered data, which included a range of areas such as politics, IT, health and celebrity gossips, and their analysis could identify rumors to 90% accuracy. The filtering was more accurate in rumors that included slanders or insults.

The research team identified three characteristics of the spread of rumors.

Firstly, rumors spread continuously. Normal news spreads widely once and is mentioned rarely again on media, but rumors tend to continue for years. 

Secondly, rumors spread through sporadic participation of random users with no connections. Rumors start from people with fewer followers and spread to the more popular. This phenomenon is often observed in rumors concerning celebrities or politicians. 

Lastly, rumors have unique linguistic characteristics. Rumors frequently include words (such as “it may be true,” “although not certain, I think,” “although I cannot fully remember”) related to psychological processes that question, deny, or infer the reliability of the information.

Professor Cha said, “This research deduced not only a statistical and mathematical model but also is an integrated research on social psychological theory on the characteristics of rumors that attract great attention from the society based on ample data.”

The results were made public in IEEE International Conference on Data Mining last December in Texas, USA.