Saturday, January 26, 2013

Yahoo scientist, Online multimedia content points to human behaviour

One look at the news feed of Facebook will give you an indication of the huge amount of data that is flowing into the cyberworld. Earlier, it was mostly text, in the form of status messages. Now it's mostly images in the form of photographs and videos. It's not surprising, considering the availability of technology. 

Thanks to good cameras on even basic phones and inexpensive data plans, it's now easy not only to shoot images and upload them to websites, but also to share them on various platforms like social networking sites or email them to friends. 
The profusion of photos, videos in combination with texts has led scientists to study this new online phenomenon. Correlations and analyses have brought out some interesting insights into human behaviour. 

Dr Alejandro Jaimes is a senior research scientist, who manages the Social Media Engagement group at Yahoo! Research in Barcelona, Spain. He was in Bangalore recently. During a chat, he threw light on some of the research studies that are being undertaken in this area and their findings. A lot of it is in the academic realm. It's too early to draw extrapolations and come to definitive conclusions. But, the insights are nevertheless interesting. 

One of the topics of study is online networking, wherein people form or join groups. A theory in sociology that explains formation of groups is the "Common identity and common bond". Groups are either "social" or "topical", depending on what prompts people to join a particular group: it's either because people want to make personal connections with other members of the group, or they are interested in the topics discussed in the group. 

Jaimes' team analysed the explicit social links and discussion patterns on Flickr groups, and was able to detect the bias of a group towards 'sociality' or 'topicality'. The research team found that it could predict the nature of the group and what was more important for the people in the group. 

Jaimes says the study shows why people join a particular group. "This kind of information is very useful for automatically recommending groups to users based on topics or friends' connections, and to understand how information propagates depending on the type of group." 

Research into activity of groups and the group sizes in Flickr showed that the Dunbar Number applies online too -- that is the maximum number of stable friends we can have, which is around 150. 

Another interesting area of study is finding correlation among people (celebrities and politicians, for example) appearing in photographs and texts. Using mathematical models, scientists have been enable to predict with some amount accuracy who will participate in a particular event, like the premier of a movie or fashion show or sports event. 

For this, the researchers built a social network based on photographs and analysed the people appearing in photos. Says Jaimes, "If you tell the system that a famous actor is at a movie premiere, our algorithm will produce a list of other celebrities or public figures that are likely to be at that event with the actor. This can be very useful for celebrity gossip and for discovering interesting relationships between public figures, among others."

No comments:

Post a Comment