Various news are in the spotlight everyday. But what topics are fashionable? Are we able to forecast them? These questions captured the attention of scientists at German Research Center for Artificial Intelligence. Tim Althoff and his colleagues presented the first comprehensive study across three major online and social media streams, Twitter, Google, and Wikipedia, covering thousands of prominent topics during an observation period of an entire year.
There are diverse reasons to explore the contents of websites mentioned above. On the one hand, the researchers seek to find out something about our social world. They think that widely discussed topics represent interests of the people. Althoff and his associates even claim that Internet trends are mirrors of our society. On the other hand, they are driven by more practical concerns. They created an instrument, which should help to foresee dynamics of future tendencies. The scholars say: “Forecasts enable us to anticipate changing information needs of users ahead of time and to allocate limited resources in trend aware multimedia systems such as recommender systems or video concept detection systems.”
The study shows that only few themes prevail longer than two weeks. Longest trending time is observed on Google. As might be expected, furor on Twitter lasts much shorter. Surprisingly, the data representing Wikipedia fashions strongly resemble the statistical distribution of Twitter trends. “This raises the question when and why exactly people are turning to Wikipedia to satisfy their information needs,” the scientists ask. It was revealed that Wikipedia and Twitter trends are not only more ephemeral, but also peaks a few days before it happens on Google.
The researchers discovered that websites tend to specialize. The most popular subjects on Google are sports and celebrities. However, themes related to politics are prevalent as well. Interestingly, Twitter has the largest proportion of trends related to the new products and technologies. Such negative topics as catastrophes and deaths of the famous person were most widely read topic on Wikipedia. “Contrary to the intuition that Wikipedia is a slowly evolving channel which people use to read up on complicated topics, especially when also considering the temporal properties of the Wikipedia channel from the analysis above, we believe that many users use Wikipedia during these trends and events to learn about or remind themselves about related topics,” the scientists elaborate.
A new forecasting technique, which is automatic and bases its forecasts on the linguistic similarity of the topics, was devised. In order to test the predictive power of their device they tried to predict trends on Wikipedia. Their predictions were quite successful. “Forecasts by the proposed approach are about 9-48k views closer to the actual viewing statistics than baseline methods,” the scholars claim.
Original research article: arXiv:1405.7452