The numbers of entries for values related to influenza epidemics correspond to the immediate public interest at the time of the epidemic. The Wikipedia model is even better than that of Google Trends
Avi Blizovsky
Can monitoring the number of entries on Wikipedia reveal how many people have had the flu? Researchers at Boston Children's Hospital have developed a method that estimates the level of spread of flu-like illnesses in the American population by analyzing the movement to specific entries relevant to the flu on Wikipedia.
The model by David McIver and Jen Braunstein, published on April 17, 2014 in the journal PLOS Computational Biology, estimates the level of influenza morbidity among the American population up to two weeks earlier compared to the availability of data from the Centers for Disease Control and Prevention. Also, they more accurately estimate the peak week in the number of patients 17% better than the Google search trend using Google Trends.
McIver and Braunstein calculated the number of times these values were read each day from December 2007 to August 2013 (the data is available on Wikipedia in open source AB). The model they developed worked well even during flu seasons that were more severe than usual and even during events such as the swine flu epidemic ( H1N1) that received high media attention.
In their article, the two wrote: "Each flu season provides new challenges and uncertainties both in the public and among those involved in the field of public health. We hope that with the help of the new flu monitoring method, we will be able to utilize information that is freely available to the public to make it easier for people to receive accurate, near real-time information about the level of morbidity. After further validation, the model may be used as a component of an automatic system that will use a model to predict the levels of the flu in the US and provide support for the existing monitoring tools."
