In light of a recent statement by US President Donald Trump that Google is manipulating searches to promote negative news articles about him, the Australian Institute of Business and Economics’ Dr Patrick O’Callaghan has explained how Google ranks webpages.
The mathematical economist, whose research focuses on the mathematics of decision making and prediction, said he would be surprised if Google had manipulated the rankings against Trump.
“Google’s rankings are demand driven, they build on the billions of searches we conduct every day,” Dr O’Callaghan said.
“However, since Google is virtually a monopoly, there is arguably a good case for an investigation simply on the grounds that market forces are not in place to keep Google honest,” he said.
But how does Google produce its page rankings and how could these be biased?
“To produce its page ranking, Google relies on two kinds of information. The first consists of the world wide web itself – two billion webpages and their links to other pages,” Dr O’Callaghan said.
“The second kind of information is proprietary, consisting of all the past searches that people have made via Google and the feedback that Google gets from the clicks searchers make on its SERPs each day.
“This is a fast moving source of data with Google handling more than five billion searches per day.
“When a new search term is entered, Google aggregates all this information to produce a plausible ranking of webpages.”
Dr O’Callaghan said an example of this would be if an identical article on Trump featured on both the CNN and the Breitbart website, and Google ranked the CNN article higher than the Breitbart article.
“Would Google be biased in this situation? Not if Google’s goal is to put forward the most plausible ranking for the current search,” he said. “If more past searchers have chosen CNN over Brietbart, then Google will tend to promote CNN.
“If the individual searcher has a tendency to choose CNN over Brietbart, the same is true.
“Finally, if more webpages link to CNN than to Breitbart, then, again, Google promotes the CNN version because this indicates that people are more likely to choose CNN.”
Dr O’Callaghan also noted that Google and Facebook may have implemented “anti-fake news” policies during the last couple of years.
“If such policies end up being biased against pro-Trump websites, then Trump might find good grounds to force Google and Facebook to treat all sources equally,” he said.
“Clearly this would mean that false information would have a higher chance of propagating around the web.”
Dr O’Callaghan is writing a paper entitled Prudent Case-based Prediction, which provides a mathematical framework for understanding how past searches are related to the most plausible ranking of pages for the current search.
It highlights some of the important steps a search engine might take to ensure it can handle novel searches consistently and quickly.
This is important because, even for Google in 2017, 15 percent of the search terms it encountered were new, but also because search engines with less rich data than Google may find this framework useful.
Media: Dr Patrick O’Callaghan, p.ocallaghan@uq.edu.au
