Researchers at Microsoft have been investigating the correlation between eye movement, or “gaze tracking”, and mouse cursor movement on search engine results pages. The findings have some interesting implications for future search engine algorithms, and may help to improve the results returned for search queries.
By measuring mouse cursor behaviour, such as clicks and hovers over particular page regions, alongside the results from an eye-tracking study, Microsoft has shown that areas of interest to a user directly correlate to mouse cursor behaviour. This is particularly prevalent for search engine results pages.
While measuring click-through behaviour is nothing new, researchers have pointed out that other mouse behaviours are just as important in determining whether a result set is accurate. Up until now, many researchers in the field have simply believed that a clicked link generally expresses user interest, while no clicks on a result page would suggest that a user has not found the answer that he or she is looking for. This is clearly wrong. For example, in particular cases a user may find an answer to a query in the description for a result link. This is known as ‘good abandonment’, since the user abandons the search because the answer was found without having to click through to any of the pages listed in the result set.
In lab environments, it is clear that users make use of the mouse cursor for a variety of other purposes, including moving the cursor to act as a reading aid or highlighting text to mark interesting results. Before now, it has been easy to monitor these behaviours in a lab, but capturing and making sense of this information on a global scale has seemed near impossible.
This application had a total size of 750 bytes, so that it would not significantly affect page load times. By carefully designing the solution to minimise the amount of data that would need to be sent by the client back to the server, the application could be used to provide detailed information about mouse cursor behaviour while browsing search engine results pages. After embedding this application into the Bing search pages used internally within Microsoft, the researchers were able to obtain over 7.5-million cursor events in under a month. The results are interesting, not least because they show the general direction that Microsoft may be taking with its future search algorithms.
So, what are the conclusions? Firstly, tracking cursor behaviours can provide a powerful tool to estimate search result relevance. For pages where users actually clicked through to a particular result, there was a stronger correlation between hover rate and human relevance judgements than with click through rate.
A combination of behaviours, however, yielded a much stronger correlation than any of the different behaviours measured. On pages where users did not click on any links, there was a high correlation between hover rate and human relevance judgements. Once again, the combination of behaviours measured yielded an even greater correlation.
This suggests that it is possible to use mouse behaviour to get a fairly accurate idea of how good the results are for a query, even if a user does not click on a link. More interesting was the conclusion that it is possible to measure good and bad abandonment on search pages by tracking how quickly and how far the cursor moves in each instance. It is clear that in good abandonment scenarios, the cursor movements were shorter; the cursor was moved less and that the cursor moved more slowly.
While many of the results from the study may be fairly intuitive, it is interesting that the correlations are so strong that we can be pretty sure that by tracking mouse movements we can make some pretty educated guesses about how relevant search results actually are to end users. In the future, we can certainly expect search engines like Bing to take mouse cursor behaviours into account within the search algorithm.
This cursor-tracking technique, deployed on to a general websites, will perhaps allow media owners and advertisers to understand who is really looking at their ads.