Predicting Clicks

Search engines make money according to how many times you click your mouse. Companies wanting to increase their revenues can capitalize on click-power by increasing the chance that someone will click on through to their web sites from a search engine. Penn State's Jim Jansen, a professor of information sciences and technology, worked with a team of colleagues to find a way to measure customer satisfaction in relation to search engine results. It is hoped that this new information can help companies maximize their commercial potential.

Jansen says that his study is the first time anyone used neural networks to analyze the interaction between user and system as research is done on a search engine. Once information is obtained on how the user interacts with the system, this knowledge can be used by search engine companies to improve the rate at which users click on through to companies. This is done by improving the algorithms which affect retrieval and ranking.

User Clicks

Jansen and his colleagues, Ying Zhang, from the industrial and manufacturing engineering department of Penn State, and Amanda Spink, at Queensland University of Technology, studied search logs from the popular search engine, Dogpile.com to try to identify factors that may translate to greater or fewer user clicks.

Basing their work on neural network data, the team found nine factors that are implicated in the rate of click-throughs. Five of these nine factors bring on an increase in user clicks, while four bring on a decrease. One factor was rated as having no discernible effect. The positive factors included: how many records turn up in the search, the total of the listing ranks, the length of a given query, the type of browser used, and response time. The negative factors were: how many links had to be clicked to get to the information, how fast a user types, the time of day for the query, and the time the user logs in to the search engine. It seems that user intent, no matter whether the user seeks information needed to make a purchase, or wants to navigate to a specific website, has no impact on the number of click throughs to commercial websites.

Jansen comments, "From a practical point of view, the more that a user reformulates the initial query, the click-through will increase, although there may be individual queries where the user clicks on no links."

Neural Networks

Neural networks can demonstrate the relationship between input and output, so it was natural for the researchers to use this as a model on which they could base their study. Taking the results from a Dogpile transaction log, the scientists were able to calculate input and output values.

The researchers discovered that more clicking is done early in the day and those who use the Internet Explorer browser clicked through more often. Those who clicked through the most were using longer queries and modifying their queries more than the average searcher. They also searched for a longer than average period of time. Those searching the web tended to click through more often than those exploring audio, video, or images. This study was published in the March 2009 issue of the Journal of the American Society for Information Science and Technology.