Other research has shown that user behavior information, such as clickthrough
data (e.g., which documents have been clicked on in the past, which rank positions
were clicked) and browsing data (e.g., dwell time on page, links followed),
can have a significant impact on the effectiveness of the ranking. This type of evidence
can be added into the inference network framework using additional operators,
but as the number of pieces of evidence grows, the issue of how to determine
the most effective way of combining and weighting the evidence becomes
more important. In the next section, we discuss techniques for learning both the
weights and the ranking algorithm using explicit and implicit feedback data from
the users.