Evidence. Next, we study how the strength of evidence
may affect recall. We compute the evidence of every
reviewer and object (xi and yj values) using IR’s bias and
controversy values (see Section 3.2). We then split the
reviewers into two groups: IR-Upper Half Evidence (the top
50 percent of reviewers in terms of evidence) and IR-Lower
Half Evidence (the rest). For each group, we measure bias
recall, which is the percentage of the l predetermined biased
reviewers that are also ranked among the top l most biased
reviewers within the group. Recall and precision are
practically the same as they both share the same denominator
ðlÞ. A similar exercise is carried out for objects.

太高深了，不懂。