The problem of outlier detection is extremely challenging in many domains
such as text, in which the attribute values are typically non-negative, and
most values are zero. In such cases, it often becomes difficult to separate the
outliers from the natural variations in the patterns in