Random Walk Similarities: Determine lowest values and make a cut

The Random Walk similarities seem to make sense, but once it comes to low-evidence cases, we find the same concepts being all given the same very low similarity. It would be better to delete these, defining a threshold, as these are random noise values, also introduced by the fact that we use a complete graph with extremely low edges for those cases not attested.