Heuristic Expansion, does it help or harm so far?

[DiegoMoussallem]

Our heuristic expansion when active gives more importance to the existing entity than the new one. For example,

curl --data-urlencode "text=<entity>Leipzig University</entity> located in <entity>Leipzig</entity> in the Free State of <entity>Saxony, Germany</entity>, is one of the oldest universities in the world and the second-oldest university (by consecutive years of existence) in <entity>Germany</entity>. Famous alumni include <entity>Leibniz</entity>, <entity>Goethe</entity>, <entity>Nietzsche</entity>, <entity>Wagner</entity>, <entity>Angela Merkel</entity>, <entity>Raila Odinga</entity>, <entity>Tycho Brahe</entity> and nine <entity>Nobel laureates</entity> are associated with this university. " -d type='agdistis' http://localhost:8080/AGDISTIS

Leipzig University is disambiguated first, but when Leipzig comes in, the word Leipzig has just appeared then [M(AGDISTIS)] considers both to be the same entity, the same happens with Saxony, Germany and Germany, both are in sequence. However, while performing a disambiguation on documents, it also helps and harms, for example, MSNBC has a lot of co-reference entities then MAG achieves 0.62 without the expansion, but when it is turned on, we got 0.78. The opposite happens to AQUAINT when it is turned on 0.67, off 0.72. We need to investigate it.