scoring mechanism for entity topics

[m1ci]

This action is follow up of https://github.com/freme-project/e-Entity/issues/44

We need to define mechanism for scoring topics assigned to each entity. Only topics which are most relevant for the entity/document should be part of the output.

[nilesh-c]

I guess we can divide this up into:

1. Having a simple (will need to think of something quick and dirty) scoring mechanism that we can implement in a couple of days, maybe by 15th?
2. Longer term, 0.4? Since we'll need to have better relevance/confidence scores for linked entities too, maybe it can be combined with this thing. The graph approach we discussed is used by AGDISTIS and seems to perform okay. We could get scores for both entities and topics in that way. Possibly.

[m1ci]

@nilesh-c this is follow up task after #44 so having the scoring by 15th is not very possible.

Some useful stats regarding the topics in DBpedia:

• 918.188 distinct topics

  SELECT COUNT(DISTINCT ?o)
  WHERE  {
   ?s <http://purl.org/dc/terms/subject> ?o .
  }

• SPARQL query sorted by number of entities with the topic

  SELECT  ?class (COUNT(?s) AS ?count ) WHERE {
    ?s <http://purl.org/dc/terms/subject> ?class 
  } GROUP BY ?class ORDER BY DESC(?count)

[koidl]

looks good to me

[m1ci]

I think this issue is dead. lets close it.