Rupert Westenthaler created STANBOL-1211:
--------------------------------------------
Summary: Improve Chunk support for Entitylinking
Key: STANBOL-1211
URL: https://issues.apache.org/jira/browse/STANBOL-1211
Project: Stanbol
Issue Type: Improvement
Components: Enhancement Engines
Reporter: Rupert Westenthaler
Assignee: Rupert Westenthaler
Both the EntityLinkingEngine as well as the LuceneFstLinkingEngine do currently
not use Chunk information very well. For now Chunks are only used to also
lookup multiple matchable tokens in the same chunk with the Vocabulary - to
increase recall in case proper-noun linking is enabled.
However chunks can also be useful to increase precision by using the span of
the Chunk as a base for calculating the confidence of the linked Entity.
A typical example are suggestions for Persons Names: If a text mentions the
Given and Family name of a Person not present in an vocabulary the
Entitylinking may suggest Entities just matching on of the two names with a
100% confidence. When using the span of the Chunk such suggestions would be
omitted as the minimum label match score is typically > 50%.
Other example include matches for "US {OrgName}" where "US" is linked when the
whole organization is not found; same with "{OrgName} {Role}" where the {Role}
(e.g. president) is linked; Also cases like "15. September, 2013" may cause
September to be suggested if present in the vocabulary.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
