پديدآورندگان :
Safdel Atefeh Department of Computer Science & Engineering & IT Shiraz University Shiraz - Iran , Fakhrahmad Mostafa Department of Computer Science & Engineering & IT Shiraz University Shiraz - Iran , Sadreddini Mohammad Hadi Department of Computer Science & Engineering & IT Shiraz University Shiraz - Iran
كليدواژه :
slot filling , knowledge base , information extraction , Freebase , entity linking
چكيده لاتين :
In recent years, several large-scale knowledge bases
have been constructed. Many applications, including Question
Answering (QA) systems, make use of these knowledge bases in
order to find their required pieces of information. Although
publicly available knowledge bases (e.g., YAGO, NELL, Wikidata,
and Freebase) are really massive, they suffer greatly from
incomplete information. For instance, Freebase has 3 million
Entities of type PERSON, in plenty of cases (subjects), it doesn’t
have any information about some relations such as PLACE OF
BIRTH (71% missing values), PROFESSION (68% missing
values). In this study, we have focused on subjects of type
PERSON and have extracted objects for each subject-relation pair
where the subjects are selected from 100K most prominent people.
For this purpose, candidate textual fillers are extracted using a slot
filling module via Wikipedia pages. An entity linking module is
then applied in order to map the answers to their entities. Finally,
the task of merging, ranking, and validation of the entities are
carried out by making use of WordNet relations. The proposed
system detects and involves identical words which could be useful
while they have different expressions. The experimental results in
terms of MRR and MAP are promising.
