Method to improve the named entity classification

• US 10,108,705 B2
• Filed: 11/09/2011
• Issued: 10/23/2018
• Est. Priority Date: 09/29/2010
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Ni, Yuan, Qui, Zhao Ming, Zhang, Lei
• Primary Examiner(s): Featherstone, Mark D
• Assistant Examiner(s): Gmahl, Navneet
• Application Number:
  US13/292,111
• Publication Number:
  US 20120185480A1
• Time in Patent Office: 2,540 Days
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification

Abstract

A method is described for providing a named entity classification in a computing system having a processor, comprising the steps of the processor reading, from an LOD (Linking Opening Data) set, an LOD node corresponding to a to-be-classified named entity. The processor also determining a type attribute of the LOD node corresponding to the to-be-classified named entity as a tagged type of the to-be-classified named entity and further reading a candidate type. Finally, the processor computing, based on the tagged type, a possibility of the to-be-classified named entity belonging to the candidate type.

Claims

1. A method of providing a named entity classification in a computing system, the method comprising:

reading, from an LOD (Linking Opening Data) set, an LOD node corresponding to a to-be-classified named entity, wherein the LOD node is associated with a uniform resource identifier and corresponds to a web page, the LOD node further comprising at least a plurality of property entries;

determining a type attribute of the LOD node corresponding to the to-be-classified named entity as a tagged type of the to-be classified named entity;

reading a candidate type;

computing a possibility of the to be-classified named entity belonging to the candidate type, comprising;

mapping the candidate type to a node of an intermediate ontology, wherein the intermediate ontology includes at least a structured correlation of generic-specific relationships, the generic-specific relationships including at least an identical relationship, a homologous relationship and a conflicting relationship;

computing an attribute matching score between the candidate type and the tagged type based on a relationship between the mapped node of intermediate ontology and the candidate type, wherein computing the attribute matching score of a mapped intermediate ontology having an identical type is based on a predetermined value, the attribute matching score of a mapped intermediate ontology having a generic-specific relationship is based on the predetermined value and a first count of nodes between two mapped nodes of the intermediate ontology, and the attribute matching score of a mapped intermediate ontology having a homologue relationship is based on the predetermined value and a second count of nodes between two mapped nodes of the of the intermediate ontology to a common source node and a third count of nodes between two mapped nodes of the of the intermediate ontology to the common source node;

performing statistical processing to attribute matching scores corresponding to the candidate type to obtain a possibility of the to-be classified named entity belonging to the candidate type, wherein performing statistical processing to each attribute matching score corresponding to a same candidate type further comprising;

converting the attribute matching scores to a node matching score based on the correspondence relationship between the attribute matching scores and the LOD node in order to reduce type attribute entry noise;

performing statistical processing to each node matching score corresponding to a same candidate type, thereby obtaining a possibility of the to-be-classified named entity belonging to the candidate type; and

selecting one or more tagged types based on satisfaction of an attribute matching score threshold.

2. The method according to claim 1, wherein performing the statistical processing to each attribute matching score corresponding to the same candidate type further comprising:

performing weight averaging to the attribute matching scores based on characteristics of the LOD nodes corresponding to the attribute matching scores.

3. The method according to claim 1, wherein reading the LOD node corresponding to the to-be-classified named entity from an LOD data set further comprising:

determining an LOD node corresponding to the to-be-classified named entity by comparing the to-be-classified named entity with a name attribute as specified in the LOD node'"'"'s attribute entry with a predicate of any of;

rdfs;

label;

dbpedia;

birthname;

or

foaf;

name.

4. The method according to claim 1, further comprising:

determining an LOD node having a redirect relationship with the LOD node corresponding to the to-be classified named entity as another LOD node corresponding to the to-be-classified named entity.

5. The method according to claim 1, wherein converting the attribute matching scores to the node matching score further comprising:

averaging a plurality of attribute matching scores corresponding to a same LOD node, in order to obtain the node matching score.

6. The method according to claim 1, wherein converting the attribute matching scores to the node matching score further comprising:

selecting a maximum value from a plurality of attribute matching scores corresponding to a same LOD node as the node matching score.

7. The method according to claim 1, wherein converting the attribute matching scores to a node matching score further comprising:

indicating presence of a conflict if an attribute matching score exists among the attribute matching scores corresponding to the same candidate type; and

setting the corresponding node matching index to indicating presence of confliction.

8. The method according to claim 7, further comprising:

determining to set the node matching score to indicating presence of confliction based on a relationship between the number of attribute matching scores indicating presence of confliction and the number of attribute matching scores indicating absence of confliction in the attribute matching scores corresponding to the same candidate type.

9. An apparatus for named entity classification, comprising:

LOD node reading means configured to read an LOD (Linking Opening Data) node corresponding to a to-be-classified named entity from an LOD data set, wherein the LOD node is associated with an uniform resource identifier (URI) and corresponds to a web page, the LOD node further comprising at least a plurality of property entries;

tagged type determining means configured to determine a type attribute of the LOD node corresponding to the to-be-classified named entity as a tagged type of the to-be-classified named entity;

candidate type reading means configured to read a candidate type;

possibility determining means configured to compute, based on the tagged type, a possibility of the to-be-classified named entity belonging to the candidate type, further comprising;

mapping means configured to map the candidate type to a node of an intermediate ontology, wherein the intermediate ontology includes at least a structured correlation of generic-specific relationships, the generic-specific relationships including at least an identical relationship, a homologous relationship and a conflicting relationship;

computing means configured to compute an attribute matching score between the candidate type and the tagged type based on a relationship between the mapped node of intermediate ontology and the candidate type, wherein computing the attribute matching score of a mapped intermediate ontology having an identical type is based on a predetermined vale, the attribute matching score of a mapped intermediate ontology having a generic-specific relationship is based on the predetermined value and a first count of nodes between two mapped nodes of the intermediate ontology, and the attribute matching score of a mapped intermediate ontology having a homologue relationship is based on the predetermined value and a second count of nodes between two mapped nodes of the of the intermediate ontology to a common source node and a third count of nodes between two mapped nodes of the of the intermediate ontology to the common source node;

performing means configured to perform statistical processing to attribute matching scores corresponding to a same candidate type, thereby obtaining a possibility of the to-be classified named entity belonging to the candidate type, wherein performing statistical processing to each attribute matching score corresponding to a same candidate type, thereby obtaining a possibility of the to-be-classified named entity belonging to the candidate type futher comprising;

converting means configured to convert the attribute matching scores to a node matching score based on the correspondence relationship between the attribute matching scores and the LOD node in order to reduce type attribute entry noise;

performing means configured to perform statistical processing to each node matching score corresponding to a same candidate type, thereby obtaining a possibility of the to-be-classified named entity belonging to the candidate type; and

selecting means configured to select one or more tagged types based on satisfaction of an attribute matching score threshold.

10. The apparatus according to claim 9, wherein the means configured to perform statistical processing to each attribute matching score corresponding to the same candidate type, thereby obtaining a possibility of the to-be-classified named entity belonging to the candidate type further comprising:

means configured to perform weight averaging to the plurality of attribute matching scores based on characteristics of the LOD nodes corresponding to the attribute matching scores.

11. The apparatus according to claim 9, wherein the LOD node reading means further comprising:

means configured to determine an LOD node corresponding to the to-be classified named entity by comparing the to-be-classified named entity with a name attribute as specified in the LOD node'"'"'s attribute entry with a predicate of any of;

rdfs;

label;

dbpedia;

birthname;

or

foaf;

name.

12. The apparatus according to claim 9, further comprising:

means configured to determine an LOD node having a redirect relationship with the LOD node corresponding to the to-be-classified named entity as another LOD node corresponding to the to-be-classified named entity.

13. The apparatus according to claim 9, wherein the means configured to convert the attribute matching scores to the node matching score further comprising:

means configured to average a plurality of attribute matching scores corresponding to a same LOD node, thereby obtaining the node matching score.

14. The apparatus according to claim 9, wherein the means configured to convert the attribute matching scores to the node matching score further comprising:

means configured to select a maximum value from a plurality of attribute matching scores corresponding to a same LOD node as the node matching score.

15. The apparatus according to claim 9, wherein the means configured to convert the attribute matching scores to the node matching score further comprising:

means configured to set, if an attribute matching score indicating presence of confliction exists among the attribute matching scores corresponding to the same candidate type, the node matching score to indicating presence of confliction.

16. The apparatus according to claim 9, further comprising:

means configured to determine whether to set the node matching score to indicating presence of confliction based on a relationship between the number of attribute matching scores indicating presence of confliction and the number of attribute matching scores indicating absence of confliction in the respective node matching score corresponding to a same candidate type.