Relation topic construction and its application in semantic relation extraction
First Claim
Patent Images
1. A method comprising:
- automatically collecting training data from manually created semantic relations using at least one computerized device;
automatically extracting rules from said training data to produce extracted rules using said computerized device, said extracting rules comprising automatically removing noisy data from said semantic relations by filtering out ones of rules that do not occur in said training data;
automatically characterizing existing semantic relations in said training data based on co-occurrence of said extracted rules in said existing semantic relations using said computerized device;
automatically constructing semantic relation topics based on said characterizing of said existing semantic relations using said computerized device; and
grouping instances of said training data into said semantic relation topics to detect new semantic relations using said computerized device.
2 Assignments
0 Petitions
Accused Products
Abstract
Systems and method automatically collect training data from manually created semantic relations, automatically extract rules from the training data to produce extracted rules, and automatically characterize existing semantic relations in the training data based on co-occurrence of the extracted rules in the existing semantic relations. Such systems and methods automatically construct semantic relation topics based on the characterization of the existing semantic relations, and group instances of the training data into the semantic relation topics to detect new semantic relations.
19 Citations
24 Claims
-
1. A method comprising:
-
automatically collecting training data from manually created semantic relations using at least one computerized device; automatically extracting rules from said training data to produce extracted rules using said computerized device, said extracting rules comprising automatically removing noisy data from said semantic relations by filtering out ones of rules that do not occur in said training data; automatically characterizing existing semantic relations in said training data based on co-occurrence of said extracted rules in said existing semantic relations using said computerized device; automatically constructing semantic relation topics based on said characterizing of said existing semantic relations using said computerized device; and grouping instances of said training data into said semantic relation topics to detect new semantic relations using said computerized device. - View Dependent Claims (2, 3, 4, 5, 6)
-
-
7. A method comprising:
-
automatically collecting training data from manually created semantic relations using at least one computerized device, said manually created semantic relations being based on explicit lexical and syntactic information; automatically extracting rules from said training data to produce extracted rules using said computerized device; automatically characterizing existing semantic relations in said training data based on co-occurrence of said extracted rules in said existing semantic relations using said computerized device; automatically constructing semantic relation topics based on said characterizing of said existing semantic relations using said computerized device; grouping instances of said training data into said semantic relation topics to detect new semantic relations using said computerized device, said new semantic relations being based on implicit lexical and syntactic information; receiving at least one input question into said computerized device; identifying ones of said new semantic relations that are present in said input question, to produce question semantic relations corresponding said input question using said computerized device; matching said question semantic relations to data source semantic relations of at least one data source maintained is a separate device external to said computerized device to generate a plurality of candidate answers using said computerized device, said data source semantic relations comprising ones of said new semantic relations present in said data source; calculating an answer score for each of said candidate answers based how closely said question semantic relations match said data source semantic relations using said computerized device; and ranking said candidate answers according to said answer score using said computerized device. - View Dependent Claims (8, 9, 10, 11, 12)
-
-
13. A system comprising:
-
a processor; and a non-transitory data storage device in communication with said processor, said non-transitory data storage device maintaining manually created semantic relations, said processor automatically collecting training data from said manually created semantic relations, said processor automatically extracting rules from said training data to produce extracted rules, said extracting rules comprising automatically removing noisy data from said semantic relations by filtering out ones of rules that do not occur in said training data, said processor automatically characterizing existing semantic relations in said training data based on co-occurrence of said extracted rules in said existing semantic relations, said processor automatically constructing semantic relation topics based on said characterizing of said existing semantic relations, and said processor grouping instances of said training data into said semantic relation topics to detect new semantic relations. - View Dependent Claims (14, 15, 16, 17, 18)
-
-
19. A non-transitory computer readable storage medium readable by a computerized device, said non-transitory computer readable storage medium storing instructions executable by said computerized device to perform a method comprising:
method comprising; automatically collecting training data from manually created semantic relations; automatically extracting rules from said training data to produce extracted rules, said extracting rules comprising automatically removing noisy data from said semantic relations by filtering out ones of rules that do not occur in said training data; automatically characterizing existing semantic relations in said training data based on co-occurrence of said extracted rules in said existing semantic relations; automatically constructing semantic relation topics based on said characterizing of said existing semantic relations; and grouping instances of said training data into said semantic relation topics to detect new semantic relations. - View Dependent Claims (20, 21, 22, 23, 24)
Specification