Discovery of implicit relational knowledge by mining relational paths in structured data
View Patent ↗Predefined relation items on paths traversing predefined entities of a knowledge base are collected and feature sets are assembled from the collected relation items. A classifier is computed for the feature sets and a relation score of a query pair of the entities is computed using the classifier.
1. A method comprising:
receiving a query for a plurality of entities of a knowledge base including predefined entities, wherein a graphical representation of the knowledge base includes nodes representing the predefined entities and edges coupling the nodes representing relationships between the predefined entities represented by the nodes, and wherein the relationships are indicated by predefined relation items;
identifying paths in the graphical representation between the plurality of entities of the query, wherein the identified paths include intermediate entities along the paths and path segments coupling the intermediate entities to other entities, and wherein each path segment is associated with a predefined relation item;
collecting the predefined relation items of the path segments of the identified paths;
assembling the collected predefined relation items into feature sets;
applying the feature sets to a machine learning classifier, wherein the machine learning classifier is trained with a training set of relation items along paths between known entities in the graphical representation; and
generating results for the query including a relation score indicating a measure of relatedness for the plurality of entities of the query using the machine learning classifier.
2. The method of claim 1 , wherein assembling the collected predefined relation items into feature sets comprises:
forming generalized path statements from the predefined relation items collected from the path segments of the identified paths; and
providing the generalized path statements as the feature set.
3. The method of claim 2 , wherein forming the generalized path statements comprises:
replacing entity names on each of the path segments with variables; and
interposing the variables between corresponding predefined relation items to form the generalized path statements.
4. The method of claim 1 , wherein the identified paths include positive and negative relations.
5. The method of claim 1 , wherein the predefined entities and relation items in the knowledge base are indicative of medical entities and relations.