Unsupervised Relation Mapping: Going from Text to Schema

The schema of a database models the knowledge content of the database. However, database users often have natural language text documents, e.g., relatively unstructured data, with information related to the database. Understanding the semantics of the text documents entails the identification of entities in the document and the relations (as specified in the schema) that connect the entities. This disclosure describes techniques to find the correct relationship in the schema for a given input pair of entities. Per the techniques, two inputs are extracted from the documents - the pairs (knowledge graph entity, input string) and a set of target attributes, e.g., binary relations between entities and other entities or values that capture particular domain semantics. A list of attributes is returned, ranked by the likelihood that the attributes capture the semantics of the input string regarded as an attribute of the input knowledge graph entity