Linguistic transformation based relationship discovery for transaction validation

Abstract

From transaction data, a graph modeling a set of relationships between transaction elements is constructed, a node of the graph representing a transaction element, an edge of the graph representing a relationship between two transaction elements. Using a part of speech analysis model, a subset of the nodes is tagged with a corresponding first part of speech tag. An alternative part of speech tag is generated for a tagged node. A set of tagged nodes is grouped into a word group including at least one tagged node having an alternative part of speech tag. The word group is validated against a grammar of a natural language. Responsive to the validating determining that the word group conforms to the grammar, a set of additional nodes representing the word group is added to the graph. Using the modified graph, a transaction in the transaction data is validated.