Extracting and recommending business processes from evidence in natural language systems

Abstract

A natural language question and answer system analyzes a question to determine key characteristics (such as focus and lexical answer type), and matches those characteristics to business processes from business process repositories. The matching business processes are ranked and at least one is presented as a recommended answer to the user. The system can offer the user a trigger to invoke the particular business process. The analysis includes examining a user profile to determine an attribute relevant to the question, and further includes named entity searching and fuzzy string matching against the business process repositories. Each business process in a repository is designated as either idempotent, non-binding or retrieve-only. The matching can include performing a factorial LDA algorithm on both extracted named entities and latent factors of the business processes in the repositories.