Self-learning user interface with image-processed qa-pair corpus

Abstract

A machine-learning component of a conversational user-interface system retrieves records of human question-and-answer sessions from heterogenous sources. A sequence of image-processing operations converts each source into a set of grayscale images. Each image is segmented into disjoint sections as a function of the textures, edges, and contours that make up the image's content. Each section is tagged with a vector of parametric values that identify characteristics of the section's content from which may be inferred semantic meaning. A cognitive function intelligently analyzes the vectors to classify each section as containing a question, an answer to a question, non-textual media, or other types of content. Another cognitive function merges and organizes the sections into question-answer pairs and the vectors associated with each pair are stored in a corpus that is submitted to the self-learning user interface during a machine-learning training session.