Semantic cluster formation in deep learning intelligent assistants

This invention describes a way for an intelligent assistant to answer questions by pulling out both text and relevant images from digital documents. It first finds a textual answer from the most important parts of the document, then extracts images from those same pages. The system then identifies the best images to include by comparing them to the text answer, often using captions, titles, or text found within the images themselves, especially from software documentation with screenshots.