Semantic correlation between documents and taxonomies

点击翻译按钮获取中文摘要

We propose to develop smart algorithms for document generation, by innovating in the field of natural language processing and document intelligence. We envision the next generation of business applications able to parse and understand documents, to compose documents automatically, and to respond intelligently to voice commands. Our industrial partner, Koneka Inc. has a document automation platform that generates documents by assembling clauses (content-blocks) together based on a set of user inputs. The database consists in a library of classified clauses which are interrelated by rules, usage patterns, document types, positioning information, and so forth. The platform relies on taxonomy trees to categorize clauses. In this project, in the first six months, we already designed and implemented a taxonomy framework that supports a weighted synonym approach to recognize that some terms are equal (i.e., "SQL Server" and "Microsoft SQL") and that other terms have a high correlation without being equal (i.e., "Linux" and "Unix"). With this work, we are able to measure the correlation between an input text and a database of objects that have been tagged with terms of the taxonomy tree. In this second part of the project, we want to extend and connect with another important component. We will develop a parser that can extract the most pertinent text segments based on a keyword dictionary in order to extract well-defined concepts from requests for proposals (RFPs). The algorithm computes a score for the relevance of the text segments. Our goal is to give labels to the extracted text segments and to analyze their content in order to group them by topic text portions in order to facilitate content analysis for the user who needs to match a RPF with a database of competencies.