Evidence based fuzzy single document analysis

Abstract

Human beings can extract meaningful information from single documents and can even summarize them depending on their interest. Computers on the other hand are used in increasingly large number of documents to process them. Even with vast number of documents we are drowning in, there will always be need of important documents which occur singly or in small numbers. For example, it is unlikely that a statistically significant number of airplanes will collide with tall buildings. So, to analyze and extract significant information from reports or documents related to this kind of scenario, it requires subjective analysis of data. This thesis uses structural fuzzy technology, subjective logic and higher order singular value decomposition to extract information from single documents, or from a small collection of documents. The idea is to analyze the language and syntax used in the document to remove uncertainty, increase confidence, and improve the reliability of decision-making which can have many applications including in the media and intelligence gathering. This is illustrated through the generation of extractive summaries using these techniques. The results are good, and validated by comparing document summaries using my techniques with human generated summaries and other machine generated summaries. My summaries are more similar to human summaries than the rest, and this is the major result captured in this thesis.