Automating Text Processing Using Analytics - Automating text classifications and financial news parsing

TLDR: This thesis presents an automation approach using machine learning in which it is shown how to improve text classification performance, and how this approach can reach practically acceptable performance levels even in certain abstract classification problems.

Abstract: Automating repetitive processes and replacing manual tasks with automated systems is an area of research that will greatly impact and transform our lives during the 21st century. Automation comes in many forms and we are now at the start of an era, after which repetitive non-creative tasks will be handled mainly by machines. In this thesis, two analytics approaches are presented that can be used to automate text processing tasks. The first is an automation approach using machine learning in which we show how we can improve text classification performance, and how we, through these improvements, can reach practically acceptable performance levels even in certain abstract classification problems. We test the developed methods on problematic web content categories, such as violence, racism, and hate. The second is an automation approach that uses network analytics to automatically process texts. We use this approach to automate processing of financial news and to automatically extract new information. We show that through automating the process, we can extract company specific sentimentrisks that a person would not identify simply by reading the news articles. Lastly, we show that the risks we have extracted can be used to identify companies that are at higher risk of stock price decrease.