Visual tracking and analysis of web content dissemination

Abstract

Due to the rising popularity and necessity of information today, it stands to reason that the enormous amount of information needs to be filtered and organized in order for humans to quickly and accurately retrieve the most important information from it. Furthermore, it is also important to track the changes in the information to discover how information about specific topics change over time. This thesis focuses on assessing and evaluating possible machine-learning algorithms in order to help automatically determine the similarity of documents and topics as well as visualization methods that allow the user to intuitively and accurately retrieve and track news article topics across multiple documents. Based on the evaluation of said machine-learning algorithms and visualization methods, a system using fundamental visualizations to promote understandability and the tracking of relationships between words and articles at the expense of requiring more user interaction was proposed. The proposed system has the main goal of helping analysts determine the significance and validity of textual content from multiple documents and sources as well as help determine other relevant documents and the possible origin of specific news article content. The system was then be evaluated through a user study where half the participants used a basic search-engine-based system and the other half used the proposed system. The results of the study was used to assess whether the proposed system can be used as an effective and efficient way for analysts and journalists to discover the relationships between different articles as well as track the provenance and evolution of the topics over time. From the results of the study, the participants using the proposed system did significantly better in terms of time and correctness of the answers in comparison to the participants who used a search-engine-based system.