Knowledge Discovery from Free Text: Extraction of Violent Events in the African Context

ABSTRACT In today’s Internet age, millions of articles in an unstructured format are created every second. Most of this information is available in the web in the form of free text articles. These unstructured free texts are difficult to process and almost impossible to use in decision making, especially those based on analysis of long term data. This research project tried to address knowledge discovery from free text through the extraction of Violent Events along with their attributes. Media monitoring tools such as the Africa Media Monitor can gather thousands of news articles every minute. These articles can be clustered and categorized. However, due to the lack of structure and the heterogeneity of information sources, access to these huge collections of information has been limited to browsing, searching, and reading through articles. To cope with that enormous amount of available data, Information Extraction Systems are needed whose goal is to extract structured information from unstructured documents or free texts. This research tried to address event extraction in one domain. The developed prototype successfully managed the extraction of the 5W (Who, What, Whom, Where, When) but could not manage the 1H (How) due to programming challenges and time constraints.