Extraction and Visualization of Temporal Information and Related Named Entities from Wikipedia

This paper addresses our process in generating a tool that extracts named entities and events from a document and visualizes them in ways beneficial to someone learning about the topic. The ultimate goal is to present a user with many of the key events and their associated people, places, and organizations within a document that will quickly give users an idea of the contents of an article. For testing, we use a set of historical Wikipedia articles which focus on topics such as the American Civil War. These articles have high occurrences of all types of named entities along with many events with clearly defined time spans. For initial named entity extraction, we incorporate the Stanford NLP CRF into our project. In recognizing location names in this subject area, it only achieves an f-measure of 57.2%. The list of locations is geocoded through Google Geocoder and will be disambiguated through a tree structure in the future. A final f-measure of 79.1% is determined which represents the precision and accuracy of our package in successfully grounding the extracted locations. The grounded locations are then grouped with other named entities related to an event through sentencelevel association. Visualization is currently done through Google Maps and the Timeline SIMILE project developed at MIT. We plan to add the capability to geospatially and temporally refine article searches in Wikipedia and make our tool usable on other online corpora.