A New Online New Event Detection Algorithm Based on Event Merging and Event Splitting

The current Online New Event Detection (ONED) algorithms based on the elements of news can effectively detect new events; however, these methods have two limitations: First, events are tend to be split into several small similar events at the beginning of events; Second, those events about the same describing object but different subject are opt to be merged into one large event. This paper proposes a new improved ONED algorithm that could effectively solve the above limitations; the new algorithm makes two improvements at the basis of the current ONED algorithms: First, at the beginning of events, it not only compares stories with detected events but also compare events with events to make sure that whether certain events should to be merged; Second, it makes a secondary analysis of those events those last for a long time to see whether they should be split. The experimental results show that comparing to the current ONED algorithms the new algorithm can effectively reduce the miss probability and false-alarm probability.