Discovery of Semantically Related Topics for Given Time Series Data

We propose a method for discovering semantically related news topics for given time series data. As compared with most existing methods which find related news mainly based on the similarity among documents, we use both textual and temporal behavior of the news, and expect to find related news which has some impact on the time series data. To this end, we first classify all of the news into events, and then decide whether an event is semantically related to the given time series based on the textual and temporal correlation between the event and the time series data. Finally, we detect when and how a certain event has impacted the time series data, by analyzing the temporal feature of the event. At the end of this paper, we show the properties of our approach by some experiments.