Methodologies for Trend Detection Based on Temporal Text Mining

We present two methodologies for the detection of emerging trends in the area of textual datamining. These manual methods are intended to help us improve the performance of our existing fullyautomatic trend detection system [3]. The first methodology uses citations traces with pruning metrics togenerate a document set for an emerging trend. Following this, threshold values are tested to determine theyear that the trend emerges. The second methodology uses web resources to identify incipient emergingtrends. We demonstrate with a confidence level of 99% that our second approach results in a significantimprovement in the precision of trend detection. Lastly we propose the integration of these methods for boththe improvement of our existing fully automatic approach as well as in the deployment of our semiautomatedCIMEL [20] prototype that employs emerging trends detection to enhance multimedia-basedComputer Science education.