ENHANCEMENTS OF HOT EVENT DETECTION METHOD FOR DETECTING EVENTS BASED ON NEWS DATA

• Main Author: Shandy
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/53776
• Institution: Institut Teknologi Bandung

Events are things that actually occur, attract attention, have a location and time, and have consequences such as the emergence of news reporting the event. Therefore, scientists have developed various methods to detect events from news documents in the hope that these events can be further utilized, such as processing location and time. One such method is called Hot Event Detection, which was developed by Qi et al. This method was chosen because it is easy to understand, implement and enhance. However, like other event detection methods, it does not support the processing of location and time of events. Therefore, this final project adds these enhancements along with other enhancements such as event merging and keyword processing. Enhancement for keyword processing is divided into analyzing keyword extraction and semantic similarity. In this final project, enhancements to the Hot Event Detection method are implemented through a combination of various techniques. The location attribute is obtained accurately through the news text, the time attribute is obtained from the time of news publication. Event merging is implemented utilizing the DBSCAN clustering algorithm. The keyword extraction analysis uses the TF-IDF dictionary to check the relevance of these keywords and WordNet ontology is used for Semantic Similarity. Enhancements such as event merging and keyword processing successfully improve Hot Event Detection performance through better cluster quality. The enhancement of the addition of the location and time attributes of the event succeeded in obtaining good results with the correct threshold and attenuation configuration (71% precision or 79% recall), but deficiencies were still found that caused the precision or recall to not be higher, for example, clusterization errors.