Abstract
This thesis proposes a method for monitoring a progressive news topic with storyline-based summarization. An ever-increasing volume of news for the progressive news topic, hence we need an incremental clustering to update clusters or events without frequently performing complete re-clustering. There are three major components of the news topic retrospection proposed in this paper. First, we use incremental clustering to identify events. Second, identify the relationship between events and evaluate the relevance of these events with the main storyline. Third, extract the representative sentences and cluster them by Chameleon to compose the summary under the main theme. Experimental results show that incremental clustering has good performance and quality, which can help the news readers comprehend the evolution of the progressive news topics.