- 學位論文
The Summarization of Chinese News Articles by Temporal or Themed Sequences
摘要中文新聞之報導-以時間或主題排序
摘要
Most of summarization can extract important sentences, but few of them concern the readability. This thesis proposes a summarization system which considers the sentences coherence and orders the sentences by the news features to facilitate readers to comprehend the news topics. There are three major components of the summarization system proposed in this thesis. First, the event clustering module identifies the events by Self-Organized Map (SOM) and the episodes by Chameleon in every event. Second, the intra-paragraph sequencing module extracts the features of every event in a news topic, and selects the composition strategy either in temporal, themed, or hybrid to compose sentences for an event as a paragraph. Third, the inter-paragraph sequencing module orders the paragraphs and calculates the topic temporal dependence to decide inter-paragraph sequence. It can order inter-paragraph by temporal or by themed based on the feature of topic temporal dependence. Experimental results show that different users may prefer different summaries using different composition methods, and there is a need of the mechanism to order sentences by different methods and choose suitable methods depending on the event’s features either in temporal, themed sequence, or both.