잡음 정제를 통한 고응집도 군집과 후반부 요약을 이용한 효과적인 뉴스 요약

초록

News articles often present high redundancy due to multiple outlets covering the same events. Clustering similar articles and summarizing them allows users to quickly understand trends and focus on articles requiring detailed reading. This paper proposes a method that begins with HDBSCAN clustering and applies Mean-Shift and noise refinement to enhance cluster cohesion. For summary generation, we introduce a strategy that selects representative summaries from the latter half of articles within clusters, evaluated using a summary metric. These are combined with lead summaries from the articles’ first-halves. Experiment results demonstrate efficiency gains in news summarization with improved clustering performance and summary quality through the inclusion of second-half summaries.

키워드