目前的多数故事线挖掘研究侧重新闻文献和事件的相似性分析,忽略了故事线的结构化表述及新闻具有的延时性,无法直观地从模型结果看出不同新闻话题的发展过程.为此,提出一种基于贝叶斯网络的无监督故事线挖掘算法.将故事线看成日期、时间、机构、人物、地点、主题和关键词的联合概率分布,并考虑新闻时效性.在多个新闻数据集上进行的实验和评估结果表明,与K-means、LSA等算法相比,该算法模型具有较高的故事线挖掘能力.%At present,most of the research on story line mining focuses on the similarity analysis of news documents and events,while ignoring the structured expression of stories and the delay of news.It is difficult to intuitively see the development of different news topics from the model results.Therefore,an unsupervised storyline mining algorithm based on Bayesian network is proposed,which considers the story line as the joint probability distribution of date,time,organization,person,place,topic and key words and considers the timeliness of news in inside.Experiments and evaluations results on multiple news datasets show that this algorithm model has a higher mining potential than the K-means and LSA algorithms.
展开▼
