融入事件知识的新闻事件对比聚类方法OA北大核心CSTPCD

News event clustering aims to mine several event clusters of different topics from massive news texts.At present,event clustering is mostly based on text semantic representation,but ignoring the guiding role of event knowledge.Due to the iterative pro-cess of representation learning and target clustering,it is easy to cause error accumulation.It can only deal with offline tasks,which limits the processing of real-time news data,and to solve the above problems,this paper proposes a contrastive clustering method of news events incorporating event knowledge.On the basis of text representation,this method incorporates event key information to enrich event representation.The cluster label is used as the representation,and contrastive learning is performed at the instance level and the cluster level.The representation and cluster assignment are jointly learned in an end-to-end manner to realize the clustering of data streams.Experimental results show that the proposed method improves by 3%compared with other baseline models.