跨文档事件检测算法

摘要

为了从海量文档中检测出特定事件，提出了一种跨文档事件检测的模型和算法。首先从文档中提取信息要素，包括主体、时间、地点、主题。然后以信息要素为基础对文档建立共现词网络图，并采用4W向量描述待检测事件，即从逆向的角度考虑，对共现词网络图进行带约束条件的深度优先搜索，寻找图中定长的环。最后判断这些环中的节点是否包含待检测事件的信息要素以实现事件的检测，并以环中节点反向获得与事件相关联的文档。实验表明该算法能从文档库中检测出事件，与其他算法相比，能同时获得较高的准确率和召回率。%In order to detect event from huge amounts of documents , it proposes a cross -document event detec-tion model, designs the algorithm .It extracts essential elements of information from all documents , builds word co-occurrence networks based on the essential elements of information , uses 4W vector to represent the event . This algorithm finds fix length acyclic with depth -first search in word co -occurrence network and constraint condition.It conducts the event detection process (EDP).In EDP the key point is to decide whether the node in the acyclic includes essential elements of information of events which are pending for check , and get the event-related documents by the essential elements in detected acyclic .Experiments show that the algorithm detects e-vents in document corpus , and is better than other algorithms in precision and recall rate .