面向Web 新闻的事件多要素检索方法

针对用户获取事件类信息的需求,在分析Web 新闻特征、事件多要素检索特点的基础上,研究了面向Web 新闻的事件多要素检索方法.首先,提出了面向Web 新闻的事件多要素检索模型;然后,使用BNF(Backus-Naur form)形式化定义了事件多要素查询项;最后,结合事件的动作要素、Web 新闻标题的重要性及事件项与约束项之间的距离,提出了事件查询项与文档相关性的计算方法.设置了16 个事件多要素查询项,基于Baidu 搜索引擎对P@n 指标进行了实验分析,所提方法得到的平均P@10 结果为0.87,平均P@20 结果为0.83.对16 个事件查询主题,通过人工标注语料的方法对F-measure 指标进行了实验分析,所提方法得到的平均F-measure 为0.74.结果表明,所提方法对事件多要素的检索较为有效.

Web News Oriented Event Multi-Elements Retrieval

To meet the demand of effectively acquiring event information, a method of Web news-oriented event multi-elements retrieval is studied through analyzing characteristics of Web news and event multi-elements retrieval process. Firstly, a model of Web news-oriented event multi-elements retrieval is proposed. Secondly, event multi-elements query terms are formally defined by using the BNF (Backus-Naur form). Finally, incorporating the importance of event action element, Web news title and the distance between event terms and constrained terms, a method of computing the relevance between query terms and the document is proposed. Sixteen event query topics are created to implement the experiments. With the proposed method, this paper evaluates the index P@n based on the Baidu search engine, getting average P@10 of 0.85 and average P@20 of 0.83. This paper also evaluates the index F-measure through manually labeling the corpus with same method, obtaining average F-measure of 0.74. The results show that the proposed method offers more effective performances.