Two-tier similarity model for story link detection
Two-tier similarity model for story link detection
复制标题
DOI：
10.1145/1871437.1871539
复制
发表时间：
2010-10
期刊：
影响因子：
--
通讯作者：
Tadashi Nomoto
中科院分区：
文献类型：
--
作者：
Tadashi Nomoto
本文提出了一种新颖的故事链接检测方法，其目标是确定一对新闻故事是否链接，即谈论同一事件。目前的工作标志着与先前工作的背离，因为我们在两个不同的文本组织水平（文档及其集合）上测量相似性，并结合两个水平的得分来确定故事的联系程度。在TDT-5语料库上的实验表明，我们称之为“两层相似性模型”的当前方法轻松击败了传统方法，如Clarity增强KL发散，同时在不同语言中表现稳健。
The paper presents a novel approach to story link detection, where the goal is to determine whether a pair of news stories are linked, i.e., talk about the same event. The present work marks a departure from the prior work in that we measure similarity at two distinct levels of textual organization, the document and its collection, and combine scores at both levels to determine how well stories are linked. Experiments on the TDT-5 corpus show that the present approach, which we call a 'two-tier similarity model,' comfortably beats conventional approaches such as Clarity enhanced KL divergence, while performing robustly across diverse languages.