Abstract: The existing studies focus a lot on the detection accuracy without considering efficiency, which leads a low performance in practical Online New Event Detection(ONED) domain. This paper analyzes the traditional ONED systems about the low performance limitation, and proposes an improved framework. A lot of storage and calculation overhead can be degraded based on the techniques of setting reasonable parameters and pre-filtering index linking tables, without decreasing detection accuracy. Experimental results demonstrate the enhanced performance of the ONED system.

Key words: Online New Event Detection(ONED), Topic Detection and Tracking(TDT), Information retrieval, pre-filtering

摘要： 现有的关于在线新事件检测(ONED)系统的研究更多地关注如何提高检测的准确率而很少考虑对资源的利用率，使ONED系统在实际应用中存在性能低下的问题。该文分析了传统的事件检测系统存在的性能上的缺点，并在此基础上进行了改进，在基本不降低识别正确率的基础上，通过合理设定技术参数以及对链表索引机制进行预筛选，降低了文档比较过程中的存储和计算开销。实验结果表明，改进的系统提升了检测性能。

关键词: 在线新事件检测, 话题识别与跟踪, 信息检索, 预筛选

CLC Number: 

• TP311.13