《Table 4 Weights of parameters and named entities》
本系列图表出处文件名:随高清版一同展现
《Hierarchical clustering based on single-pass for breaking topic detection and tracking》
To find optimal threshold,the algorithms are compared against several competing methods with different thresholds.Weights of parameters,title,named entities are listed in Table 4.Weights of non-title and nonnamed entity are 1.These weights are selected according to different characteristics and significance of parameters in experimental data.For example,a great quantity of irrelevant reports are released in the same time,therefore,the importance of temporal expressions is diminished slightly,which is assigned to 0.9.Effectiveness evaluation is performed using experimental data in Section 3.1.CMissand CFalseare predefined as 0.8and 0.2.Results of miss rate,false positive rate and error cost are shown in Figs 2,3,4.Error cost is used to evaluate error degree of topic detection and tracking.The smaller the error cost is,the better the algorithm performs.By error cost analysis,the performance of 4algorithms is the best when the threshold is in the interval[0.25,0.35].Therefore,experiments are performed near this interval.
图表编号 | XD0020606200 严禁用于非法目的 |
---|---|
绘制时间 | 2018.12.01 |
作者 | 李风环、Zhao Zongfei、Wang Zhenyu |
绘制单位 | School of Computers,Guangdong University of Technology、School of Software Engineering,South China University of Technology、School of Software Engineering,South China University of Technology |
更多格式 | 高清、无水印(增值服务) |