信息自适应过滤的效用指标优化方法

Optimization Method of Adaptive Filtering Utility Target

自适应过滤是文本检索会议(TREC)过滤任务的重要子任务,也最接近真实的环境。对评测指标的优化是自适应过滤任务中非常重要的研究方向。论文以TREC的评测指标为目标函数,对在阈值调整中的极大似然估计法和局部优化法进行了比较分析,提出了结合极大似然估计法的局部优化方法,克服了采用单一方法的缺点,实验结果表明这个方法对提高过滤性能是有效的。

TREC plays an important role in the research area of text retrieval and filtering,and the adaptive filtering track is one of the most important tasks in TREC.Evaluation measure optimization with threshold is a challenging topic in adaptive filtering.In the paper,TREC evaluation measure is regarded as our target function,and it is compared with maximum likelihood estimation and local target optimization on the threshold tuning.As a result,a local optimization approach combining maximum likelihood estimation is presented,and it overcomes the shortcomings of above single method.Based on experiments,it proves that this approach can improve effectively the quality of text filtering.