摘要
针对现有的话题跟踪中由于数据不平衡性和话题漂移性造成误报率和丢失率较高的问题,在类别选择的基础上,以话题为单位的邻近报道最大平均相似度代替最大相似度和作为跟踪报道的判定依据,更新话题特征向量,从而提出一种基于改进KNN的话题跟踪算法.所提方法解决了由于数据不平衡和话题漂移带来的跟踪代价较高的问题.实验结果中识别代价降低了7%,验证了算法的有效性.
At present, the miss rate and false rate of topic tracking are high because of the unbalancedness of data and excursion of data. Concerning this problem, based on class selection,the average similarity instated of the sum of similarity to judge which topic is the report belong to, which similarity was calculate with the topic as for unit. Updated topic feature vector, and then proposes a track- ing algorithm which based on KNN algorithm. The proposed method solves the problem of unbalanced data and topic excursion. In the experimental results, the cost was reduced by 7% , showed the validity of the proposed algorithm.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第8期1722-1725,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61363027
61063032)资助
广西自然科学基金项目(2012GXNSFAA053225)资助
