【作者】 王辉； 左万利；

【Author】 Wang Hui Zuo Wanli (Jilin University,Changchun 130012)

【机构】 吉林大学；

【摘要】 本文给出三种启发式规则用于获取根集文档的质心向量。通过使用给定的正例文档,本文首先构建一个与其对应的质心向量,然后利用它提取未标识数据集中的可靠正反样例。文本分类器使用这个逐步扩展的训练集进行增量训练。在 Reuters-21578数据集上的实验结果表明,Sum 质心向量可以协助分类器提高其分类精度。最后,本文通过实验验证了论文中提出的方法比 PEBL 算法更高效。更多还原

【Abstract】 In this paper,three heuristic rules were brought forward for obtaining a centroid vector corresponding to a root set.Using positive training examples,this article first built the centroid vectors and then utilized them to extract reliable positive and negative examples from an unlabeled data set.A text classifier was trained iteratively using the expanded training data.Experimental results on Reuters-21578 data set demonstrate that classification accuracy is improved with the aid of our proposed Sum centroid vector.In the end,this paper verifies that our proposed technique is more effective than the PEBL algorithm.更多还原