基于文档相关性等级区分优化的网页排序方法  

Page Ranking Based on the Optimization of the Document's Relevance Class

作　　者：余烨[1] 张东站[1] 

机构地区：[1]厦门大学数据库实验室,厦门361005

出　　处：《现代计算机》2013年第7期13-16,共4页Modern Computer

摘　　要：利用机器学习方法自动构建排序模型,在Pairwise方法上平等化每个查询,扩充训练集加大文档不同相关性等级间的区分度和减少不相关文档的噪声影响,利用交叉熵计算误差函数来提高排序算法的性能。在公开数据集LETOR 4.0上的实验结果显示该方法可以提高排序结果的准确率,证明本方法的有效性。Uses the machine learning methods to automatically build a ranking model, just like the widely used Pairwise approach. Based on the Pairwise approach, the new approach is equal for each query＇. It uses the cross-entropy to calculation the loss and selects the Top N related documents to expand the training set to increase the document distinction and reduce the noise impact of the irrelevant documents so that can improve the performance of the ranking approach. Experi- mental studies were conducted using the LETOR 4.0 data set which improved the ranking accu- racies and demonstrated the effectiveness of the proposed method.

关 键 词：网页排序 神经网络 样本对 搜索引擎 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术] TP181[自动化与计算机技术—计算机科学与技术]