基于决策树技术的个性化服务Agent  

Individual Search Agent Based on Machine Learning

作　　者：陈红英[1] 杨宜民[1] 

机构地区：[1]广东工业大学自动化学院

出　　处：《微电子学与计算机》2006年第3期8-10,15,共4页Microelectronics & Computer

基　　金：广东省自然科学基金资助(970237)

摘　　要：文章采用了机器学习技术,从提交更精确地反映用户兴趣的检索串入手,研究如何提高搜索引擎查准率。文中采用决策树方法进行学习,对决策树方法应用于网页检索中出现的几个问题:缺少属性值的训练实例处理问题;如何使不同权值的属性,具有不同的表现力问题;树的重建问题;过度拟合问题;扩充检索串返回的网页时,属性值的取舍问题等进行了分析和研究,给出了具体的解决方法。性能提高后的决策树,用验证集检验,正确率由70%提高到75.4%,较好地学习到了用户的兴趣。This paper uses methods of machine learning to enhance the exactness of search engine. It aims to submit a character string which is similar to users＇ interesting. Decision-tree is used to learn the users＇ interesting. There are several problems using decision-tree in web pages： how to deal with when the samples have not enough characters; many characters have different importance, how to reflect it in decision-tree; when and how to recreate the decision-tree; perfect tree problem; when we delete character string from decision-tree, which should be chosen; These situations are investigated and solved. Performance of new decision-tree is： validity has enhanced from 70% to 75.4%.

关 键 词：搜索引擎 个性化服务 决策树算法 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]