基于决策树技术的个性化服务Agent

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在准率。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。

Individual Search Agent Based on Machine Learning

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%.