基于分类方法的Web站点实时个性化推荐

提出一种新的基于分类方法的实时个性化推荐方法，该文首先根据用户访问事务文法生成序列访问事务集，用于得到每个用户访问的序列特性并且便于分类器进行分类，然后利用该事务集训练一个多类分类器，作者通过推荐引擎得到每个用户的当前访问序列和用户当前请求页面，然后把该序列送入分类器进行分类，以得到用户的下面一些可能访问的页面，这些推荐页面的地址被附加到用户当前请求的页面的底部由推荐引擎返回以进行推荐，在这种方法中，用户不需要注册信息，推荐不打扰用户，可以为用户提供实时个性化的服务，实验表明这种方法是成功的。

Real Time Personalization Recommendation Based on Classification

To using user path characteristics to provide the personalization recommendation, this paper presents a new approach of real time personalization recommendation based on the classification approach in web usage mining. The sequence access transaction set is generated by the user access transaction grammar defined by this paper. The grammar is educed from the regular grammar and can get the sequence characteristic from the user access transaction and the result can facilitate the classification. The set can be used to train a classifier that can process the multiple classes. The k -nearest neighbor classification approach is chosen. Authors use recommendation engine to identify the active user, his current access sequence, and his next request. The sequence and the next request are input into the trained classifier to get the new possibly accessed web page. The recommendation web page address is annexed to the requested page and it is returned to the user by the engine. Each user is provided personalization web recommendation. Authors' approach does not require the profile information about the user and the recommendation process will not disturb the user. It can provide the real time personalization recommendation and the experiment manifest the approach is successful in speed and precision.