Abstract: Web-page classification can be utilized to categorize massive web-pages and thus can be utilized in lots of areas. There are quite a few existing automatic web-page classification methods, among which there is large performance improvement space for the commonly-used web-content-based method, due to the impurity of page content. In this paper, based on query log, a novel web-page-classification method NQPC（Novel Query log-based web-Page Classification） is proposed. Its novelty is that: a low-dimensional feature vector extraction method is proposed to avoid the “curse of dimensionality”; web-page classification is based on high-quality query log, which has purer content than web-page content; a filter method is proposed to improve the classification accuracy. Experimental results show that the web-page-classification method has excellent performance, which gives it good application prospects.

Key words: query log, web-page classification, machine learning, text classification, feature extraction

摘要： 网页分类可对海量网页进行分门别类，可应用于许多方面。现存的网页自动分类方法较多，其中常用的基于网页内容的方法由于网页内容的不纯，导致其存在较大的性能提升空间。基于查询日志，提出了一种新型的网页分类方法NQPC。该方法提出一种低维特征向量抽取方法，从而避免“维度灾难”；基于优质的查询日志进行网页分类，查询日志相对网页内容而言，具有内容较纯的优点；提出一种提升分类准确率的过滤方法。实验结果表明，提出的网页分类方法具有优异的性能表现，使其具有良好的应用前景。

关键词: 查询日志, 网页分类, 机器学习, 文本分类, 特征抽取