融合DOM树结构向量的行为类别标签预测模型  

Behavior Category Tag Prediction Model Fused with DOM Tree Structure Vector

作　　者：王宝亮 陈伟宁 潘文采 WANG Bao-liang;CHEN Wei-ning;PAN Wen-cai(Information and Network Center,Tianjin University,Tianjin 300072;College of International Engineers,Tianjin University,Tianjin 300072;School of Electrical Automation and Information Engineering,Tianjin University,Tianjin 300072)

机构地区：[1]天津大学信息与网络中心,天津300072 [2]天津大学国际工程师学院,天津300072 [3]天津大学电气自动化与信息工程学院,天津300072

出　　处：《计算机仿真》2022年第9期257-262,281,共7页Computer Simulation

基　　金：赛尔网络下一代互联网技术创新项目(NGII20170108)。

摘　　要：针对传统的校园网络行为日志分析仅考虑文本内容而忽视结构信息的问题,提出了融合压缩DOM树结构向量的行为类别标签预测模型。模型通过融合网页文本特征向量和DOM树结构向量并使用分类器进行分类,对于网页文本特征向量,利用TF-IDF方法,计算每个词的权重,然后进行加权平均,最终表示整段文本的特征向量。对于DOM树结构向量,提取网页中的全部DOM树结构并进行压缩,降低向量冗杂度,提高向量的表达能力。实验结果显示,与现有方法相比,行为类别标签预测模型能有效提升类别预测准确率。The analysis of traditional campus network behavior logs only considers the text content and ignores the problem of structural information. A behavioral category label prediction model fused with compressed DOM tree structure vector is proposed. The model combines web page text feature vectors and DOM tree structure vectors and uses classifiers for classification. For webpage text feature vectors, the TF-IDF method is used to calculate the weight of each word, and then a weighted average is perform to finally represent the feature vector of the entire text. For the DOM tree structure vector, all the DOM tree structure in the webpage are extracted and compressed to reduce the vector redundancy and improve the expression ability of the vector. Experimental results show that compared with the existing methods, the average increase rate of F1 value is 1.6%. Therefore, the behavior category label prediction model can effectively improve the accuracy of category prediction.

关 键 词：用户画像 标签预测 日志分析 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]