CAJ | 학술논문

Web日志分析预处理阶段的一个重要步骤是事务分割，把一系列的操作合并成完整的事务。文章提出了一种基于统计模型的事务分割算法。统计模型上采用了网站链接的边界访问概率和事务内的链接转移概率矩阵。算法将包含了若干个事务的日志片断作为一个整体，采用最大似然的准则来获得整体最优划分结果。在一个内网办公系统日志采集的数据集上，文章采用True Positive Rate和Precision两个统计量来比较不同事务划分算法的性能，结果证明了文章提出的算法的有效性。
Web일지분석예처리계단적일개중요보취시사무분할，파일계렬적조작합병성완정적사무。문장제출료일충기우통계모형적사무분할산법。통계모형상채용료망참련접적변계방문개솔화사무내적련접전이개솔구진。산법장포함료약간개사무적일지편단작위일개정체，채용최대사연적준칙래획득정체최우화분결과。재일개내망판공계통일지채집적수거집상，문장채용True Positive Rate화Precision량개통계량래비교불동사무화분산법적성능，결과증명료문장제출적산법적유효성。
Transaction identification is very important in pre-processing stage of Web log data mining. It combines several log accessing entries into a meaningful operation. The paper proposed a transaction identiifcation algorithm based on statistical model. The statistical model used is marginal statistics of URL visiting and transition probability matrix of two adjacent URL within a transaction. The algorithm ifnds a global optimized solution with maximum likelihood. On a dataset collected from an internal ofifce system, using true positive and precision, experiment result proves the effectiveness of the proposed algorithm.