基于操作序列挖掘的OLAP查询推荐方法

Operation sequence mining based OLAP query recommendation method

针对联机分析处理(OLAP)操作复杂导致的用户使用效率低下问题,提出基于操作序列挖掘的OLAP查询推荐方法.首先从多维表达式(MDX)查询语句记录中提取整数数列形式的查询序列,再利用PrefixSpan方法对查询序列进行频繁序列模式挖掘,并基于挖掘出的模式及其子模式建立概率矩阵,最后通过搜索与用户当前查询操作或查询序列匹配的候选模式对其下一步查询操作进行预测,并将预测结果按概率大小分级推荐.在7位OLAP专业分析人员的查询分析日志数据集上对提出的查询推荐方法进行性能评价,实验结果表明:使用用户相关模型前5推荐内容的平均正确率为92.20%,其中第1推荐的平均正确率为77.06%.

An operation sequence mining based OLAP（online analytical processing） query recommendation method is proposed to counter the low efficiency problem caused by the complexity of OLAP query operations.First,query sequences in the form of numerical array are extracted from continuous MDX（multidimensional expression） query operations.Then,the PrefixSpan mining algorithm is exploited to obtain the frequent sequential patterns from query sequences,and a matrix of probabilities is established upon mined patterns and their sub-patterns.Finally,the next operation of current user is predicted by searching candidate patterns matched with the user＇s query operation or query sequence,and the prediction results are ranked according to the magnitude of probabilities.The performance of the proposed query recommendation method is evaluated with an OLAP query operation dataset recorded by seven professional OLAP users.The results show that with user-specific recommendation models,the average accuracy rates of the top five recommendations and the first recommendation are 92.20% and 77.06%,respectively.