数据挖掘中的一种次优化动作知识提取方法  

Sub-optimal Actionable Knowledge Extraction Method in Data Mining

作　　者：吕强[1] 李兆荣[1] 陈崚[1] 朱留存[1] 

机构地区：[1]扬州大学信息工程学院,江苏扬州225127

出　　处：《小型微型计算机系统》2017年第5期977-982,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61502412,61379066)资助;江苏省自然科学基金项目(BK20150459)资助;江苏省高校自然科学基金项目(15KJB520036)资助;扬州大学高层次人才科研启动经费项目(5013/137010725)资助

摘　　要：到目前为止,尽管数据挖掘研究已取得了惊人的进展,特别是已达到很高的泛化精度和效率,但是在从模型中提取有意义的决策行动方面仍然只有有限的进展.然而,在许多应用中,如客户关系管理等,用户不仅需要准确预测的模型,也期望能够得到一些建议的行动.提出一种从随机森林模型提取次优化动作知识的方法.其基本思想是首先将提取动作知识问题形式化为一个优化问题,然后证明该问题等价于状态空间搜索中最短路径搜索问题,并提出一种次优化状态空间搜索算法来求解该问题.实验结果表明,该次优化算法在求解效率和提取的动作知识质量上达到了很好的平衡.Although date mining has made amazing progress, especially in achieving high accuracy and efficiency, there is still few work on extracting meaningful actionable knowledge. However, users in many applications, such as customer relationship management, social network, recommendation systems, advertisement, etc., usually need not only a learnt model with accurate prediction,but also advice on actions to obtain an anticipant goal （e. g. high advertise hit rates ）. Only few existing work was trying to extract such action- able knowledge and limited to simple models such as decision tree. In this paper,a sub-optimal actionable knowledge extraction meth- od in random forest model is proposed. This framework first formulates the actionable knowledge extraction （ AKE ） problem to an opti- mal problem, and proves that the optimal problem equivalents to the shortest path finding problem in a state space graph. Then, we pro- pose a sub-optimal state space search algorithm to solve the problem. We also present a heuristic based on the random forest model to improve the efficiency of the algorithm. Our experimental results demonstrate the sub-optimal state space search algorithm can achieves a good balance between efficiency and solution quality.

关 键 词：动作知识提取 数据挖掘 随机森林 状态空间搜索 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]