Actionable Knowledge Extraction Research In Data Mining

With the development of the technology of data mining in the past decades,date mining has made amazing progress,especially in achieving high accuracy and efficiency.However there is still few works on extracting meaningful decision policy and this methods were limited to simple action models.It is more complex and harder for this models to achieve optimal solution in many real applications.In many applications such as customer relationship management,social network,recommendation systems,advertisement,etc.,users need not only a prediction model,but also suggestions on courses of actions to achieve desirable goals.In existing methods,users often need to spend a lot of time manually finding out really useful knowledge from data mining models,such as actionable knowledge which could get their own goals.The actionable knowledge extraction method of data mining model is a new problem encountered in the subsequent model processing research of the traditional data mining process.In many practical applications,it is necessary to extract action automatically,and tradition methods such as manually finding and image display could not be able meet the growing demand.This paper proposes an actionable knowledge extraction method which extract high-precision and high-efficiency decision actions from data,it is mainly through three aspects that include single attribute value changing,multiple attributes value changing and offline preprocessing-online searching system.The main research include:(1)An approach of combining markov decision process with actionable knowledge extraction.Combining reinforcement learning,Markov decision process is used to extract actionable knowledge.It learns the decision strategy by using autonomous interaction in their own environments,the decision strategy makes maximized the long-term cumulative reward.The method of combining Markov decision processes is used to solve this optimization problem.The experimental results also prove that the change of the single attribute can extract the actionable knowledge which can transforme the state into target state more possibly.(2)A sub-optimal actionable knowledge extraction method is based on state space search.The random forest model is used as the basic model of knowledge extraction,and the problem of knowledge extraction is transformed into an optimization problem.On this basis,it is proved that the optimization problem is equivalent to the shortest path search in the state space.Finally,a sub-optimal state space search algorithm is proposed to solve the problem of the knowledge extraction,and the heuristic function is used to further improve the performance of the suboptimal state space search algorithm.Experimental results also prove that sub optimal state space action knowledge extraction method can guarantee the quality of solution and efficiency of search simultaneously at the same time.(3)Achieving data-driven actionability knowledge extraction by combining learning and planning.By combining machine learning and automatic planning which are two core areas of artificial intelligence the knowledge extraction of data driven action is realized,and the prediction results of the machine learning model are transformed into the ability of action knowledge.Formal definition of action knowledge extraction,and then preprocessing to find the approximate optimal set of target states.Finally,an approximate optimal action sequence is obtained through online search for arbitrary input data.