Trend Knowledge Discovery In Sequence Data

The trend features are important objects in sequence data mining research.The basic research issues include the expression and recognition of trends,the similarity measure of trends,the trend clustering,the trend forecasting and so on.This paper focus on the trends knowledge discovery problem,studied the description and identification of trend.The specific research results are as follows.(1)A trend expression and identification method based on inertia are proposed.Reveals that "the preorder nodes have a significant influence on the following ones" in the sequence,which is the connotation essence of the trend,named "inertia",helping to redefine the concept of trend.In the case of considering the "overlapping trend",an "inertial test" algorithm is proposed to automatically identify the trend contained in the sequence.(2)Aiming at the problem that it is difficult to directly measure the similarity due to the difference of the length,scale and shape among the trends,a trend similarity calculation algorithm based on multiple warping distance is proposed.Based on the structural similarity of the sequence data,the method performs the operation of "shifting","order warping" and "value warping" with the help of the mathematical characteristics of the movement function,and transforms the trend into the closest form."Isomorphic evolution process "and" inverse evolution process" concept are proposed,combined with the theory of sequence key points,reduces the complexity of similarity calculation.On this basis,an improved curve integral distance is used to complete the trend similarity calculation.(3)Aiming at the problem of dimension selection,discretization and boolean in multi-dimensional association rules discovery,a multi-dimensional association rule discovery method based on trend events is proposed.By changing the transaction into dimension,combining the results of trend identification,the"sequence of trend events" is constructed.The "hierarchical clustering method based on multiple warping distance" is used to merge the number of trend events.Combing the types of "time lag",this paper gives a method to calculate the optimal lag coefficient in the application of association rules discovery.