The Research Of CP-nets Structure Learning On Streaming Data

Preference learning is an important research content in many scientific fields such as decision theory,operations research and modern computer science theory.As an intuitive expression tool for qualitative preferences,Conditional Preference networks(CP-nets),have become a hot issue in artificial intelligence research.Its main research content includes structural learning,modeling,and reasoning.At present,the research on CP-nets structure learning mainly focuses on the mining of static databases,and the preferences of user agents are basically unchanged.However,in real applications,user preferences often change dynamically over time and most CP-nets learning methods are difficult to quickly and efficiently mine streaming data in real-time applications,especially when the transaction volume is large,the time to learn the CP-nets structure will also increase.Aiming at the current problem of learning CP-nets structure with streaming data,this paper proposes a method based on time-sensitive sliding window and based on inverse matrix and frequent pattern tree to learn CP-nets structure.The main research work is as follows:(1)Research on CP-nets structure learning based on time-sensitive sliding windowFor dynamic preference data stream,this paper proposes a method to mine conditional preference relations and dynamically learn the structure of CP-nets based on a time-sensitive sliding window model.The method includes a storage structure for obtaining all possible preference relations and a data structure for cumulative counting of preference relations.By comparing the size of the basic block and the sliding window,the conditional preference relationship is inserted and updated in real time,and finally the CP-nets structure is generated.On this basis,the TSCPL algorithm based on time-sensitive sliding window learning CP-nets structure is proposed,and the effectiveness of the TSCPL algorithm is verified through experiments.(2)Research on CP-nets structure learning based on inverted matrix and frequent pattern tree methodWith the increase of attributes in the preference database,this paper adopts the IMFP algorithm based on reverse matrix and frequent pattern tree to mine conditional preferences on streaming data and learn CP-nets.The algorithm uses the transaction layout of the reverse matrix to establish FP-Tree for candidate preference items,which can quickly track the preference information related to a certain attribute in the streaming data,and improve the efficiency of mining condition preference.Finally,it is verified by simulation data experiment and real data experiment.Compared with other methods of learning CP-nets structure,the IMFP algorithm can quickly obtain accurate CP-nets,and shows good performance in multi-attribute fields and large transaction databases.In summary,the TSCPL algorithm based on time-sensitive sliding window learning CP-nets structure is an incremental learning method of CP-nets structure,which can dynamically learn preferences in streaming data.In addition,in the field of multi-attribute,the use of IMFP algorithm can quickly obtain accurate CP-nets structure.Finally,the content of the full text and future research directions are summarized and prospected.