Research On Incremental Learning Of Conditional Preference Networks Based On Sliding Windows

Preference information mining is an important research content in the field of artificial intelligence,and has been widely studied in recent years.CP-nets have a wide range of applications in the field of artificial intelligence,mainly used in collaborative filtering,recommendation systems,product configuration,and so on.However,with the current social development and the rapid growth of digital data,preference data in real-time applications is rapidly generated in the form of data streams and changes over time.Increasingly,data mining needs need to adapt to this streaming data change.Most of the traditional preference learning methods are applied to static data.The theories related to CP-nets dynamic learning are not yet perfect,and the ability to learn CP-nets in the data stream is relatively inefficient.In response to this problem,this paper starts from the basic theory of CP-nets and uses the sliding window model to study the incremental learning problem of CP-nets in the data stream.The main research content of the paper includes the following two parts:(1)Research on incremental learning of CP-nets based on bit sequence tables.When the data stream is continuously updated,the increasing data becomes massive data gradually.Due to the limited memory and CPU,the data may not be stored in large batches,and each data tuple must be ignored after being stored for a certain period of time.The paper proposes a method of incremental learning CP-nets based on bit sequence table to process the preference data stream.In this method,a sliding window and sliding time are set as the data storage structure,and the data stream is input in batches and can be increased or decreased.The bit sequence table is used to incrementally store binary preference information,and the CP-nets learning algorithm is used to learn the preference relationship between attributes.The experimental results of the algorithm on simulated data sets and real data sets show that the algorithm can obtain a more accurate CP-nets model based on user preference information.Compared with traditional algorithms,the algorithm has lower time complexity and higher algorithm efficiency under the premise of ensuring accuracy.(2)Research on CP-nets incremental learning based on MMPC algorithm.Due to the randomness of user needs and observation data,there is a problem of attribute diversity in training samples.This paper proposes an incremental learning method of CP-nets based on the MMPC algorithm that can mine preferences in dynamic data.This method reduces the incremental attributes of the dynamic data samples through the sliding window and establishes the initial parent-child relationship structure,and then uses the incremental scoring algorithm to optimize the structure to obtain the optimal CP-nets structure.The experimental results of the algorithm on simulated data sets and real data sets show that the algorithm has certain accuracy and efficiency,and can obtain the optimal CP-nets structure in a limited time.In summary,the research on incremental learning of CP-nets based on chain data structure and the CP-nets incremental learning based on MMPC algorithm are both incremental learning of CP-nets through a sliding window model.The former stores the preference between the learning attributes of the preference data through the bit sequence table storage structure,and the latter learns the preference between the attributes through the incremental learning method based on the MMPC algorithm.Both methods can obtain accurate and optimized CP-nets model with lower algorithm time complexity.