The Research Of Acyclic CP-nets Learning Based On Degree Of Attributes Dependence And Evolutionary Programming

Preference processing is a hot research field in the field of artificial intelligence.Conditional Preference networks(CP-nets)is a graph model that can represent the preference relationship between variables and variables.Its main research contents include representation,reasoning,aggregation and learning of CP-nets.In recent years,the study of CP-nets is attracting more and more attention and has achieved a series of results.The idea of CP-nets learning is to extract the preference structure and multiple preference parameters by observing the user's query record.However,the related methods of CP-nets learning have not been perfected,and it is difficult to obtain an accurate graph model of CP-nets,especially when the amount of data increases sharply,the solution time is also greatly extended.In view of the existing research problems,this paper proposes a structural learning problem for CP-nets based on degree of attribute dependence and evolutionary programming methods.The main research work is as follows:(1)Research on CP-nets structure learning based on attribute dependence in preference databaseThis paper proposes a CP-nets structure learning method based on the preference database,including the following aspects: The CP-nets learning problem is formalized based on the preference database.The calculation method of dependency between attributes is proposed.Based on this,an ADLA algorithm for generating CP-nets structure from the preference database is proposed.After obtaining the structure of CP-nets,a learning method of ring elimination processing and condition preference table is also proposed.Based on this method,a more accurate CP-nets structure can be obtained,and the effectiveness of the algorithm is verified by experiments.(2)Research on Structure Learning of CP-nets Based on Evolutionary Programming MethodsAs the attributes in the preference database increase,the time complexity of the ADLA algorithm increases exponentially,making it difficult to learn the exact CP-nets structure.To solve this problem,the evolutionary programming method is used to learn the CP-nets structure to obtain its approximate optimal structure.The main work includes: Based on the degree of attribute dependency,the evaluation method of CP-nets structure is proposed.Combined with the evolutionary programming method to optimize the search process,a new Evaluate-EP algorithm for learning CP-nets structure is proposed.Through experimental verification,while obtaining the approximate optimal CP-nets structure,the running time is significantly reduced.In conclusion,the ADLA algorithm based on attribute dependency can certainly learn a compact acyclic CP-nets structure in preference database.At the same time,in view of the computational difficulties caused by the increase of the number of attributes in the database,we use the Evaluate-EP learning method to obtain the approximate optimal CP-nets structure in a relatively short time.Finally,the main contents of this paper are summarized and the future research directions are given.