Reseach On Trust Model Based On Data Mining Technology

With the development of internet,Trust is playing an increasingly important role in decision support,malicious node detection,bad conspiracydetection and so on.The current research on trust model has been relatively mature,however,the internet environment has become more complicated as the data overload is becoming more and more serious,traditional trust model encounters unprecedented challenges.Data mining technology is an effective way to analyze massive data which converts data into knowledge,the combination of trust model and data mining technology can effevtively solve the bottleneck of traditional trust model.This paper presents a trust model based on data mining technology,the innovation and work are as follows:Firstly,an adaptive FCM clustering algorithm based on improved PSO is proposed.By combining particle swarm optimization and fuzzy C-means clustering algorithm,the optimal initial center of FCM is found,which overcomes the problem of being sensitive to initial center.At the same time,in order to improve optimization ability of PSO,adaptive inertial factor is introduced to make the algorithm adapt well to each satge.In view of prematurity and concussion in later period,corresponding improvement is made.Secondly,a trust model based on PSO-LSTM is proposed.A LSTM neural network is generated based on training set,which replaces traditional trust function and works as a trust classifier.This network can capture continuity and overcomes the problems of gradient disappearance and gradient explosion.Aiming at the performance instability caused by LSTM random initialization weights and thresholds,the PSO algorithm is used to initialize the network optimally.Finally,a trust model prototype system based on data mining technology is proposed.The clustering algorithm is used to pre-process the original training set to ensure the validity and balance of the training set.The window length of LSTM is determined according to the characteristics of the training set to ensure the best effect.Both clustering algorithm and neural network are optimized by intelligent algorithms to avoid falling into local minimum.