Research On Network Management For Mobile Internet And Some Key Technologies

During the development of 3G network,mobile internet shows the trend of the development of high speed.The performance and security management of the traditional network management have blocked the development of mobile internet.Hence,the research on the network management in mobile internet is of great theoretical and practical value.First,to improve accuracy of the evaluation of quality of experience(QoE),this paper proposes an algorithm to evaluate quality of experience based on the random forest algorithm.For the purpose of the improvement of the accuracy of our algorithm,it uses the random forest algorithm to evaluate QoE,and the ensemble algorithm is applied to it.Because of the subjective of QoE,it uses the fuzzy analytic hierarchy process algorithm(FAHP)to evaluate QoE.Sec-ond,for the purpose of effective construction of the detection model of mobile virus,this paper proposes a closed sequential pattern mining algorithm.It pro-poses a method to judge whether some sequence is a closed sequence,and this method can decrease the number of redundant sequential patterns in the process of mining sequential patterns.Therefore,it can speed up the construction of the detection model of mobile virus.At last,to get useful information from the log in mobile internet,this paper proposes weighted sequential pattern mining and useful sequential pattern mining to mine log files in order to achieve more useful information.The main content of this paper includes:(1)For the improvement of the accuracy of the evaluation of QoE,a new algorithm,called Random Forest for the evaluation of Quality of Experi-ence(RForestQoE),is proposed.It uses random forest and adaboost boost-ing(AdaBoost)algorithms to evaluate QoE.Because of the subjective of QoE,the FAHP algorithm is applied to RForestQoE in order to make the subjective of QoE into our algorithm.Experiments on real data show that RForestQoE can improve the accuracy of the evaluation of quality of ex-perience significantly compared with the decision tree algorithm.(2)For the effective construnction of the detection model of mobile virus,a new algorithm,called closed sequential patterns mining algorithm in time order(CloTSP),is proposed.Based on the characters of closed sequen-tial patterns,CloTSP proposes a method to judge whether some sequence is a closed sequence.It can decrease the number of redundant sequen-tial patterns in the process of mining sequential patterns.Hence,it can speed up the construction of the detection model of mobile virus.Exper-iments on synthetic data show that CloTSP can shorten run-time signifi-cantly compared with closed sequential pattern mining(CloSpan),and it is not affected by variation of attribute numbers with fine augmentability.(3)Because log files contain the information that has different weight,a new algorithm,called ItemSet-interval Weighted Sequences(ISiWS),is pro-posed.In the ISiWS algorithm,a matrix structure,called Transaction Bit Matrix(TBM),represents a sequence.Based on TBM,ISiWS designs search an location operations to find the positions of sequential patterns.It utilizes projected technology to discover weighted sequences,and an approximate sequence match algorithm is applied to calculate support of sequences based on their itemset-intervals.To improve the performance of ISiWS,it proposes a new pruning method to select promising items.In experiments,we analysis the performance of ISiWS from run-time,the number of sequential patterns in results and maximum difference.The results of experiments show that ISiWS can shorten run-time compared to the weighted sequential pattern mining(WSpan)algorithm.Then,they show that ISiWS can decrease the number of the long sequences that con-tain little useful information.Therefore,the experiments shows that ISi-WS can effective mine useful information from log files.(4)For the explosion of the number of the sequential patterns in the results of log analysis,a new algorithm,called Useful sequential pattern min-ing using Hidden markov model(UspHmm),is proposed.The UspHmm algorithm first uses the KMeans algorithm to divides sequential patterns into several clusters based on their position information.Then,a mea-surement about difference between sequences,called Difference Useful-ness Measurement(DUM),is proposed,and the DUMs of sequences are calculated based on Hidden Markov Model(HMM).Based on the DUM-s of sequential patterns,UspHmm can select the sequential patterns that conform to constraints from a large number of the sequential patterns in results.Therefore,UspHmm can solve the problem about the explosion of the number of the sequential patterns in results.In experiments,weanalysis the performance of UspHmm from the number of sequential pat-terns in results,the usefulness of sequences and sensitivity to the num-bers of clusters.The results of experiments show that UspHmm produces a significantly less number of sequences,and the discovered sequences contain the most useful information compared to bi-directional extension based frequent closed sequence mining algorithm(BIDE)and sequential pattern mining algorithm(SPAM).The research content of this thesis has been applied in network manage-ment for mobile internet with machine learning and sequence mining.The CloSpan and RForestQoE algorithms have been applied in practice with an ef-fective solution and project practice guidance.