Research On Dynamic Acquisition Of Context Mobile User Preference Based On Telecommunication Data

3G network not only enhances the data rate of transmission, but also supports a variety of forms of media data. The application of the cloud computing makes the user can get the storage resources, computing resources, and the corresponding hardware and software resources according to his requirement by the terminal. In addition, the mobile terminal is easier to carry, so the mobile user can finish the complexity operation by the mobile terminal with simple function wherever he is. On the other hand, due to the rapid development of the triple play, pervasive computing and the Internet of Things technology and application, the mobile communication network nutures the information services of the traditional Internet in the process of gradual integration with Internet. It can provide more colorful mobile network services for mobile user than the traditional communications business. However, the mobile terminal has some drawbacks, such as the small interface, difficult input and output, power of short duration and so on. Therefore, in order to satisfy the personalized requirements of mobile user, how to obtain timely the accurate mobile user preference has become the hotspot of the academia and industry in recent years.Compared with the desktop users, it’s more obvious that the mobile user preference is affected by context. In recent years, the researchers add the context to the process of mobile user acquisition to accurately position the mobile user preference. Although the introduction of context can accurately position the mobile user preference, it also brings some challenges to the acquisition method of context mobile user preference. After the introduction of context, the amount of the mobile user preferences will increase in direct proportion to the amount of the context instances. Therefore, the response time of obtaining the mobile user preference will increase and can’t satisfy the personalized needs of the mobile user. In addition, the introduction of context makes the user-item matrix extend to user-item-context matrix. This exacerbates the sparsity probems and drops the accuracy of the mobile user preference prediction. In order to obtain the real-time and accurate mobile user preference, we utilize the available data to acquire the mobile user preference according to the characteristics of the mobile network. In the paper, the research includes how to calculate the trust between the mobile users according to the available data in mobile network, how to adaptively learn the context mobile user preference, how to relieve the affect of the sparsity and cold start problem of collaborative filtering to the mobile user preference prediction and how to online acquire the context mobile user preference. Based on the above study, we obtain the following research results.(1) In the mobile network, the implicit method obtaining the trust mainly performs some simple computations according to the communication behaviors between the mobile users. The existing methods ignore the affect of the context information, the social influence of mobile user as well as the similarity between mobile user preferences to the trust. In addition, these methods rarely make the in-depth study to the propagation distance of the trust. To address the problem, in the paper, a calculation method of the trust based on the telecommunication data is proposed. This proposed method takes into account the affect of the context mobile user behavior, the social influence of mobile user and the similarity between context mobile user preferences to the trust. In the method, the input data includes the communication behaviors and spending time between the mobile users, the mobile network service used by the mobile user as well as the corresponding context information (time, position). Firstly, the direct trust is calculated according the mobile user behavior and obtained the weight of the corresponding context. Based on the methods existed and the theory of the six degrees of separation, a calculation of the propagation distance of the trust is proposed and we can compute the indirect trust according to the propagation distance. Secondly, we can construct the mobile social network according to the obtained trust and employ the cohesion subgroup knowledge to divide the constucted mobile social network. Based on the obtained community structure, a method to calculate the social influence of the mobile user is proposed. Finally, we calculate the similarity between the context mobile user preferences and merge the obtained similarity with the obtained trust and the social influence.(2) In the mobile network, it has the higher demands for the performance of the personalized mobile network services. However, the research existed is unable to update the context mobile user preference adaptively and provide the real-time, accurate personalized mobile network services for mobile user. In order to resolve the problem, an adaptive learning method of context mobile user preferences is proposed, and the method can ensure the accuracy and reduce the response time. Fristly, through analyzing the context mobile user behaviors, the method judges whether mobile user preference is affected by context or not and detects whether the context mobile user preference change. And then a context quantization method based on the weight and the similarity matrix of context is proposed. If the context mobile user preference does not change, it only needs to modify the confidence of the user preference. If the context mobile preference changes, we employ the classifier to learn the context mobile user preference. Since only part of the context mobile user preferences needs to learn, it reduced the response time of the study. In order to ensure the accuracy of the context mobile user preference and further speed up the response time of the study, the context is introduced into the least square support vector machine classifier. And we employ the increment context least square support vector machine to learn the changed context mobile user preference.(3) The collaborative filtering is the most common method to predict the mobile user preference. However, the traditional collaborative filtering has the sparsity and cold start problem. In the mobile network, the introduction of the context further exacerbates the sparsity problem. On the basis of previous studies, an improved collaborative filtering method based on the timestamp is proposed to predict the mobile user preference to the unused mobile network service. Firstly, we select the context mobile user preferences that meet the requirements to calculate the similarity between the mobile users according to the follow time. And then we combine the trust with the similarity between the mobile users to select the nearest neighbors. Before predicting the mobile user preference, we select the mobile network services that will most likely be used by the target mobile user according to the follow time and the credibility of mobile user preference. And we predict the mobile user preferences to unused mobile network service according to the preferences of the nearest neighbors. Finally, in order to solve the cold start problem caused by the new mobile network services, we judge whether the mobile user is a fashion user according to the mean value of follow time to the new mobile network services. And the item-based collaborative filtering method is employed to predict the preference of the fashion user to the new network services. Since the proposed method makes the choice to the context mobile user preferences and the most likely used mobile network services when predicting the target mobile user preference, it can alleviate the sparsity of the data. The proposed method can guarantee the accuracy and reduce the response time of the prediction. Therefore, it can better to accord with the real-time requirements of the mobile user.(4) In the existing auquisition methods of the context mobile user preference, most of them obtain mobile user preference in a static environment. However, due to the real-time characteristics of the mobile network, it is necessary to obtain the mobile user preference accurately and timely. In order to solve the above problem, this paper presents a context mobile user preference online acquisition method based on the sliding window. Firstly, we set the sliding window and the basic window by the time-based method. And then the context mobile user preferences are divided into three categories according the frequency of the mobile network services used:the preference to unused mobile network services, and we use the improved collaborative filtering method to predict; the preference to the mobile network services were used previously, but recently have not been used, and we employ the forgotten function to learn; the preference to the mobile network services that are used frequently, and we use the online context least squares support vector machine to learn. Finally, the experimental results show that the online acqusition method is superior to the offline learning.