Research On Context-aware Mobile User Personalized Requirements Dynamic Acquisition Techniques

With the development of mobile communication technology, more and more users can use mobile terminals to meet their needs anytime and anywhere(e.g., mobile web services, mobile applications or check-ins). As the premise of mobile personalized service, the research on context-aware mobile user person-alized requirements dynamic acquisition techniques has attracted wide attention from industry and academia. Compared with the traditional user requirements,there are many characteristics of the mobile user requirements, such as context awareness, dynamic, personalization, diversity. A lot of researchers have stud-ied requirement engineering and traditional user requirements acquisition tech-niques. However, their works mainly focused on domain system users and In-ternet users respectively, and they seldom considered the contexts. It is difficult to apply these methods directly to the context-aware mobile user personalized requirements dynamic acquisition techniques. Therefore, this thesis focuses on how to acquire mobile user requirements timely and precisely according to the context awareness, dynamic, personalization and diversity of mobile user requirements. The main contributions of this thesis are as follows.1) This thesis proposes a method of context-aware mobile user group re-quirements acquisition. There are many implicit and overlapping user groups in the mobile communication networks. Identifying these groups effectively and obtaining their needs are the key steps of providing mobile personalized services for them. However, traditional user group requirements acquisition techniques focused on the existing user groups and the sizes of them are small.Moreover, these methods only considered the disjoint groups. In addition, most context-aware mobile user personalized requirements dynamic acquisition tech-niques aimed at the single user, few of them considered the mobile user group requirements acquisition, especially for the overlapping mobile user group re-quirements acquisition. To address these problems, this thesis identifies the overlapping groups of mobile users and predict preferences for these groups according to the mobile user communication data, the proximity time of mobile users and mobile application usage logs. More specifically, this method can deal with the overlapping user groups and disjoint user groups. Finally, this thesis conducts a series of experiments on the MIT reality mining dataset and the synthetic dataset. The experimental results demonstrate the good perfor-mance of this method in accuracy.2) This thesis proposes a method of mining user requirements dynamically based on multi-dimensional cloud model. Currently, most mobile user require-ments dynamic acquisition techniques based on Collaborative Filtering (CF)method on location-based social networks (LBSNs) measured the similarity of users by common rated locations, and they neglected the influence of time con-text. With the purpose of modeling the features of mobile user preferences un-der different time context, this thesis firstly extends the two-dimensional cloud model to the multi-dimensional cloud model. At the same time, it introduces the multi-dimensional cloud model into evaluating the similarity of mobile user behaviors and prove the correctness of this method. In particular, it assesses the similarity of social ties based on common friends and common location el-ements. Secondly, in order to improve the accuracy of context-aware mobile user requirements dynamic acquisition techniques, it integrates the similarity of mobile user preferences, mobile user behaviors and mobile user friends into CF method. Finally, it is parallelized with Mapreduce framework for the im-provement of efficiency. The experimental results on Yelp Academic dataset demonstrate the performance gains of this method in accuracy and efficiency.3) This thesis presents a method of mobile user personalized requirements acquisition based on maximum entropy model. Existing methods of context-aware mobile user personalized requirements acquisition on LBSNs assumed that mobile user requirements follow to a certain distribution or model, e.g.,Power-law distribution (PD) or Multi-center Gaussian Model(MGM). Then they employed the unified model to extract mobile user preferences of new locations.However, these methods ignored that mobile users with different backgrounds have different requirements, that is, mobile user requirements are personalized and the influences of contexts on mobile user requirements are unique. Consid?ering the non-parametric characteristic of maximum entropy model, it not only utilizes it to model the influence of geographical context, but also introduces mobile user social links and mobile user preferences into maximum entropy model for the improvement of accuracy. In particular, it assesses the similarity of social ties based on common friends and common location categories. Fi-nally, in order to improve the efficiency of the method, it is parallelized with Mapreduce framework. The experimental results on Yelp Academic dataset demonstrate the performance gains in accuracy and efficiency of this method.4) This thesis presents an adaptive scheme of context-aware mobile user diverse requirements acquisition. Many works of context-aware mobile user diverse requirements acquisition on LBSNs introduced location category into assessing the similarity of mobile users. However, they neglected the dynamic and diversity of mobile user preferences. It firstly divides mobile user prefer-ences into mobile static preferences and mobile dynamic preferences according to location category, and utilizes the TF-IDF(Term Frequency-inverse Docu-ment Frequency) and two-dimensional cloud model to estimate them respec-tively. Secondly, it integrates the similarity of mobile user static preferences,mobile dynamic preferences and social ties into user-based CF method. At the same time, it develops a method of measuring location similarity, which con-siders the location category tree, geographical distance and the common users.Finally, it predicts mobile user preferences of new locations according to user-based CF method, location-based CF method, the category of new location and the categories of the mobile user visited. In particular, in order to improve the efficiency of the method, it is parallelized with Mapreduce framework. The experimental results on Foursquare dataset demonstrate the performance gains in accuracy and efficiency of this method.