Revealing User Mobility Patterns Based On Matrix Factorization And Neural Network

With the continuous development of global technology, mobile phone users, especially the number of smart phone users increased year after year. In 2016, the amount of mobile phone sers has exceeded 2 billion, accounting for one-third of the total number of the world. As the statistics shows that, as of September 2016, the total number of mobile phone users in China is close to 1.316 billion. Among them, the mobile broadband users, that is, 3G and 4G users surged 179 million, and the total is close to 885 million. Data from Ministry of Industry and Information Technology shows that, in 2015, China's mobile data traffic consumption is up to 4.187 billion GB, the mobile phone outgoing calls is about 235.22 billion minutes, and the nationwide mobile text messages volume is near 54.08 billion. The mobile user data is more related to the location of the user, so the research on the user location data has been developing rapidly,and many valuable achievements have been made.Based on the goal of conducting research using the user location data,we choose the user location data provided by the mobile operator of a city in our country for one month, and carry on the comprehensive research on the data.Aiming at solving the problem that if only single source data is used in the learning process, the result obtained is not precise enough. And in this way a model which can fuse multi-source data and discover the regional function is proposed. In the process of urban evolution,areas with different functions appear. At the same time according to life experience,each region usually contains a variety of functions, and these functions have different strength. In our research, the topic model method of natural language processing is introduced into the analysis of regional function discovery. And a matrix decomposition model which can fuse multi-source data for function discovery is proposed. This method has a significant improvement in the area partitioning effect.Finally, as the user's trajectory can be seen as a time series data, and by taking advantage of the results from the previous study users tracks which are defined in accordance with functional labels can be obtained,therefore, the recurrent neural network in the neural network is used to predict the above two trajectories respectively. Compared with the traditional Markov prediction model, the accuracy of trajectory prediction using recurrent neural network was increased by 88% and 122%,respectively.The experimental results show that the results of this research will provide valuable data for regional function monitoring, traffic flow warning, and location recommendation in smart city.