Research On User Relationship And Mobilbility Behavior In Multi-source And Cross-domain Mobile Internet

With the rapid development of Mobile Internet, the study of perception, memory, analysis and application of multi-source heterogeneous data is a challenging research subject in the big data era. On the one hand, a rapid increase of amount of data leads to a gradually drop in performance of tradition algorithm for data mining and a variety of data category shields tradition methods for data fusion and data storage from having to deal with heterogeneous data. On the other hand, low data profit-density results in the loss of strengths of tradition methods for data analysis and it presents a challenge for real-time data processing and analysis in the respect of fast data updating and high effectiveness. Moreover, the I/O of these multi-source data's real-time demand is rigorous. A series of tradition methods of perception, memory, analysis for data obviously couldn't meet user's demand for processing real-time data in the four aspects:volume, multi-source heterogeneous, high-value and high effectiveness. If the distributed processing technique of cloud-computing is adopted for mass data and context is built for multi-sourece heterogeneous data, and appropriate rules are set for the I/O of high-value information and efficient algorithms are coupled with equipments. It may relieve stress of users'demands for data in big data era and better meet information needs of users.In this paper, the exsiting problems are based on perception, fusion, memory, analysis and application for data. The research is done with the goal of user behavior and user mobility on mobile internet, and discussed on user relationship and mobility behavior in urban.The main contributions of this paper are summarized as follows:1) For the problem of user's following behavior in social network, we propose a methodology for user credibility. The main idea of this algorithm is that firstly the five key factors for evaluation of user credibility are extracted which are number of followers, friends, tweets, favorites and bi-followers. These factors are adopted for computing the scores of user credibility. The scoring procedures include self-evaluation and mutual evaluation so that the evaluation model is consist of two sub-models which are self-evaluation model and mutual evaluation model.Through computational process of the model, user credibility of every user is drawn and then the score is sorted from top to bottom. And after that, user credibility is derived;2) For the problem of user mobility in the research of urban computing, we propose W5 model in location-based social network (LBSN). The model could well describe mobility behavior of users in our daily life. Dynamic context is built that it could explain well for some people (Who), some time (When), some place (Where), some thing (What) and some reason (Why). Compared with exsisted W4 model, the advantage of W5 model is dynamic context which not only could explain user behavior in the current context, but also could explain user behavior in the last context and predict user behavior in the next context. The main idea of the methodology is firstly the five key context factors of users'mobility behavior are extracted, i.e.5W. Thereinto, Who-. When and Where correspond to user id,check-in time and GPS in the dataset, respectively. Because they could be extracted directly from data items, so they are called explicit factors in the paper. What and Why should be extracted from micro-blog text using keyword extraction through the technology of text extraction because they are related to tweets and subjects which are tweeted by users. So the two factors are extracted indirectly and called implicit factors. After the five context factors extracted, calculations of joint probability for the factors are presented. The joint probability model could be deduced successively to solve the problem of context explanations and context prediction;3) For the problem of POIs recommendation in the research of urban computing, we propose a framework called PMR, which is in the research of mobile location-based social network (MLBSN). A set of entire computing principle is present by the framework which is about context-awareness, context memory, context recommendation and user feedback. The main idea of the methodology is that an existing framework called PMJ which have advantange on artificial intellegence could be used for reference and some computing methods about context-awareness could be combined. After that, a recommendation system framework used for context awareness, context memory and context recommendation is built. Then computing methods and running process of every module is respectly present. In the end, performance evaluation of recommendations is shown by the unit of user feedback. So these procedures make the framework be self-correcting to satisfy users'actual demands;4) For the problem of user mobility in the research of urban computing, we propose a mutual prediction model based on user roles and urban functional zones which is departed into two sub-models.And they are MUR?RC and MRC?UR MUR?RC is used to infer functional zones based on context factors and user roles; MRC?UR is used to infer user roles based on context factors and functional zones. The main idea of the model is as follows:through user roles, their routine activity areas and exsiting functional zones of a city, it could infer a user is being in which urban functional zones at some time quantum. For the same reason, it also could infer user roles according to urban functional zones of and a user's frequent activity in some area of a city at some specific times. This prediction model could discover users' latent activity traiectories among urban functional zones which is changed over time. In addition, this model could replenish lost data items on the basis of exsiting data itmes.In the research of above four aspects, they are all based on user behaviors. Thereinto, the research of user credibility belongs to the research field of user relationship behavior.The research of W5 model and POIs recommendation belongs to the research area of user mobility behavior. And the research of mutual inference model for user roles and urban functional zones belongs to the comprehensive research area of user relationship behavior and user mobility behavior.The experimental results of the four aspects show that user behavior in multi-source and cross-domain mobile internet could be exact description and prediction through superior mathematical modeling. And then these research achievements could be transformed into actual products for improving people's urban lives.