Search

Social media data have been used in many studies in the recent years. Compared to the traditional household travel survey data, social media data have a lower cost, and they can be obtained from abundant sources. However, several pre-processing tasks are required before social media data can be used for mobility analyse. For instance, distinguishment needs to be made between location and activity. In this thesis, text analytics, machine learning, and “Tweet Block” are applied in order to differentiate between location and activity and to extract more accurate information from social media data which can then be used for mobility analyse. In Section 3, the focus is to extract and analyse users’ movement and lifestyle. Unlike the state-of-the-practice, this research clearly distinguishes between location and activity. Text mining technique was applied to identify location and activity information respectively, and a clustering algorithm was applied to analyse the lifestyle of users. The strict distinguish between activity and location led to a result that the identified data is limited compared to traditional ways of labelling. To solve this problem, the information extracted from data was enriched by applying a method called “Tweet Block”. Tweet Block enable to identify 1,745 location and 98 activity which were not identified in text mining process. With the enriched data in hand, a method was purposed to infer information of user’s movement from the data point that is previously unusable (i.e. a single record from a day.) The average generated trip rate using this method was increased by 26%-50% compared to the method used in previous research. Travelling track was also generated to analyse the movement of these users. In Section 4, the primary purpose is to build a valid activity prediction model from the data. Machine learning algorithms were applied to build an activity prediction model from the data. Land use data were overlapped to the original data set, which acted as a supportive data to location information. Random Forest (RF) and Neural Network (NN) algorithms were used to build models and NN models were kept after model selection. A Stratified K-fold cross-validation was used to validate the model.

Chinese migrant workers constitute the backbone of the export economy, accounting for approximately 56% of the urban workforce. Yet, these workers are widely considered among the most exploited and vulnerable groups. A key issue is: Are these workers best regarded as part of an underclass or a new section of the working class. An examination of migrant worker job mobility helps answer this question: no evidence of upward job mobility would support the underclass thesis while significant job mobility would favour the working class thesis. Construction of an 11-point social class (occupational status) scale and survey data collected from over 4,000 migrants in the strong exporting Pearl River Delta region enabled answers to several questions. First, compared to local workers, a higher proportion of migrants (started their urban career at the lowest status level. However, arising from job change, a greater minority of migrant workers moved upward, even though most of the improvement was short range. Comparatively, respondentsâ income change exceeded their status change, with more than half of better off workers having increased their annual income by more than 30%. Using regression analysis, this study reveals the key factors influencing upward job mobility and changes in income, including age, hukou, educational level and employer type and size. Adopting the technique of latent class analysis, this study maps out the typical job pathways traversed by migrant workers (one-time movers and two â time movers) after arrival in the city. Overall, the most typical mobility path of one-time movers represent a steady and progressive line while that of two-time movers shows more downward-inclined mobility pattern and fluctuations. In conclusion, this study shows that although migrant workers began from a lower base compared to local urban workers, job mobility enabled status and income improvement, in effect, facilitating recruitment into the working class rather than being consigned to permanent impoverishment at the bottom of the class structure as suggested by the underclass thesis.

A mobile ad hoc network is a collection of wireless nodes, all of which may be mobile, that dynamically create a wireless network amongst them without using any infrastructure. Ad hoc wireless networks come into being solely by peer-to-peer interactions among their constituent mobile nodes, and it is only such interactions that are used to provide the necessary control and administrative functions supporting such networks. Mobile hosts are no longer just end systems; each node must be able to function as a router as well to relay packets generated by other nodes. As the nodes move in and out of range with respect to other nodes, including those that are operating as routers, the resulting topology changes must somehow be communicated to all other nodes as appropriate. In accommodating the communication needs of the user applications, the limited bandwidth of wireless channels and their generally hostile transmission characteristics impose additional constraints on how much administrative and control information may be exchanged, and how often. Ensuring effective routing is one of the greatest challenges for ad hoc networking. As a practice, ad hoc routing protocols make routing decisions based on individual node mobility even for applications such as disaster recovery, battlefield combat, conference room interactions, and collaborative computing etc. that are shown to follow a pattern. In this thesis we propose an algorithm that performs routing based on underlying mobility patterns. A mobility pattern aware routing algorithm is shown to have several distinct advantages such as: a more precise view of the entire network topology as the nodes move; a more precise view of the location of the individual nodes; ability to predict with reasonably accuracy the future locations of nodes; ability to switch over to an alternate route before a link is disrupted due to node movements.