This talk does not have an abstract.

This talk does not have an abstract.

As the Internet of Things (IoT) get bigger and more complex, efficient and effective management solutions must be developed and put into operation. IoT management is even more crucial if, in addition to managing physical sensors, we want to have a unified IoT management that is able to cope with virtual (software) sensors and human-based sensors, which provide contextualised data via Online Social Networks (OSN). In this paper we present and explore approaches to this unified management, resorting to open and widely adopted standards for both data and device management, namely OMA-LwM2M and FIWARE. We present a proof-of-concept implementation that shows that the management of the referred three types of sensing is feasible from the functional and performance points of view.

The rapid growth of mobile traffic and the vast variety online activities demand enhancement of mobile networks. Software-Defined Networking (SDN) is considered a promising solution for the core network. But upgrading to SDN from a legacy network is a multi-period process. From both dynamic and global perspectives, during the migration it needs to simultaneously consider optimizing the local goal at each intermediate step and the global goal. This involves in some essential questions that have to be answered: which legacy devices to upgrade and when to, and how to place controllers. Due to interaction between the local goals and the global goal, however, to answer these questions together is challenging. In this paper we study the SDN migration problem and answer all of these questions together. We formulate the SDN migration problem as a time-varying dual-objective dynamic optimization model, in which the timing factor is taken into account, and the objective of optimization varies with time. Then we introduce a penalty item to convert the dual-objective dynamic optimization problem to a series of single step optimization problems that can be solved directly by CPLEX for small scale networks. The simulation results based on real network topology show that our model can obtain a tradeoff between the local goals and global one.

In recent years, recommendation based on explicit feedback data has been extensively studied. However, in the field of Point-of-Interest (POI) recommendation, check-in information is usually implicit feedback, that is, we can only observe positive data where users interact with POIs. The lack of negative samples brings difficulties to the research of POI recommendation. Although there have been recently studies converted the rating prediction into the POIs ranking by constructing pairwise preference assumption, they only consider the optimization of the ranking of one POI pair, which the value of negative data is underutilized. In addition, the geographical influence has not been fully utilized. Hence, we propose a recommendation model based on geographic influence and extended pairwise ranking (GIEPR). Extensive empirical studies on two publicly available datasets show that our method performs significantly better than state-of-the-art methods for POI recommendation.

With the rapid development of online social networks, marketing through online social platforms attracts a lot of attention. Recently, a special social marketing method is prevailing, i.e., e-commerce celebrities(ECs). ECs run their social network accounts to attract followers and then sell products to them directly. While the sales of ECs have dominated the e-commerce marketing in China, there is, however, a lack of accurate measurement and model about it. In this paper, we first conduct a large-scale cross-platform measurement on two of the biggest online social network platforms and e-commerce platforms in China, i.e., Sina Weibo and Taobao. We then characterize the typical behavioral patterns of ECs and build a machine learning model to quantitatively represent the relationship between the social network behavior and their product sale volumes. Experimental results show that we can accurately predict an EC's sale volume based on the 41 social network behavior features (F1 score can reach 0.83). Furthermore, we obtain the top-10 most important features that affect the sales. Our measurement and modeling results provide beneficial insights in understanding and optimizing social marketing for ECs.

In recent years, news providers, as well as politicians, made massive efforts to publish and disseminate their content on online social media networks to reach a wider audience. Considering the ubiquity of mobile devices, social media users started to consume daily news directly on the platforms, mainly in an incidental manner. This way of information consumption bears many risks. Social media users tend to overly trust information distributed by their friends in the network and therefore fail to evaluate the credibility and trustworthiness of information sources. The results are lower levels of political learning and the risk of partisan echo chambers, which are user groups that only amplify and reinforce their own beliefs. To measure and understand these phenomenons one has to analyze the user behavior on these platforms, requiring a virtually complete data set of Twitter usage of the targeted community. In this work, we focus on collection techniques to obtain a complete data set of sub-groups of Twitter users, i.e. specifically targeting German-tweeting Twitter users. Therefore, we collect a representative snapshot of the German Twitter traffic over two months, including 77 million tweets and $6.9$ million users. Finally, we report on exhaustive evaluations on the virtually complete data. Our examination also showed the massive impact of political events, such as the 2019 European Parliament election, on the German Twitter-sphere.

This talk does not have an abstract.