|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Spontaneous Wireless Networks (SWN)
Spatio-Temporal Prediction of Cellular Data Traffic
The understanding of human behaviors is a central question in multi-disciplinary research and has contributed to a wide range of applications. The ability to foresee human activities has essential implications in many aspects of cellular networks. In particular, the high availability of mobility prediction can enable various application scenarios such as location-based recommendation, home automation, and location-related data dissemination; the better understanding of future mobile data traffic demand can help to improve the design of solutions for network load balancing, aiming at improving the quality of Internet-based mobile services. Although a large and growing body of literature has investigated the topic of predicting human mobility, there has been little discussion in anticipating mobile data traffic in cellular networks, especially in spatiotemporal view of individuals. We address the problem of understanding spatiotemporal mobile data traffic demand for individuals and perform an theoretical and empirical analysis of jointly predicting human whereabouts and mobile data traffic, by collaboratively mining human mobility dataset and mobile data traffic dataset. Our contributions are summarized as follows:
-
We investigate the limits of predictability by measuring the maximum predictability that any algorithm has potential to achieve based on tools of information theory. Our theoretical analysis shows that it is theoretically possible to anticipate the individual demand with a typical accuracy of 75% despite the heterogeneity of users and with an improved accuracy of 80% using joint prediction with mobility information. This work was published at the IEEE LCN 2017 international conference and the Technical report RT-0483 brings a full description of the work, which is being prepared for a journal submission.
-
We evaluate the state-of-the-art predictors and propose novel solutions for predicting mobile data traffic via machine learning algorithms. Our data-driven test on the performance of these predictors show that the 2nd order Markov predictor outperforms all the legacy time series predictors. It can achieve a mean accuracy of 62% but can hardly have an enhancement from knowing human mobility information. Besides, based on machine learning techniques, our proposed solutions can achieve a typical accuracy of 70% and have a 1% 5% degree of improvement by learning individual whereabouts (what confirms the predictability theoretical results). Finally, our analysis show that knowing mobile data traffic of a user can significantly help the prediction of his whereabouts for 50% of the users, leading to an improvement up to 10% regarding accuracy. The Technical Report hal-01675573 brings more details on this work. A conference paper is also in preparation.
All those works were performed in the context of the Guangshuo Chen's PhD thesis, who will defend in March 2018.
Human Mobility completion of Sparse Call Detail Records for Mobility Analysis
Call Detail Records (CDR) are an important source of information in the study of diverse aspects of human mobility. The accuracy of mobility information granted by CDR strongly depends on the radio access infrastructure deployment and the frequency of interactions between mobile users and the network. As cellular network deployment is highly irregular and interaction frequencies are typically low, CDR are often characterized by spatial and temporal sparsity, which, in turn, can bias mobility analyses based on such data. In this paper, we precisely address this subject. First, we evaluate the spatial error in CDR, caused by approximating user positions with cell tower locations. Second, we assess the impact of the limited spatial and temporal granularity of CDR on the estimation of standard mobility metrics. Third, we propose novel and effective techniques to reduce temporal sparsity in CDR, by leveraging regularity in human movement patterns.
These works have been published as invited papers at the ACM CHANTS 2016 workshop (in conjunction with ACM MobiCom 2016) and at the IEEE DAWM workshop (in conjunction with IEEE Percom 2017). A journal version (also registered as TR: hal-01646608) is in revision at the Computer Communication Elsevier Journal, and got the first notification asking for minor revisions. Finally, a new completion methodology improving the previously described that leverages tensor factorization was designed and will be submitted to a journal: the technical report hal-01675570 describes this work.
Sampling frequency of human mobility
Recent studies have leveraged tracking techniques based on positioning technologies to discover new knowledge about human mobility. These investigations have revealed, among others, a high spatiotemporal regularity of individual movement patterns. Building on these findings, we aim at answering the question “at what frequency should one sample individual human movements so that they can be reconstructed from the collected samples with minimum loss of information?”. Our quest for a response leads to the discovery of (i) seemingly universal spectral properties of human mobility, and (ii) a linear scaling law of the localization error with respect to the sampling interval. Our findings are based on the analysis of fine-grained GPS trajectories of 119 users worldwide. The applications of our findings are related to a number of fields relevant to ubiquitous computing, such as energy-efficient mobile computing, location-based service operations, active probing of subscribers' positions in mobile networks and trajectory data compression. to an international conference in the next months. This work was published at the IEEE Globecom 2017 international conference.
We are improving the currently published sampling approache by incorporating human behavioral features at the sampling decisions to make it more adaptive. This is an on-going work with Panagiota Katsikouli, who spent 5 months in our team working as an internship and is currently doing a Post-Doc at the AGORA Inria team.
Inference of human personality from mobile phones datasets
Personality research has enjoyed a strong resurgence over the past decade. Trait-based personality theories define personality as the traits that predict a person's behavior through learning and habits. Personality traits are relatively stable over time, differ across individuals, and most importantly, influence behavior. In psychology, the human personality has been modeled into a set of independent factors that, together, accurately describe any individual: The Five Factors Personality Model. This personality model presents the Big Five personality traits, often represented by the OCEAN acronym: Openness: appreciation for a variety of experiences; Conscientiousness: planning ahead rather than being spontaneous; Extraversion: being sociable, energetic and talkative; Agreeableness: being kind, sympathetic and happy to help; Neuroticism: inclined to worry or be vulnerable or temperamental.
This is a very recently started work, where we are firstly analysing the relationship between smartphone usages (i.e., social interactions, content interest, mobility, and communication) and personality traits in the Big Five Model. Most of the studies on personality traits were performed by social scientists and in particular, by psychologists. Studies reveal that one of the most distal influences shaping personality lie in the environment where development occurs. Nevertheless, the identification of precise environmental sources impacting personality is still an open research. More recently, computer science researchers have tried to extract personality from datasets collected through smartphones. Although laying the ground work to understand human personality from smartphones usage, much still remain to be investigated. Thus, we are performing analysis to study the correlation between traits and technological features. We plan then to establish a methodology to infer traints from features and consequently, to investigate how different traits influence different features.
This is an on-going work with Adriano di Luzio, who spent 4 months in our team working as an internship, and Julinda Stefa, an invited research visitor at Infine.
Predicting new places to visit in human mobility decision
Most location prediction methods need a large user mobility history to accurately predict the next location of a user (markov chains, rnn). These methods are particularly good for predicting locations that are frequently visited by users, but not as good for predicting new places or how a user?s trajectories change in case of random events. We amend this by using contextual information to manage new places and random events and the movement patterns of users who exhibit similar behaviours. In this context, we plan to use the user?s profile and social ties to identify the most probable next category of locations (type of actions: entertainment, social, food etc.). Then, use subtrajectory similarity to predict the route taken to the identified area. This is an on-going work with the intern Maria Astefanoaei and her advisor, who spent 5 months in our team.
Data offloading decision via mobile crowdsensing
With the steady growth of smart-phones sales [1], the demand for services that generate mobile data traffic has grown tremendously. WiFi offloading has been considered as a promising solution to the recent boost up of mobile data consumption that is making excessive demands on cellular networks in metropolitan areas. The idea consists in shifting the traffic off of cellular networks to WiFi networks. Characterizing the capacity and availability of a chaotic deployed dense WiFi network is crucial to understand and decide where and when to o oad data. This is the first goal of this work, where the MACACO dataset was considered in the characterization. Our final goal is the design of a decision strategy allowing a mobile phone of a user to decide if offload or not her traffic, i.e., when, where (using what Access Point in her usual mobility) and how (if the traffic will be offloadied to one or more Access Points). This is an on-going work with the intern emanuel Lima and his advisor, who spent 4 months in our team.
Infering friends in the crowd in Device-to-Device communication
The next generation of mobile phone networks (5G) will have to deal with spectrum bottleneck and other major challenges to serve more users with high-demanding requirements. Among those are higher scalability and data rates, lower latencies and energy consumption plus reliable ubiquitous connectivity. Thus, there is a need for a better spectrum reuse and data offloading in cellular networks while meeting user expectations. According to literature, one of the 10 key enabling technologies for 5G is device-to-device (D2D) communications, an approach based on direct user involvement. Nowadays, mobile devices are attached to human daily life activities, and therefore communication architectures using context and human behavior information are promising for the future. User-centric communication arose as an alternative to increase capillarity and to offload data traffic in cellular networks through opportunistic connections among users. Although having the user as main concern, solutions in the user-centric communication/networking area still do not see the user as an individual, but as a network active element. Hence, these solutions tend to only consider user features that can be measured from the network point of view, ignoring the ones that are intrinsic from human activity (e.g., daily routines, personality traits, etc). In this work, we plan to investigate how human-aspects and behavior can be useful to leverage future device-to-device communication. This is a recently started PhD thesis subject, aiming the design of a methodology to select next-hops in a D2D communication that will be human-aware: i.e., that will consider not only available physical resoures at the mobile device of a wireless neighbor, her mobility features and restrictions but also any information allowing to infer how much sharing willing she is.