Link prediction in human mobility networks

Yang, Yang; Chawla, Nitesh V.; Basu, Prithwish; Prabhala, Bhaskar; Porta, Thomas La

doi:10.1145/2492517.2492656

Cited by 13 publications

(17 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This gives us the ability to view mobility traces in a hierarchical network model by examining the trace data in a different granularity. In this scheme the incidental places and their associated traces will be absolved into other (significant) place-to-place transitions using the technique of link predictions [18]. Thus we can utilize the users' mobility network with a smaller number of places by combining several transitions.…”

Section: Summary and Discussionmentioning

confidence: 99%

Spatial and temporal considerations in next place predictions

Prabhala

Porta

2015

2015 IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS)

Self Cite

View full text Add to dashboard Cite

We present synthesized findings from a systematic study of user mobility based on a well grounded data set through mining attributes of place-to-place transitions. Next place predictions are the atomic units in constructing end-to-end user mobility trajectories based on historical trace data. These trajectories in turn form models for opportunistic networks to be utilized for providing location and communication services. We start with a baseline of the user's current place, start time, and end time to predict the next place. We demonstrate the efficiency of an algorithm called PeriodicaS through aggregated average prediction accuracies across all users over a large set of diverse participants. PeriodicaS mines periodicity intelligently in users' mobility traces and further improves prediction accuracies with additional classification rules. We derive these classification rules by applying explicit semantic annotations (home, work place and public transportation points associated with places visited), and accompanying group information. We propose novel ways of transforming bits of information in the mobility traces, defined to be inherent semantic annotations, as features for mobility modeling in PeriodicaS. Inherent semantic annotations are computed with temporal variations from visited places such as end time only, and measuring duration time. We deduce more inherent semantic annotations from place rankings by frequency of visits. By progressively employing these two types of semantic annotations, explicitly stated in the data set and deduced from the mobility traces, we improve next place prediction accuracies up to 54% compared to baseline predictions.

show abstract

Section: Summary and Discussionmentioning

confidence: 99%

Spatial and temporal considerations in next place predictions

Prabhala

Porta

2015

2015 IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Previous research focusing on predicting link dissolution is much less common than for link formation prediction. Recent research includes [24], which studied unfollowing behavior on twitter, [25] which studied unfriending behavior on Facebook and [22,23] which studied link dissolution on Wikipedia. In all of these previous studies, it was shown that predicting link dissolution is harder than predicting link formation.…”

Section: Link Dissolution Predictionmentioning

confidence: 99%

Semi-supervised Graph Embedding Approach to Dynamic Link Prediction

Hisano

2018

Springer Proceedings in Complexity

View full text Add to dashboard Cite

We propose a simple discrete time semi-supervised graph embedding approach to link prediction in dynamic networks. The learned embedding reflects information from both the temporal and cross-sectional network structures, which is performed by defining the loss function as a weighted sum of the supervised loss from past dynamics and the unsupervised loss of predicting the neighborhood context in the current network. Our model is also capable of learning different embeddings for both formation and dissolution dynamics. These key aspects contributes to the predictive performance of our model and we provide experiments with three realworld dynamic networks showing that our method is comparable to state of the art methods in link formation prediction and outperforms state of the art baseline methods in link dissolution prediction.

show abstract

“…The essential challenge facing academic researchers is thus to find a way to anticipate future changes in human activities (human mobility) that will enable them to make reliable predictions of future urban energy demand. A growing body of research has explored the predictability of human mobility [31][32][33][34][35], and new models are continually being proposed that provide more reliable predictions of energy consumption [6,18,23,36,37]. Studies that have focused primarily on predicting human mobility [33,34,38,39] and building human mobility-based predictive models [13] have taken a number of different approaches.…”

Section: Related Workmentioning

confidence: 99%

Urban energy flux: Spatiotemporal fluctuations of building energy consumption and human mobility-driven prediction

Taylor

2017

Applied Energy

View full text Add to dashboard Cite

Urbanization is causing a significant increase in the amount, diversity, and complexity of human activities, all of which have a substantial impact on energy consumption. Current approaches to predicting energy demand at different spatiotemporal levels are functions of the characteristics of either individual buildings or cities and their occupancy levels, or are data-driven, typically taking the form of sensor-based modeling. Nevertheless, accounting for both the spatial and temporal effects of heterogeneous human behavior patterns on buildings' energy use at the city level remains a challenge. In this paper, we examine the temporal manifestation of the fluctuations of energy use in urban buildings driven by spatial mobility patterns of the population. We then present an urban-level spatiotemporal approach for predicting buildings' energy demand. Using a full year of individual positional records from an online social networking platform (Twitter), we introduce a multivariate autoregressive model in reduced principle component analysis space to create monthly predictions of residential building electricity demand generated across 801 spatial divisions that account for 68% of the electricity used by buildings in the City of Chicago, through a spatial autoregressive model. This model represents an important step forward, incorporating the spatiotemporal energy use fluctuations of urban population activities to create more reliable predictions of demand in future cities.

show abstract

Link prediction in human mobility networks

Cited by 13 publications

References 22 publications

Spatial and temporal considerations in next place predictions

Spatial and temporal considerations in next place predictions

Semi-supervised Graph Embedding Approach to Dynamic Link Prediction

Urban energy flux: Spatiotemporal fluctuations of building energy consumption and human mobility-driven prediction

Contact Info

Product

Resources

About