Predicting the popularity of social media content in real time requires approaches that efficiently operate at global scale. Popularity prediction is important for many applications, including detection of harmful viral content to enable timely content moderation. The prediction task is difficult because views result from interactions between user interests, content features, resharing, feed ranking, and network structure. We consider the problem of accurately predicting popularity both at any given prediction time since a content item's creation and for arbitrary time horizons into the future. In order to achieve high accuracy for different prediction time horizons, it is essential for models to use static features (of content and user) as well as observed popularity growth up to prediction time. We propose a feature-based approach based on a self-excited Hawkes point process model, which involves prediction of the content's popularity at one or more reference horizons in tandem with a point predictor of an effective growth parameter that reflects the timescale of popularity growth. This results in a highly scalable method for popularity prediction over arbitrary prediction time horizons that also achieves a high degree of accuracy, compared to several leading baselines, on a dataset of public page content on Facebook over a two-month period, covering billions of content views and hundreds of thousands of distinct content items. The model has shown competitive prediction accuracy against a strong baseline that consists of separately trained models for specific prediction time horizons.

Video accounts for a large proportion of traffic on the Internet. Understanding its geographical viewing patterns is extremely valuable for the design of Internet ecosystems for content delivery, recommendation and ads. While previous works have addressed this problem at coarse-grain scales (e.g., national), the urban-scale geographical patterns of video access have never been revealed. To this end, this article aims to investigate the problem that whether there exists distinct viewing patterns among the neighborhoods of a large-scale city. To achieve this, we need to address several challenges including unknown of patterns profiles, complicate urban neighborhoods, and comprehensive viewing features. The contributions of this article include two aspects. First, we design a framework to automatically identify geographical video viewing patterns in urban neighborhoods. Second, by using a dataset of two months real video requests in Shanghai collected from one major ISP of China, we make a rigorous analysis of video viewing patterns in Shanghai. Our study reveals the following important observations. First, there exists four prevalent and distinct patterns of video access behavior in urban neighborhoods, which are corresponding to four different geographical contexts: downtown residential, office, suburb residential and hybrid regions. Second, there exists significant features that distinguish different patterns, e.g., the probabilities of viewing TV plays at midnight, and viewing cartoons at weekends can distinguish the two viewing patterns corresponding to downtown and suburb regions.

The exploding volumes of mobile video traffic call for deploying content caches inside mobile operator networks. With in-network caching, users' requests for popular content can be served from a content cache deployed at mobile gateways in vicinity to the end user. This inherently reduces the load on the content servers and the backbone of operator's network. In light of the increasing trend in virtualization of network functions, we propose a cost-effective caching as a service (CaaS) framework for virtual video caching in 5G mobile networks. In order to evaluate the pros and cons of our CaaS approach, we formulate two virtual caching problems, namely maximum return on investment(MRI) and maximum offloaded traffic (MOT). MRI aims at maximizing return on caching investment by finding the best trade-off between the cost of cache storage and bandwidth savings from caching video contents in the mobile network operator (MNO)'s cloud. Likewise, MOT aims to maximize the traffic offloaded from the MNO's core and backhaul within given budget constraints. More specifically, taking the popularity and size of video contents into account, MRI and MOT aim to find the optimal caching tables which maximize the ratio of transmission bandwidth cost to storage cost and the offloaded traffic for a given budget, respectively. We reduce the complexity of the proposed problem formulated as a binary-integer programming (BIP) by using canonical duality theory (CDT). Experimental results obtained using the invasive weed optimization (IWO) have shown significant performance enhancement of the proposed system in terms of return on investment, quality, offloaded traffic, and storage efficiency.

P2P sharing amongst consumers has been proposed as a way to decrease load on Content Delivery Networks. This paper develops an analytical model that shows an additional benefit of sharing content locally: Selecting close by peers to share content from leads to shorter paths compared to traditional CDNs, decreasing the overall carbon footprint of the system. Using data from a month-long trace of over 3 million monthly users in London accessing TV shows online, we show that local sharing can result in a decrease of 24-48% in the system-wide carbon footprint of online video streaming, despite various obstacle factors that can restrict swarm sizes. We confirm the robustness of the savings by using realistic energy parameters drawn from two widely used settings. We also show that if the energy savings of the CDN servers are transferred as carbon credits to the end users, over 70% of users can become carbon positive, i.e., are able to support their content consumption without incurring any carbon footprint, and are able to offset their other carbon consumption. We suggest carbon credit transfers from CDNs to end users as a novel way to incentivise participation in peer-assisted content delivery.

Peer-assisted content delivery networks have recently emerged as an economically viable alternative to traditional content delivery approaches: the feasibility studies conducted for several large content providers suggested a remarkable potential of peer-assisted content delivery networks to reduce the burden of user requests on content delivery servers and several commercial peer-assisted deployments have been recently introduced. Yet there are many technical and commercial challenges which question the future of peer-assisted solutions in industrial settings. This includes among others unreliability of peer-to-peer networks, the lack of incentives for peers’ participation, and copyright issues. In this paper, we carefully review and systematize this ongoing debate around the future of peer-assisted networks and propose a novel taxonomy to characterize the research and industrial efforts in the area. To this end, we conduct a comprehensive survey of the last decade in the peer-assisted content delivery research and devise a novel taxonomy to characterize the identified challenges and the respective proposed solutions in the literature. Our survey includes a thorough review of the three very large scale feasibility studies conducted for BBC iPlayer, MSN Video and Conviva, five large commercial peer-assisted CDNs - Kankan, LiveSky, Akamai NetSession, Spotify, Tudou - and a vast scope of technical papers. We focus both on technical challenges in deploying peer-assisted solutions and also on non-technical challenges caused due to heterogeneity in user access patterns and distribution of resources among users as well as commercial feasibility related challenges attributed to the necessity of accounting for the interests and incentives of Internet Service Providers, End-Users and Content Providers. The results of our study suggest that many of technical challenges for implementing peer-assisted content delivery networks on an industrial scale have been already addressed in the literature, whereas a problem of finding economically viable solutions to incentivize participation in peer-assisted schemes remains an open issue to a large extent. Furthermore, the emerging Internet of Things (IoT) is expected to enable expansion of conventional CDNs to a broader network of connected devices through machine to machine communication.

The exploding volumes of mobile video traffic call for deploying content caches inside mobile operators network. With in-network caching, users' requests for popular content can be served from a content cache deployed at mobile gateways in vicinity to the end user, therefore considerably reducing the load on the content servers and the backbone of operator's network. In practice, content caches can be installed at multiple levels inside an operator's network (e.g., serving gateway, packet data network gateway, RAN, etc.), leading to an idea of hierarchical in-network video caching. In order to evaluate the pros and cons of hierarchical caching, in this paper we formulate a cache provisioning problem which aims to find the best tradeoff between the cost of cache storage and bandwidth savings from hierarchical caching. More specifically, we aim to find the optimal size of video caches at different layers of a hierarchical in-network caching architecture which minimizes the ratio of transmission bandwidth cost to storage cost. We overcome the complexity of our problem which is formulated as a binary-integer programming (BIP) by using canonical duality theory (CDT). Numerical results obtained using the invasive weed optimization (IWO) show that important gains can be achieved, with benefit-cost ratio and cost efficiency improvements of more than 43% and 38%, respectively.

Mobile data offloading can greatly decrease the load on and usage of current and future cellular data networks by exploiting opportunistic and frequent access to Wi-Fi connectivity. Unfortunately, Wi-Fi access from mobile devices can be difficult during typical work commutes, e.g., via trains or cars on highways. In this paper, we propose a new approach: to preload the mobile device with content that a user might be interested in, thereby avoiding the need for cellular data access. We demonstrate the feasibility of this approach by developing a supervised machine learning model that learns from user preferences for different types of content, and propensity to be guided by the user interface of the player, and predictively preload entire TV shows. Testing on a data set of nearly 3.9 million sessions from all over the U.K. to BBC TV shows, we find that predictive preloading can save over 71% of the mobile data for an average user.

The last 5 years have seen a dramatic shift in media distribution. For decades, TV and radio were solely provisioned using push-based broadcast technologies, forcing people to adhere to fixed schedules. The introduction of catch-up services, however, has now augmented such delivery with online pull-based alternatives. Typically, these allow users to fetch content for a limited period after initial broadcast, allowing users flexibility in accessing content. Whereas previous work has investigated both of these technologies, this paper explores and contrasts them, focusing on the network consequences of moving towards this multifaceted delivery model. Using traces from nearly 6 million users of BBC iPlayer, one of the largest catch-up TV services, we study this shift from push- to pull-based access. We propose a novel technique for unifying both push- and pull-based delivery: the Speculative Content Offloading and Recording Engine (SCORE). SCORE operates as a set-top box, which interacts with both broadcast push and online pull services. Whenever users wish to access media, it automatically switches between these distribution mechanisms in an attempt to optimize energy efficiency and network resource utilization. SCORE also can predict user viewing patterns, automatically recording certain shows from the broadcast interface. Evaluations using our BBC iPlayer traces show that, based on parameter settings, an oracle with complete knowledge of user consumption can save nearly 77% of the energy, and over 90% of the peak bandwidth, of pure IP streaming. Optimizing for energy consumption, SCORE can recover nearly half of both traffic and energy savings.

In search of scalable solutions, CDNs are exploring P2P support. However, the benefits of peer assistance can be limited by various obstacle factors such as ISP friendliness - requiring peers to be within the same ISP, bitrate stratification - the need to match peers with others needing similar bitrate, and partial participation - some peers choosing not to redistribute content. This work relates potential gains from peer assistance to the average number of users in a swarm, its capacity, and empirically studies the effects of these obstacle factors at scale, using a month-long trace of over 2 million users in London accessing BBC shows online. Results indicate that even when P2P swarms are localised within ISPs, up to 88% of traffic can be saved. Surprisingly, bitrate stratification results in 2 large sub-swarms and does not significantly affect savings. However, partial participation, and the need for a minimum swarm size do affect gains. We investigate improvements to gain from increasing content availability through two well-studied techniques: content bundling-combining multiple items to increase availability, and historical caching of previously watched items. Bundling proves ineffective as increased server traffic from larger bundles outweighs benefits of availability, but simple caching can considerably boost traffic gains from peer assistance.

Using nine months of access logs comprising 1.9 Billion sessions to BBC iPlayer, we survey the UK ISP ecosystem to understand the factors affecting adoption and usage of a high bandwidth TV streaming application across different providers. We find evidence that connection speeds are important and that external events can have a huge impact for live TV usage. Then, through a temporal analysis of the access logs, we demonstrate that data usage caps imposed by mobile ISPs significantly affect usage patterns, and look for solutions. We show that product bundle discounts with a related fixed-line ISP, a strategy already employed by some mobile providers, can better support user needs and capture a bigger share of accesses. We observe that users regularly split their sessions between mobile and fixed-line connections, suggesting a straightforward strategy for offloading by speculatively pre-fetching content from a fixed-line ISP before access on mobile devices.

Everyday, millions of users save content items for future use on sites like Pinterest, by ''pinning'' them onto carefully categorised personal pinboards, thereby creating personal taxonomies of the Web. This paper seeks to understand Pinterest as a distributed human computation that categorises images from around the Web. We show that despite being categorised onto personal pinboards by individual actions, there is a generally a global agreement in implicitly assigning images into a coarse-grained global taxonomy of 32 categories, and furthermore, users tend to specialise in a handful of categories. By exploiting these characteristics, and augmenting with image-related features drawn from a state-of-the-art deep convolutional neural network, we develop a cascade of predictors that together automate a large fraction of Pinterest actions. Our end-to-end model is able to both predict whether a user will repin an image onto her own pinboard, and also which pinboard she might choose, with an accuracy of 0.69 ([email protected] of 0.75).

Modeling human mobility is crucial in the analysis and simulation of opportunistic networks, where contacts are exploited as opportunities for peer-to-peer message forwarding. The current approach to human mobility modeling has been based on continuously modifying models, trying to embed in them the mobility properties (e.g., visiting patterns to locations or specific distributions of inter-contact times) as they arose from trace analysis. As a consequence, with these models it is difficult, if not impossible, to modify the features of mobility or to control the exact shape of mobility metrics (e.g., modifying the distribution of inter-contact times). For these reasons, in this paper we propose a mobility framework rather than a mobility model, with the explicit goal of providing a flexible and controllable tool for modeling mathematically and generating simulatively different possible features of human mobility. Our framework, named SPoT, is able to incorporate the three dimensions–spatial, social, and temporal–of human mobility. The way SPoT does this is by mapping the different social communities of the network into different locations, whose members visit with a configurable temporal pattern. In order to characterize the temporal patterns of user visits to locations and the relative positioning of locations based on their shared users, we analyze the traces of real user movements extracted from three location-based online social networks (Gowalla, Foursquare, and Altergeo). We observe that a Bernoulli process effectively approximates user visits to locations in the majority of cases, and that locations that share many common users visiting them frequently tend to be located close to each other. In addition, we use these traces to test the flexibility of the framework, and we show that SPoT is able to accurately reproduce the mobility behavior observed in traces. Finally, relying on the Bernoulli assumption for arrival processes, we provide a thorough mathematical analysis of the controllability of the framework, deriving the conditions under which heavy-tailed and exponentially-tailed aggregate inter-contact times (often observed in real traces) emerge.

The problem of identifying the optimal location for a new retail store has been the focus of past research, especially in the field of land economy, due to its importance in the success of a business. Traditional approaches to the problem have factored in demographics, revenue and aggregated human flow statistics from nearby or remote areas. However, the acquisition of relevant data is usually expensive. With the growth of location-based social networks, fine grained data describing user mobility and popularity of places has recently become attainable.

Modeling human mobility is crucial in the performance analysis and simulation of mobile ad hoc networks, where contacts are exploited as opportunities for peer-to-peer message forwarding. The current approach to human mobility modeling has been based on continuously modifying models, trying to embed in them the newest features of mobility properties (e.g., visiting patterns to locations or inter-contact times) as they came up from trace analysis. As a consequence, typically these models are neither flexible (i.e., features of mobility cannot be changed without changing the model) nor controllable (i.e., the exact shape of mobility properties cannot be controlled directly). In order to take into account the above requirements, in this paper we propose a mobility framework whose goal is, starting from the stochastic process describing the arrival patterns of users to locations, to generate pairwise inter-contact times and aggregate inter-contact times featuring a predictable probability distribution. We validate the proposed framework by means of simulations. In addition, assuming that the arrival process of users to locations can be described by a Bernoulli process, we mathematically derive a closed form for the pairwise and aggregate inter-contact times, proving the controllability of the proposed approach in this case.