Efficient techniques for time-constrained information dissemination using location-based social networks

Abstract

Social networks have undergone an explosive growth in recent years. They constitute a central part of users׳ everyday lives as they are used as major tools for the spread of information, ideas and notifications among the members of the network. In this work we investigate the use of location-based social networks as a medium of emergency notification, for efficient dissemination of emergency information among members of the social network under time constraints. Our objective is the following: given a location-based social network comprising a number of mobile users, the social relationships among the users, the set of recipients, and the corresponding timeliness requirements, our goal is to select an appropriate subset of users so that the spread of information is maximized, time constraints are satisfied and costs are considered. We propose LATITuDE, our system that investigates the interactions among the members of the social network to infer their social relationships, and develop scalable dissemination mechanisms that select the most efficient set of users to initiate the dissemination process in order to maximize the information reach among the appropriate receivers within a time window. Our detailed experimental results illustrate that our approach is practical, effectively addresses the problem of informing the appropriate set of users within a deadline when an emergency event occurs, uses a small number of messages, and consistently outperforms its competitors.

Introduction

Recently, we have observed the explosive growth of social networks such as Facebook [1], Twitter [2] and Google+ [3], that enumerate large amount of subscribers. For instance, Google+ has reached over 1.6 billion active users, while Facebook follows with over 1.2 billion users and Twitter with more than 600 million users.¹ These networks have been utilized as major tools for the spread of ideas, information and notifications among their members. Studies reveal that social networks can be exploited not only for “viral marketing” [4] (i.e., promote products to targeted sets of users that further propagate them through the word-of-mouth effect to reach a larger audience), but also for discovering emergent topics [5] and for emergency events alerting, management and public safety [6]. For example, people located in the vicinity of earthquakes share via Twitter, a well known social service for exchanging short text messages, anecdotal information related to the dissemination of seismically activity, that earthquake alerts lag behind firsthand notification [7], [8]. Studies reveal that depending on the size and location of the earthquake, scientific alerts can take between 2 and 20 min to publish, while using Twitter׳s notification capabilities people were notified about the occurrence of the earthquakes shaking within seconds of their occurrence. Recently, Facebook, one of the most popular social networks, announced the release of a tool, Safety Check, to be used by users in the proximity of a disaster zone to notify their loved ones about their safety during emergency situations. The application has been developed based on observations of user activity in the network in Japan during the tsunami of 2011.²

The use of social networks during emergencies and their efficiency in communicating important information, even when traditional communication medium fail, has been verified in numerous disastrous events in the recent years. As reported by the administrator of the US Federal Emergency Management Agency [9] with respect to the catastrophic 2010 Haiti earthquake, even when an area׳s physical infrastructure was completely destroyed, the cellular tower bounced back quickly, allowing survivors to request help from local first responders and emergency managers to relay important disaster-related information via social media sites.

Thus, social networks (i.e., Google+, Facebook, and Twitter) can play a major role in effective emergency notification due to their ability to (1) effectively reach millions of users, especially family and friends, (2) become alternative communication mediums when the wireless and telecommunication networks are overloaded during emergencies, and (3) provide cost-effective solutions since they are able to reach large amounts of social users without additional infrastructure costs. Furthermore, the study of social relationships and interactions in social networks may provide important insights for gathering information and planning evacuations during rescue efforts, as demonstrated in [10]. However, adopting location-based social networks as an effective communication medium for emergency alerting raises considerable challenges. Challenges lie in the level of availability and responsiveness expected from these infrastructures in delivering notifications under time constraints to reach all recipients interested in receiving the information (these can be people located in the area of the event i.e., students in a campus, as well as their relatives and friends).

In this paper we study the problem of using location-based social networks for efficient dissemination of information under time constraints. Specifically, we examine how efficiently a location-based social network, such as Twitter, can be deployed for emergency notification. Twitter has the ability to broadcast and forward messages to users and is primarily used via mobile devices, allowing users to be informed at anytime, anywhere, as long as they can access the network. We have chosen the network of Twitter over other Online Social Networks as it is indisputably one of the most widely adopted social network with more than 80% of its users being active mobile users, and over 500M tweets published daily on Twitter,³ as opposed to Facebook that presents 55M status updates per day⁴ and Google+ that, despite its growth, 90% of the users have never published a post.⁵ Consequently, we consider Twitter to be the most appropriate network for emergency related information propagation, since users on Twitter are more actively engaged to the network. Albeit we focus on the network of Twitter, we note that other social networks may be exploited to infer users׳ relationships and propagate emergency information.

Our objective is stated as follows: Given a location-based social network comprising a number of mobile users, the social relationships among the users, the set of recipients, and the corresponding timeliness requirements, our goal is to select an appropriate subset of users to propagate the information such that (1) the expected spread of information among users related to the event is maximized, (2) time constraints in the dissemination of the information are satisfied, and (3) costs are considered. Cost is defined as the amount of messages exchanged among users. Thus, it could be either monetary (for an SMS) or resource allocation cost. Our primary focus is on information that needs to be propagated under time constraints, such as emergency information, where the notification about the event needs to be propagated under strict time constraints.

We approach the problem in two phases. First, we use a crawling phase where User Profiles are built, social relationships are inferred and effective dissemination paths among the users of the social network are computed. In the second phase, namely reaction phase, we aim at reducing the search space by considering only users in the social network that are interested in the event (i.e., users related to the event) since the event may not be of interest for all users in the network. Then, we select a small number of seed users that will allow us to efficiently disseminate the emergency information to all interested recipients during the emergency event. Reduction of costs is accomplished by avoiding push-based broadcasts, which is important in emergency events as communication is typically over-utilized in such scenarios [8].

Existing information dissemination techniques are not adequate to solve these problems. The problem of maximizing the spread of influence in social networks has been addressed in [11], [12], [13], [14], but none of these works consider time constraints. Only recent efforts recognize that time plays an important role in the influence spread [4], [15]. However, contrary to our approach, these efforts assume that the influence flow is known and aim at maximizing the influence in the entire network rather than identifying and informing an appropriate subset of users that would be most interested in the information. Furthermore, both works study cases of viral marketing campaigns or voting systems, rather than emergency response situations that have to operate under tight time deadlines and resource savings. In [16] an approach is proposed to maximize the influence in a subset of users, rather in the entire network, while minimizing the number of users involved in initiating the propagation process. However, time constraints are ignored.

Emergency response outside social networks has also been studied. The use of geographical notification systems has been considered in [17]. The purpose of the presented system is to construct overlays that support location-based regional multicasting while they also consider issues of providing reliable storage of social information under extreme regional conditions. Traditional approaches such as multicast [18], [19] and publish/subscribe systems [20] are not appropriate for our setting since they will inform only subscribed users, while we need to alert all users associated with the emergency event, that is, all users that would be interested in the event and not just subscribed users. Furthermore, the set of users to be informed by our system is determined based not only on locational criteria, but also on relationship criteria, so it is not considered to be a strictly location-based approach. On the other hand, approaches like flooding and gossiping [21], [22], [23] will inform most of the users interested in the event, but they will also produce a lot of traffic and load in the network and undesirable spamming to the rest of the users. Contrary to our work, none of these approaches take real-time constraints into account and no social networks are exploited in the dissemination process.

We summarize the main contributions of our work below:

• We present LATITuDE (using Location based sociAl networks for Time constrained InformaTion DissEmination), our system that solves the problem of efficient dissemination of emergency information in location based social networks under time constraints. We show that the problem of selecting an appropriate influential set of individuals to maximize the spread of information is NP-hard and provide a greedy algorithm to solve it. Unlike previous works, we do not aim in influence maximization but rather at reachability maximization for a subset of users under real-time constraints.

• We investigate a number of metrics for estimating the appropriate set of users to initiate the propagation of the emergency information on the network. Specifically we estimate effective propagation routes among users by considering a number of metrics for penalizing long routes and present their effect in the seed selection process and the information diffusion.

• We extend our approach to be able to consider the authoritativeness of users in the spread of information. We investigate different approaches for determining the authorities on a social network and exploit the impact of authorities in the dissemination process. We determine authorities based on one of the following criteria:

  • Their popularity on the network, expressed as the number of followers.

  • Their importance in the connectivity of the network, expressed by the betweenness centrality metric.

  • Their ability to reach a large volume of users, determined by the spread achieved if they were to initiate propagation.

• We perform an extensive experimental study to validate our approach. We verify our solution using multiple datasets:

  • We use a dataset of tweets of general interest in the area of Dublin. The dataset consists of approximately 176,000 tweets collected from December 2012 to March 2013. The dataset presents tweets exchanged among users in the area of Dublin and in various cases and is not directly connected to emergency events.

  • We exploit a dataset of tweets related to a major emergency event. That is a dataset of tweets related to the Sandy Hurricane, a major emergency event that occurred in 2012 and severely affected the area of New York City [24]. The dataset consists of 73,325 tweets, collected from October 22 to December 2012.

  • We also conducted experiments using a dataset of tweets related to the floods in Germany. That is an emergency event of large scale that received nation-wide attention. The dataset consists of 5.5 million tweets and contains tweets from 22 November 2012 to 24 July 2013.

  • To further investigate the impact of the structure of the underlying network on the performance of LATITuDE, we used a snapshot of the DBLP citation network. Due to the reciprocity of the relationship among the users, the network of authors in the DBLP dataset differs significantly from the network of the Twitter users. The dataset consists of 954,951 authors and 3,338,893 connections among them.

• Finally, we demonstrate the performance of LATITuDE when applied to the Twitter stream by exploiting streaming data from Twitter related to traffic events in the area of DublIn:

  • The streaming data consist of 9357 tweets in total, published between 22 January 2015 and 3 June 2015 from 5090 users.

Our detailed experimental results illustrate that when an emergency event occurs, our approach is practical and effectively addresses the problem of informing a large amount of users within a deadline, with the least messages, while outperforming its competitors under various circumstances.