3
A Mobile Platform For Sociability-based Continuous Identification Fazel Anjomshoa , Burak Kantarci ? , Melike Erol-Kantarci ? , Stephanie Schuckers Department of Electrical and Computer Engineering, Clarkson University, NY, USA ? School of Electrical Engineering and Computer Science, University of Ottawa, ON, Canada E-mails: {anjomsm, sschucke}@clarkson.edu, ? {bkantarc, merolka2}@uottawa.ca Abstract—By exploiting the ever-expanding amount of data on smart devices and transforming it to meaningful information, it may be possible to uniquely identify a person via user behavior while interacting with their own smart devices. In previous works, identification of users via biometric properties such as fingerprint, iris has been successfully employed to increase the security of access. However, with the popularity of social networks, identification based on behavior over social networks is emerging as a novel concept. In this paper, We use real traces of data collected over several months. The feature set used in the paper includes location of users, their data usage, number of sessions and session duration for Facebook, Linkedin, WhatsApp, Skype and Twitter applications. The collected feature set is aggregated over time and analyzed using machine learning techniques. Index Terms—Behaviometrics; Continuous authenticatino so- ciability; Mobile computing I. I NTRODUCTION With the proliferation of smartphones and popularity of social network services (SNS), studying human behavior while they are interacting with their smartphones has been taken into consideration [1]. With mobile social networks, users can share their interest contents, looking for their desire contents, connect with other people and so on. By increasing the popularity of mobile social networks, these frameworks have captured considerable amount of information about users’ behavior that previously was not reachable. According to [2], these user data could be categorized into 6 different types; 1) Service data, 2) Disclosed data, 3) Entrusted data, 4) Incidental data, 5) Behavioral data and 6) Derived data. In this demo project we study behavioral data of users while they are interacting with their mobile social networks and use this type of data to analyze user identification based on social interactions. Traditional biometric-based user identification relies on the biometrics of a person such as fingerprint [3], iris [4], face [5] and so on. It performs recognition to allow access to a certain person or group of persons which is usually one-time and requires repeated interaction for validation of identities. On the other hand, continuous authentication is based on recognition of user behavior and contextual patterns. In order to achieve this goal, we apply behaviometrics referring to the behavior of a user which in our case this corresponds to mobile users’ activities on the most popular online social networks including Facebook, Figure 1: TrackMaison framework architecture and applica- tion functionality Linkedin, Skype, Twitter and WhatsApp, location of users and their Wi-Fi data usage. II. SYSTEM DESIGN The proposed framework is based on TrackMaison frame- work [6]. Figure 1 shows the architecture of the proposed framework. The framework consists of two main components: the frond-end android application which is mainly responsible for monitoring and capturing user data and the back-end behaviometric cloud server which is in charge of storing, mining and learning the data. A. TrackMaison Front-End The frond-end application is exclusively responsible for rolling up spatiotemporal information about social network services which is used on mobile devices. Moreover, the application also periodically monitoring users’ location for our future aim which is to have the behaviometric framework and the app communicates with the server through REST commands. The application is developed using Java for Android and works on Android versions 5.0 and above.

A Mobile Platform For Sociability-based Continuous ... · A Mobile Platform For Sociability-based Continuous Identification ... The application is developed using Java for Android

Embed Size (px)

Citation preview

A Mobile Platform For Sociability-basedContinuous Identification

Fazel Anjomshoa†, Burak Kantarci†?, Melike Erol-Kantarci†?, Stephanie Schuckers†† Department of Electrical and Computer Engineering, Clarkson University, NY, USA

? School of Electrical Engineering and Computer Science, University of Ottawa, ON, CanadaE-mails: †{anjomsm, sschucke}@clarkson.edu, ?{bkantarc, merolka2}@uottawa.ca

Abstract—By exploiting the ever-expanding amount of data onsmart devices and transforming it to meaningful information, itmay be possible to uniquely identify a person via user behaviorwhile interacting with their own smart devices. In previous works,identification of users via biometric properties such as fingerprint,iris has been successfully employed to increase the securityof access. However, with the popularity of social networks,identification based on behavior over social networks is emergingas a novel concept. In this paper, We use real traces of datacollected over several months. The feature set used in the paperincludes location of users, their data usage, number of sessionsand session duration for Facebook, Linkedin, WhatsApp, Skypeand Twitter applications. The collected feature set is aggregatedover time and analyzed using machine learning techniques.

Index Terms—Behaviometrics; Continuous authenticatino so-ciability; Mobile computing

I. INTRODUCTION

With the proliferation of smartphones and popularity ofsocial network services (SNS), studying human behavior whilethey are interacting with their smartphones has been taken intoconsideration [1]. With mobile social networks, users can sharetheir interest contents, looking for their desire contents, connectwith other people and so on. By increasing the popularityof mobile social networks, these frameworks have capturedconsiderable amount of information about users’ behavior thatpreviously was not reachable. According to [2], these user datacould be categorized into 6 different types; 1) Service data,2) Disclosed data, 3) Entrusted data, 4) Incidental data, 5)Behavioral data and 6) Derived data. In this demo project westudy behavioral data of users while they are interacting withtheir mobile social networks and use this type of data to analyzeuser identification based on social interactions. Traditionalbiometric-based user identification relies on the biometrics ofa person such as fingerprint [3], iris [4], face [5] and so on.It performs recognition to allow access to a certain personor group of persons which is usually one-time and requiresrepeated interaction for validation of identities. On the otherhand, continuous authentication is based on recognition of userbehavior and contextual patterns. In order to achieve this goal,we apply behaviometrics referring to the behavior of a userwhich in our case this corresponds to mobile users’ activitieson the most popular online social networks including Facebook,

Figure 1: TrackMaison framework architecture and applica-tion functionality

Linkedin, Skype, Twitter and WhatsApp, location of users andtheir Wi-Fi data usage.

II. SYSTEM DESIGN

The proposed framework is based on TrackMaison frame-work [6]. Figure 1 shows the architecture of the proposedframework. The framework consists of two main components:the frond-end android application which is mainly responsiblefor monitoring and capturing user data and the back-endbehaviometric cloud server which is in charge of storing, miningand learning the data.

A. TrackMaison Front-End

The frond-end application is exclusively responsible forrolling up spatiotemporal information about social networkservices which is used on mobile devices. Moreover, theapplication also periodically monitoring users’ location for ourfuture aim which is to have the behaviometric framework andthe app communicates with the server through REST commands.The application is developed using Java for Android and workson Android versions 5.0 and above.

Figure 2: TrackMaison Application UI

1) Social Network Services Monitoring: TrackMaison reapsthe usage data of the five SNSs. The app runs in the backgroundand monitors usage services by built-in monitoring capabilityavailable in an Android operating system. As showed in Fig. 1the data collection is stored in JSON format. The app storesthe data in sessions where each session contains three essentialparts: the duration of the session, the amount of data usedduring the session, and the initial location of the session. Theduration of the session refers to the duration from when oneof the the social network applications opens until it closesagain. The amount of data used is the amount of cellular orWi-Fi data consumed by the social network service application.As observed in Fig. 2the app provides the total data used byeach application visually in time granularity of hourly, daily,weekly and monthly. Also thanks to the Google map API, amap displaying the most recent location where the session wascreated.

2) Continuous Location Monitoring: The TrackMaisonfront-end application also is equipped with the capability ofmonitoring the users’ location periodically (every 5 minutes).The location information will send to the server for furtheranalyses. The feature is set to update the location within 30meters distance in minimum.

B. TrackMaison back-end

The back-end server is a local cloud behaviometric serverwhich is responsible to listen to the app messages always andacts toward the message appropriately. The server is in chargeof storing the session data, modeling the sociability metrics

Figure 3: Biometric authentication probability under DB-SCAN algorithm for sample user

(sociability rate and sociability factor), analyzing the databased on defined sociability models, normalizing the data andlearning from the data. The server is also in charge of sendingback the identification results along with the sociability ratesto the mobile app.

1) Server specification and functionality: The proposedserver is based on LAMP stack consists of open source softwareincluding Linux, Apache, MySQL and PHP. Since the ingressdata flow is continuous at the back-end of TrackMaison, theserver is equipped with on-demand analysis of incoming dataas shown in Figure 1. In more details, upon receiving the JSONsession file from the frond-end application, the server decodesthe JSON file and stores the data. Following that, the storeddata is sent to analysis module for the mining procedure usingPython. The normalized data is then stored in the server andupon receiving the normalized data, the server initiates theconnection to the application and sends the normalized databack to the users’ smartphone. Below is the explanation of thesociability rate and sociability factor metrics.

• Social Activity Rate: It refers to the amount of data that isconsumed by a user using social networking apps. The rateis normalized by considering the data from all participantsin the TrackMaison platform. To do that, we defined fourmetrics naming instantaneous, short term, overall andnormalized social activity rate. The instantaneous raterefers to the amount of data at each session where theshort term representing the amount of data in a dailybases. The weighted sum of consecutive short term socialactivity rates provide the overall social activity rate andthis is normalized by the maximum social activity rate inpool of active users.

• Sociability Factor: It refers to the amount of time that usersspend on social networking applications It is calculatedagain by the four metrics of instantaneous, short term,overall and normalized Sociability Factor.

A sample result is shown in Fig 3 which indicates anidentification performance by our platform for a random userunder the DBSCAN algorithm. At the end of the trainingperiod, the authentication error probability for the usersis below 0.1%. This stands for the case where biometricauthentication is triggered. A performance metric, namelythe authentication error probability is defined in order toevaluate the disruption probability of continuous authentication

of users on connected mobile devices. This corresponds to thesituation when user behavior on social network applicationshas been identified as an anomaly so the back-end server sendsa biometric authentication triggering signal to the front-enddevice. Hence we use the biometric authentication probabilityand authentication error probability interchangeably.

III. SUMMARY

Identification of users is critical for providing access and pre-venting unauthorized access to restricted systems. Traditionalidentification technologies rely on human biometrics and theyare repeated for each access request. As the growth in personaldevice usage steadily continuing, rich amount of data collectedfrom those devices may as well be used for identificationpurposes. We present a mobile platform framework to monitorthe usage of five social networking services on mobile smartdevices. We use real traces of data collected over severalmonths. The feature set used in the paper includes locationof users, their data usage, number of sessions and sessionduration for Facebook, Linkedin, WhatsApp, Skype and Twitterapplications. The proposed framework monitors the user data,trains the classifier and identifies users with high accuracy.

ACKNOWLEDGMENT

This material is based upon work supported by the Center forIdentification Technology and Research (CITeR) and the U.S.National Science Foundation under Grant No IIP-1068055.

REFERENCES

[1] N. Lathia, V. Pejovic, K. K. Rachuri, C. Mascolo, M. Musolesi, and P. J.Rentfrow, “Smartphones for large-scale behavior change interventions.”IEEE Pervasive Computing, vol. 12, no. 3, pp. 66–73, 2013.

[2] B. Schneier, “A taxonomy of social networking data,” Security & Privacy,

IEEE, vol. 8, no. 4, pp. 88–88, 2010.[3] S. B. Pan, D. Moon, Y. Gil, D. Ahn, and Y. Chung, “An ultra-low memory

fingerprint matching algorithm and its implementation on a 32-bit smartcard,” IEEE transactions on consumer electronics, vol. 49, no. 2, pp.453–459, 2003.

[4] S. Thavalengal, P. Bigioi, and P. Corcoran, “Iris authentication in handhelddevices-considerations for constraint-free acquisition,” IEEE Transactions

on Consumer Electronics, vol. 61, no. 2, pp. 245–253, 2015.[5] D.-J. Kim, K.-W. Chung, and K.-S. Hong, “Person authentication using

face, teeth and voice modalities for mobile device security,” IEEE

Transactions on Consumer Electronics, vol. 56, no. 4, pp. 2678–2685,2010.

[6] F. Anjomshoa, M. Catalfamo, D. Hecker, N. Helgeland, A. Rasch,B. Kantarci, M. Erol-Kantarci, and S. Schuckers, “Mobile behaviometricframework for sociability assessment and identification of smartphoneusers,” in Computers and Communication (ISCC), 2016 IEEE Symposium

on. IEEE, 2016, pp. 1084–1089.