Fatima projected for identifying Fake Profiles in OSNs. The

Fatima Qayyum Mahnoor Shaukat Department of Computer Science Department of Computer Science Fatima Jinnah Women University, Pakistan Fatima Jinnah Women University, Pakistan E-mail: [email protected] E-mail: [email protected] Abstract- Online Social Networks (OSNs) are a great place for scammers to copy the identities of users via creating fake profiles. Fake profiles are a widespread means for the impostors used to carry out nasty activities for example harming persons’ repute and privacy in (OSN). Henceforth, identifying the identities of fake profiles is a serious security complication in OSNs. In this paper, a detection mechanism called Fake Profiles Recognizer (FPR) is projected for identifying Fake Profiles in OSNs. The detection procedure is based on the functionality of Regular Expression and Deterministic Finite Automaton (DFA). The testing is done on three Online Social Networks: Facebook, Google+, and Twitter. The outcomes discovered high accuracy, efficiency, and low False Positive Rate. Index Term- Security, Privacy, Deterministic Finite Automata, Regular Expression, Online Social Networks I. INTRODUCTION1 Emerging web technologies allowed users on OSNs to create and exchange their personal information. OSNs such as Facebook, Google+ & Twitter allow users to present themselves as online profiles and setup social connections with other users. Due to the exposed nature of OSNs, users can appear in different Profiles views. Hence, verifying user’s identities are one of the serious issues from the security and privacy viewpoint. For saving users from intruders and scammers many of the researchers proposed systems to identify the fake identities. In this paper, a new system is introduced which is far better than the systems introduced before and is known as Fake Profiles Recognizer which is based on the two concepts of Automata I.e. Regular Expression and Deterministic Finite Automata. . . II. FRAME OF REFERENCE Some definitions and concepts on which the security framework is based are as follows: 1) Each node is represented with a unique Regular Expression; known as “a Friend Pattern”. 2) Each friend in the profiles friend list is marked with an instance of Friend Pattern to identify the identity. 3) Each node is linked with a predesigned Deterministic Finite Automaton to verify the identity of instances which can be obtained from Friend Pattern (Friend Pattern Processor FPP). FPP == DFA = (Q, ?, ?, q0, F) 4) Fake identity of any profile appears as a duplicated profile from the user’s friend list. For example, in a social network there are 2 user’s profiles and each profile is represented as a friend pattern, 1) ?= (a,b)? PA == FA = (a | b*) 2) ?= (k,l) ? PB == FB = (k | l)* From these we can make Friend List of each profile 1) PA == FA = (a | b*) ? {?,a,b,bb,bbb, ….. } 2) PB == FB = (k | l) * ? {?,k,l,kk,kl,lk,ll,kkk, ….. } III. PROPOSED SYSTEM (FAKE PROFILE RECOGNIZER) Copying user’s identity in Online Social Networks (OSNs) is one of the problems from security and privacy viewpoint. Scammers create fake profiles over OSNs to deceive a specific Ego profile of a specific user by mimicking his/her friend list. For this, a detection mechanism called Fake Profiles Recognizer (FPR) is proposed for identifying fake profiles in OSNs. And its functionality is based on Regular Expression & DFA (Deterministic Finite Automata). In the paper, the regular expression is known as Friend Pattern and DFA as Friend Pattern Processor. The regular expression forms the Friend Pattern for each profile in the social network. It is used to separate the profile from redundancy and duplication in OSN. For example, Bob is in the friend list of Alice but in an OSN maybe there are hundreds of profiles with the name Bob and 3 even having same features as Bob’s profile then how will Alice recognize the real Bob profile. So, for this purpose the social graph of OSN is remodeled in which every user’s identity is specified by a Friend Pattern and by this there will be decrease in the existence of fake profiles. This system will also help the users to identify their true friends and detect the fake profiles in a secured manner. The general architecture of Fake Profile Recognizer (FPR) is used to explore the Friend Pattern and verifies the incoming stream of instances of different Friend Patterns and then the associated FPP (Friend Pattern Processor) decides to accept or reject these instances. IV. RESULTS For testing the proposed system experiments were being conducted on three social networks i.e. Facebook, Google+ and Twitter. A fixed number of profiles were selected from all the social networks i.e. 1500 profiles as real ones and 1500 as fake ones. The source code of proposed system is in C++ and is implemented on Visual Studio. The results of evaluating the Completeness and the Exactness of proposed mechanism proved that it can effectively recognize the fake profiles, and achieved precision value 95.25% in average and recall value 94.53% in average. On the other hand, the mechanism attained low False Positive Rate value. The results proved that a small number of fake profiles couldn’t be recognized by the mechanism. V. CONCLUSION In this paper, a novel detection mechanism called Fake Profile Recognizer (FPR) is introduced for recognizing the identity of profiles. The functionality with FPR is using Regular Expression approach i.e. Friend Pattern to represent the identity of profiles in the social network, and using Deterministic Finite Automaton approach i.e. Friend Pattern Processor to recognize the identity of profiles. The mechanism is assessed on three networks: Facebook, Google+, and Twitter, and the results proved that FPR is effective to recognize fake profiles in OSNs. In addition, the proposed detection mechanism attained strong competitive results compared with other detection mechanisms in the literature. 4 CRITIQUE I. PURPOSE Fake accounts are a favorite tool for malevolent users of online social networks to send spam, commit deception, or abuse the system. A single malevolent actor may create hundreds to thousands of fake accounts to scale their operation to reach the maximum number of authentic users. Detecting and acting on these accounts is important as to protect authentic users and sustain the reliability of the network. For this purpose, an approach based on automata is used to identify and detect the fake profile in the OSN. II. HYPOSTHESIS Online Social Networks (OSNs) are gradually manipulating the way people communicate with each other to share personal information. Thus, the OSNs are gaining the interest of the malicious users who are trying to exploit the weaknesses of the OSNs. Increasing reports of the security threats in the OSNs is gaining the interest of security researchers who are trying to detect and lessen the threats for the users. As OSN have a lot of user’s personal data that can be used against them, preventing users from these attacks is a main challenge. To protect the users a lot of researchers proposed solutions to maintain the privacy but most of them failed to provide the right privacy to the users of OSN. After seeing the failure of most of the systems the authors tried to develop a mechanism which can protect the users by identifying the fake profiles present in the social networks. III. METHODOLOGY Nowadays people all around the world depend on online social networks (OSNs) to share knowledge; seek information; and expand personal connections. And thus, malicious acts like making faking accounts on OSN to destroy a person repute is very common so for this a very scalable approach is described to finding groups of fake accounts registered by the same actor. The main methodology is from Automata which are Regular Expression and DFA (Deterministic Finite Automata). The Regular Expression is used to generate the Friend Pattern of each user while on the other hand the DFA’s are used as a Friend Pattern Processor that recognizes all the instances that can be derived from a particular Friend Pattern. IV. RESULTS To check the correctness of the proposed system a test was conducted on three most used online social networks i.e. Facebook, Google+ and Twitter. The Friend Pattern which was used while testing was / (0* | 1) (1 | 01 * 0) /. From all the three OSN’s 1500 profiles were selected as genuine ones and 1500 as fake ones. And the three OSN’s were evaluated against Recall, Precision, F-Measure (Fl Score), Accuracy, Specificity, Fall-Out (False Positive Rate), and Area under the Curve (AUC) and the results obtained were great. The results showed the success of the proposed system by showing the accuracy as 94.91% which means that only 5.09% fake profiles couldn’t be recognized by the system. But the overall results were far better than the results of all the old proposed techniques. V. SHORTCOMINGS The possible short comings are that if the proposed system is to be used by the OSN’s then the social graph needs to be remodeled. For the research purposes only three of the OSN’s were remodeled and by remodeling we do not mean that the system is implemented on these remodeled OSN’s but we meant to say that the remodeling is done only for testing purpose and fortunately the proposed mechanism showed high accuracy but unfortunately remodeling the whole OSN graph is a great deal of work which is somehow not possible or we can say is very difficult. So, we can conclude by saying that the greatest difficulty is to remodel the OSN graph otherwise the system is great to use as it is providing security to the users by sensing the fake accounts. VI. ALTERNATIVE APPROACHES There are many alternative approaches for solving this problem but unfortunately, they are not providing higher accuracy rate as compared to this mechanism. Otherwise the concepts of Data Mining are also used to solve the problem of the security dilemma by which clusters of fake accounts can be detected. Another thing to be noticed is that the other approaches are only used on one or two OSN’s and thus are giving high accuracy but this system is for all the online social networks and still will give high accuracy rate in detecting the fake profiles is implemented. VII. COMPARISON Similar work is done with the names as: Support Vector Machine SVM, Social Honeypots, J48 and Naive Bayes. All of these systems provided good accuracy like Support Vector Machine SVM achieved accuracy 70.1% and Social Honeypots achieved accuracy 88.98% but it is still less than the FPR accuracy rate. And the FPR has low Fall-Out value compared with J48 (about 5.2% Fall-Out), and Naive Bayes (about 12.6%). According to F-Measure and Area under the Curve (AUC), FPR mechanism unfortunately achieved low values (about 94.89%, and 94.93% in average respectively) compared with J48 and Naive Bayes which achieved Fmeasure value around 96.5% and AUC value around 98%. 5 VIII. LESSON LEARNT We learnt that after purely gaining the concepts of Automata we can use the concepts of automata to solve a lot of daily life problems which cannot be solved otherwise. REFERENCES 1. Mohamed Torky, Ali Meligy, Hani Ibrahim. Recognizing Fake Identities in Online Social Networks Based on a Finite Automaton Approach. 2. Lee, K, Caverlee, J, and Webb, S. Uncovering social spammers: social honeypots+ machine learning. Proceedings of the 33rd international ACM SIGIR conference on Research and devel-opment in information retrieval. ACM, Geneva, Switzerland July 19 – 23, 2010. 3. Benevenuto F, Magno,G, Rodrigus,T , and Almedia, V, Detecting Spammers on Twitter, In 7th annual CoUaboration, Electronic Messaging, Anti-Abuse and, Spam Conference (CEAS) Vol. 6, Redmond, Washington, US, (2010).