AUTOMATIC TUNING AND SEMANTIC BASED CLONE DETECTION OF ANDROID APPLICATIONS

Transcription

1 International Journal of Mechanical Engineering and Technology (IJMET) Volume 8, Issue 10, October 2017, pp , Article ID: IJMET_08_10_020 Available online at ISSN Print: and ISSN Online: IAEME Publication Scopus Indexed AUTOMATIC TUNING AND SEMANTIC BASED CLONE DETECTION OF ANDROID APPLICATIONS Vishnu Priya T Department of Information Technology, Veltech University, Chennai, India Stephanie Vianna G Department of Information Technology, Veltech University, Chennai, India ABSTRACT A smart phone mainly depends on the app market. To avoid plagiarism and for detecting the similar Android application we use a tool named AnDarwin. It estimates various apps that have been collected from 17 markets. The AnDarwin includes major benefits as follows: it rejects evaluating apps pairwise, it examines only the app code. This can find both app that has partial and full similarities, and it can mechanically discover and eliminate library code from the similarity analysis. These indicate that the tool, AnDarwin is an effective tool to discover app which is rebranded and cloned and this could aid in developing the health of the market. However, the tool has various parameters based on the dataset the AnDarwin applied. Accurate selection of these parameter values would decrease the number of false positive and false negative rates. In order to increase the accuracy of the parameter selection, we use a method called Automatic Tuning and Semantic Based Clone Detection of Android Applications. Keywords: similar app detection, fully or partially cloned. Cite this Article: Vishnu Priya T and Stephanie Vianna G, Automatic Tuning and Semantic Based Clone Detection of Android Applications, International Journal of Mechanical Engineering and Technology 8(10), 2017, pp INTRODUCTION Android apps plays immense role in current world. The Android operating system is core for smart phone; most of the Android users depend on the Android market for their applications. The Android has both official and third-party markets which creates platform for users where they can download their application like social networking, games and many more apps. But in today s world the main issue is that the third-party markets do not provide healthy environment for developers. Many developers release their apps for paid or ad-supported version. The Hackers may crack the application s code and clone them and release them for editor@iaeme.com

2 Automatic Tuning and Semantic Based Clone Detection of Android Applications free. By plagiarizing the application hackers can get the revenue for that application. Thus the developers who actually choose to release the app for paid ones end up with financial crisis. Another threat to the application market is the low quality app called as spams. When the spam rate increases it will distort the healthy ecosystem. To ensure healthy market, an automatic tuning and semantic based clone detection method is followed. The apps can be compared for plagiarism using a tool called AnDARWIN. This tool rejects evaluating app pairwise, this increases scalability; it examines app code and does not depend on app s marketplace, signature; it can identify both full and partial cloned app; Further this tool is very efficient to discover cloned and rebranded apps. The AnDarwin tool compare the applications based on the parameter. Accurate selection of these parameter values would decrease the false positive and false negative rates. In order to increase the accuracy of the parameter selection, a method called Automatic Tuning and Semantic Based Clone Detection of Android Applications. In the proposed methodology, particle swarm optimization approach based automatic tuning of parameters has been introduced which will tune the parameter values in the optimized manner. The automatic tuning of parameters is done using the methodology called particle swarm optimization approach. Thus when the proposed technique of automatic tuning is introduced into the AnDarwin tool, the accuracy of the detection of App clones is increased by reducing the false positive and false negative rates. The paper is structured as follows: Section I illustrates a few introduction to AnDarwin tool and its need. Section II says about the background on the system. Section III highlights the threats faced in today s world and proposes a methodology in Section IV. The paper is concluded in Section V. 2. BACKGROUND 2.1 Android System Android users will get their apps from the app markets such as Google play store and other third party markets. The developer will get a developer key to upload his app to a market. But most of these markets are not efficient in finding the cloned apps, spams or malware. 2.2 Program Dependency Graph In a Program Dependency Graph each node is a statement and edge is the dependency between statements. Dependencies can be classified into two data and control dependency. If a variable in s2 depends on the value of s1 then those two statements are data dependent. If the second statement is controlled by the true value of the first statement then those two statements are controls dependent 3. THREAT MODEL Android applications though developed by different developers, share a huge amount of same code. Our goal is to find those cloned codes. The main targets are to find the plagiarists, spammers who release copied codes. These clones are detected only based on the codes, not by using meta data such as market, developer etc. The similarity is noted between the source code or the semantic of the application. 4. METHODOLOGY AnDarwin tool works by finding the parameter. Every application has several parameters which can be tuned with the dataset. These parameters which are used for LSH-Based clustering were found using trial and error method. In order to reduce the false positive and editor@iaeme.com

3 Vishnu Priya T and Stephanie Vianna G false negative rate accurate selection of parameter should be done. Particle Swarm Optimization algorithm can be used for automatic tuning of parameter in a optimized manner. These parameters are considered as Read external storage, read call log, Delete packages, Wake lock and so on. Extrema can be found using this algorithm in dimensional search space using swarm intelligence. Each particle consist of position xt, velocity vt. In AnDarwin parameters are considered as particles and change of parameters in each phase is the velocity. The factor K is used here to maintain the reasonable velocity as in Eq.1: K=2/ 2-φ-φ 2-4φ (1) Where, φ=c1+c2; φ>4. The AnDarwin tool consist of four stages as shown in Figure 1, first extracting semantic vectors form the apk files by using PDG Graphs, second is the identification of similar code by using LSH algorithm, third excluding the library code for simplicity and finally detecting the application similarity. a) Extraction of Semantic Vectors AnDarwin computes Program Dependence Graphs (PDG) for extracting the semantic vectors. In these graphs, each node becomes a statement along with edge that gives a dependency for those statements. PDG dependencies can be classifies into two, one is data PDG and the other is control PDG. If one statement s2 depends on s1 for its value then it is data dependent. Likewise, if the statement is executed on the true evaluation of other statement then it is said to be control dependent. Each PDG can be further split into numerous components that remain connected, where multiple computations whose data is independent, will occur in the similar method. Components that are connected and are called as semantic blocks. This represents the semantic information that are stored in dependency graph. AnDarwin calculates semantic vector for each of the semantic block. This information is gathered by calculating the node frequency from which semantic vectors are found. The threshold is set manually by analyzing all the apps and then discovering the similarities. Even a few lines of code can result in fairly large semantic blocks. Figure 1 System design editor@iaeme.com

4 Automatic Tuning and Semantic Based Clone Detection of Android Applications b) Parameter Selection PSO algorithm The Optimization algorithm, Particle Swarm Optimization is used to calculate the extrema in a very large dimensional search space using the swarm intelligence. The position of each particle is given a parameter set for the tool and its the change in the parameter set among generations is represented as velocity. There is a memory of previous fitness value and parameter value, pi that is highest among particles. The swarm algorithm is already alert of the fitness value. The velocity of the particle is focused on its parameters that is best, by a factor denoted as cognitive component, c1, and the best of swarm s parameters is its social component, c2. With an aim to reduce the effect when it is travelling in a velocity that is constant and to explore the search space in better way, the velocity contributions are increased by a random factor, r. in addition, the premature convergence is prevented and maintained in a reasonable velocity. This is referred to as a constriction factor. It serves in governing the velocity change between generations in the form of overall weight as shown in Eq.2: K= Where, φ = c1 + c2; φ > 4. At last, the velocity and parameter vector for each particle at each generation, t, after the first is computed as: v =K(v +c r (p x +c r (p x x =x +v c) Similar Code Identification by LSH In this, when two semantic blocks are checked and found the code to be clones, then they will share common nodes and, hence, their semantic vectors will remain same. From this, nearest neighbors for a semantic vector is found, potential code clones are identified. The Locality Sensitive Hashing (LSH) is an algorithm that can efficiently find approximate nearest neighbors even when there is a huge number of vectors. This is achieved by LSH by hashing the vectors using various hash functions from a special family. If the vectors are similar, then there is a high probability of collision. In order to find nearest neighbors, firstly LSH will hash all vectors using a special hash function and then will look for nearest neighbors within the hash collisions. Through this LSH is able to identify clusters approximately which AnDarwin will use to detect apps that are similar. By this way, LSH is able to cluster group quickly, because the individual group is smaller than the vectors set. Moreover, LSH computation is independent in each case which allows all the groups to be run in parallel. From LSH, AnDarwin is able to identify various number of clusters of semantic vectors blocks that could be code clones. Specifically, an index value is assigned to each clone and then represent it with a binary vector where the dimension is 1. This value says that the semantic vector is a member of the semantic block and 0 otherwise. This binary vector has many dimensions, each one to represent different code clones. d) Exclusion of Library Codes Using the clustered values, AnDarwin was able to find the libraries automatically. These libraries map to semantic vectors using the semantic blocks. When a single app contains a library then it will inherit all the semantic vectors resultant from that library code. These features derived from library code will tend to collide with other feature vectors too. In addition to that, features which represent boilerplate code and compiler-generated constructs will look much more frequent. These uncharacteristic features are excluded in AnDarwin by omitting the feature that appear in more than a threshold value that is set. (2) editor@iaeme.com

5 Vishnu Priya T and Stephanie Vianna G e) Detection of Similar Applications The characteristic features are used to find the similarities using two approaches, one for finding full app similarity and the other to find partial app similarity. Full app similarity detection, AnDarwin groups each app into a feature sets. In a simpler way, the feature sets for two very similar apps will be overlapping in most cases. Whereas in dissimilar apps, the feature sets will have a very little overlapping. This similarity among apps feature sets F A and F B can be found in the Jaccard Similarity, which reduces the problem of finding similar app to that of finding similar sets using (Eq.3) (,=! # (3)! # Partial app similarity detection: the tool, AnDarwin figures out all set of apps and their features. If two features are found to be similar between the app sets, based on the Jaccard similarity, these those features are said to be common for the apps set. It was able to find a non-trivial code, if a significant number of features share a same app. AnDarwin can detect partial cluster by creating clusters of features and comparing the similarities. Next is to find sets that are similar; this requires similarity stets. A comparison is performed among pair of sets. MinHash is a Locality Sensitive Hashing algorithm that starts initially by working out a signature for each and every feature set based on hashing and random permutations. Next the nearest neighbors are found that have a hash collision. Using Minhash, AnDarwin can efficiently detect both full and partial similarities among apps. The output of MinHash is a list of sets that pairs similar sets. With each set, a rapid cluster merging and element lookup is initialized, with each set in a cluster by itself. Each pair is processed, say (X, Y) and they get merged to the set, if they are not present previously. By repeating this process, the average similarity between sets and each cluster gets decreased with each pair processed. f) Detecting Commonly Injected Code Once the clustering is done, AnDarwin was able to detect for similarities among apps, if any code was injected into other app. Introduction of two concepts: cores and Diffs is done. Each and every cluster will contain a core that represents code that remain similar to all apps. AnDarwin will find the core by calculating the feature sets that intersect in the app cluster. This core is made as reference, and difference between sets are found by subtracting from the feature sets. The new set that was found behaves as a superset of the code that is modified. AnDarwin computes the cluster cores and their corresponding diffs among apps in each app cluster to detect both full and partial similarities. Later it clusters the app based on the MinHash to invade the apps that have similar code. g) Online Clone Detection Online clone detection methodology works in three steps. Firstly, it defines feature vector for each app. Among the semantic blocks, the semantic vector may or may not be seen. In case a semantic vector was already seen before, then no further processing is done for that particular vector which is already known. If a semantic vector is new then, LSH is used to figure out all nearest neighbors for that vector. It neither merges the semantic vector into an existing cluster based on its neighbors nor thus assign it to a feature, nor, if it has no close neighbors, it creates a new feature. The process is applies to each semantic vector and it forms a new app for that vector value. Once AnDarwin determines the feature vector for each app, it then eliminates those features that were previously identified as library features. Finally, the tool is able to find apps that exhibit either full or partial similarities editor@iaeme.com

6 Automatic Tuning and Semantic Based Clone Detection of Android Applications 5. CONCLUSION The problem of plagiarism in android application has to be resolved and it can be performed using the AnDarwin tool. This enhances the originality of android applications that is been developed at rapid speed. The advantage of the system is that it can find both full and partial similarities among the applications using semantics. REFERENCES [1] G. Eason, B. Noble, and I.N. Sneddon, On certain integrals of Lipschitz-Hankel type involving products of Bessel functions, Phil. Trans. Roy. Soc. London, vol. A247, pp , April [2] J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892, pp [3] I.S. Jacobs and C.P. Bean, Fine particles, thin films and exchange anisotropy, in Magnetism, vol. III, G.T. Rado and H. Suhl, Eds. New York: Academic, 1963, pp [4] Y. Yorozu, M. Hirano, K. Oka, and Y. Tagawa, Electron spectroscopy studies on magneto-optical media and plastic substrate interface, IEEE Transl. J. Magn. Japan, vol. 2, pp , August 1987 [Digests 9 th Annual Conf. Magnetics Japan, p. 301, 1982]. [5] M. Young, The Technical Writer s Handbook. Mill Valley, CA: University Science, 1989 [6] Namrata A. Patel, Vijay Ukani and Nilay Mistry, A Novel Methodology for Analyzing Malicious Behavior of the Android Applications, International Journal of Advanced Research In Engineering And Technology (IJARET), Volume 5, Issue 4, April (2014), pp [7] Kirandeep and Anu Garg, Implementing Security on Android Application, International Journal of Computer Engineering & Technology (IJCET), Volume 4, Issue 2, March April (2013), pp editor@iaeme.com