Proceedings of the International Conference on Big Data Cloud and Applications Tetuan, Morocco, May 25 - 26, 2015 Using Big Data Classification and Mining for the Decision-making 2.0 Process Rhizlane Seltani1,2 Noura Aknin1,2 1,2 1,2 Information Technology and Modeling Systems Research Information Technology and Modeling Systems Research Unit, LIROSA Laboratory Unit, LIROSA Laboratory Faculty of Science, Abdelmalek Essaadi University Faculty of Science, Abdelmalek Essaadi University Tetuan, Morocco Tetuan, Morocco sel.rhizlane@gmail.com aknin@ieee.org Souad Amjad1,2 Kamal Eddine El Kadiri2 1,2 2 Information Technology and Modeling Systems Research Computer Science, Operational Research and Applied Unit, LIROSA Laboratory Statistics Laboratory Faculty of Science, Abdelmalek Essaadi University Faculty of Science, Abdelmalek Essaadi University Tetuan, Morocco Tetuan, Morocco amjad_souad@uae.ma elkadiri@uae.ma Abstract—Web 2.0 is a revolution that has affected all decision-making systems, to generate more meaningful and areas, especially those of the new technology. Several new relevant decisions. concepts have emerged, and a large number of innovative applications continue to come out every day. However, the To classify and process data, various algorithms and social networking remains the racehorse of web 2.0, giving the techniques can be used. These methods differ depending on user at the same time, a space for communication and for data types. In the case of big data, to retrieve information, information sharing, which generates too much data, variable there are various analysis techniques with different and characterized by a great creation speed. So, we can call orientations and results, such as Representation-learning them big data, and consider them a very rich and interesting methods based geometric information, Stream Classification basis for decision-making. Algorithms, Associative Classifiers, etc. Big Data is a type of data which are characterized by the veracity, important volumes, and increasing variety and In this paper, we discuss some methods that we can use velocity, which makes their treatment and their processing by to classify big data in order to elaborate decisions, report the traditional database management tools a very difficult task. To strengths and the weaknesses. And therefore, present our overcome this problem, we opt for the big data classification global framework of decision-making 2.0 based on big data process. classification by describing the key pillars to be considered, In this paper, we make a study of some big data to lead well the classification process for the purpose of classification methods, which are the most significant to be decision-making. used to classify big data dedicated to decision-making, we detect their points of strength and weakness. Then we propose II. WEB 2.0 a framework summarizing the process of the formulation of the decision from the web 2.0 content, based on the big data A. Definition classification, and we specify the criteria to be taken into account when choosing the big data classification methods The web 2.0 is a combination of technologies, business intended for the decision-making. plans and social skills, which allow users to create web content, and to be more involved in the process of the Keywords—Web 2.0; Big Data; Decision-making; Data management of this content. It has brought many creative Classification concepts and techniques that did not exist before and which made the electronic life simpler and more enjoyable [1][2]. I. INTRODUCTION With the web 2.0, a new era of web use is born. Several applications have been developed and which have enriched The large variety of applications that appeared after the our lives by allowing more of interactivity and collaboration, emergence of the web 2.0, produce a huge mass of various such as blogs and social networks [3]. and diverse data. This wealth of information is a very important resource that we want to exploit to enrich our 29 B. Architecture and Principals  Data Volume: refers to a very important quantity of Web 2.0 is based on a varied and robust architecture, generated information. Data is considered as big founded on the introduction of new principles such as data if their size is very large, so we cannot control collaboration and interactivity, and the use of new them to make analysis easily. applications like web interface design techniques, those of  Data Variety: This makes analyzing this type of content syndication, XHTML, URL, etc [4]. data a very difficult mission. We have more There are several emerging principles with the different data presentation formats: text, audio, appearance of web 2.0, the most notable: image, etc.  Collaboration: This is an important aspect which  Data Velocity: It refers to the speed of creation and describes when a user has the opportunity to generation of data, which have been increased with contribute in the creation of the web content by the different new web applications. creating its own content.  Data Veracity: Data veracity refers to the  Interactivity: one of the introduced principles by the anomalies in data. Veracity in data analysis web 2.0, interactivity is reflected by the interaction constitutes the biggest challenge to overcome, of the user with the web content and with other because, veracity of data sources can largely affect users. the precision of analyzes. These two principles constitute new trends that have changed our lives and our way of working, they are the basis IV. BIG DATA CLASSIFICATION FOR DECISION-MAKING of social networks, blogs, wikis, etc. A. Clustering III. BIG DATA Clustering (also called Cluster Analysis), is a task of data mining, which means the mission of assembling a set of A. Definition objects, by the way that, objects which belong to the same The term big data refers to data sets exchanged by group have more similarities than with those belonging to connected objects in the web, and whose volumes are others groups. A group is called a cluster. The clustering was important and the variety and the velocity are increased [5]. used for the first time in the classification tasks by Cattell in It is a compilation of data sets which are characterized by 1943 for personality psychology classification [7]. Many complexity and large volume, so their management and clustering algorithms exist. Making the choice about which processing constitute a difficult task if we use traditional algorithm we must use, depends on the used cluster models [8]. Among the most distinctive cluster models, we find: database management tools [6]. Centroid models, Distribution models, Group models, and Connectivity models. B. Characteristics Compared to other types of data, big data are different In addition to its important role in the classification task, and have some specifications. These differences concern clustering has several advantages, such as the definition of several facets as the data format, their volume, the time information relating to the data, which were not revealed required for their creation, and their nature. before, as associations, so we can look for new patterns. Also, clustering provides a logical structure which makes The principal features are: Data volume, data velocity, results read and interpreted easily. But it is not the case, if we data variety, and data veracity. We can consider these opt for a large scale of clusters, because there are no elements as the characterizing pillars of big data (Fig. 1.), definitive methods to determine precisely the suitable and which make their processing and their analysis a special number of clusters. challenge. B. Decision Trees The decision tree is a technique which we can use for classification tasks, by creating a model to predict the output value based on a number of input values [9] [10]. To use decision trees for classification, we construct trees starting by the root of the tree, and subsequently, proceeding down to its leaves. A classification rule is developed based on example objects, which are known by their values of a collection of attributes. Then, the decision tree is expressed in function of the same attributes [11]. Decision trees constitute a good Fig. 1. Characterizing Pillars of Big Data way to well represent decisions. An example of a decision tree form is shown in the Fig. 2. 30  Classification by Emerging Patterns: based on emerging patterns from a sample, which means event associations whose supports vary, depending on the dataset [21].  Classification based on High-Order Pattern: is a classification system, which uses the algorithm of high-order pattern discovery, which detects considerable connection or association patterns using residual analysis in statistics [22].  Associative Classifiers based on the Apriori Algorithm: the Apriori Algorithm is an algorithm which proceeds by determining the prevalent items in the database. So, we can define association rules to wrap up trends in the database, many applications in various domains were done using this technique, such as market basket analysis [23]. Associative classification provides a high accuracy and Fig. 2. A General Form of a Decision Tree it is easy to understand. However, it presents some challenges, like the lack of obvious criteria to classify The decision trees are characterized by the robustness objects. Since it is based on a large number of rules, the and the simplicity of understanding and interpreting. What process of its elaboration is a time-consuming task, and it is important about decision trees is that we can treat becomes a difficult task to select the suitable ones to categorical and numerical data. On the other hand, decision develop the classifier. trees are instable, since a miniature change in the input data can affect the entire tree, by causing large changes in it. V. BIG DATA CLASSIFICATION AS A BASIS OF DECISION- C. Support Vector Machines MAKING 2.0 Support vector machines, more usually SVMs, were A. The Data Generation Process introduced the first time for binary classification. They refer to a collection of methods used for regression Web 2.0 is a very important source of information. The and classification, to analyze data in order to verify to which user interacts continuously with the web content through category an element belongs [12]. They can be used in collaborative applications, such as blogs, social networks, several ways depending on the nature of their application, etc. With the increase of the number of actors on the web, such as, text categorization, recognition of images, hand- the rate of information circulating on its channels increases. writing code, bioinformatics, etc. This large data flow generates the phenomenon of big data. Hence, web 2.0 is a rich platform of information, which can Some of the advantages of using SVM algorithms are: be treated to generate significant data. The user is primarily the robustness, the ability to learn well using a few parameters, and the computational efficiency. On the other a passive actor, becomes in an instant an active actor, by hand, apply SVM can at times require taking into transmitting opinions, which we propose to treat to ensure consideration many aspects of learning methods [13], SVM the mission of decision-making. These opinions can take, is oriented to be applicable directly in the case of two-class for example, the form of: tasks. For that reason, when we deal with a multi-class task,  A solution to a particular problem: a problem can be we must use algorithms that can reduce it to a set of binary solved quickly and efficiently if the process of the problems, or take account of all the classes at once by giving generation of the solution is collaborative. So the one formulation of optimization for all the data. Different methods of treating multi-class support vector machines reviews, including those of experts, about an issue continue to emerge [14]. may be of great use to make decisions to solve a given problem. D. Associative Classification  A feedback to a given subject: any feedback Associative classification refers to a classification which contains in itself a notice that we can use to extract is based on the use of association rules, by combining both useful information which enriches the process of the classification and mining of associations [15] [16]. decision making. Compared to other approaches, it is considered a highly  A proposal for improvement: in any field, accurate and competitive method, and can be applied in application, or system, we always look for ways of different ways [17] [18] [19] [20]. We can define three types improvement, especially in the case of business. of associative classification systems: Opinions of clients and in particular those which are 31 the most affected by the service, constitute a very In the decision-making 2.0 process, the classification important resource of inspiration to make the right serves as a passage from the raw data to the classified ones, decision of improvement. which will be used later to generate decisions. Data which circulate across the web, especially in social networks,  A complaint about a process, a product, a service: as blogs, etc, are difficult to track and manage. So to overcome with proposals for improvement, complaints also this problem, our classification process should follow some lead to the generation of significant decisions about specifications to properly carry out this mission. a product, a process, a service, etc. Taking into consideration our aim, which is decision- B. Decision-Making 2.0 Based Big Data Classification making based on the content reflected by the comments and Model the feedbacks of users, and to provide relevant decision, To exploit the generated data on the web 2.0, it is which must be generated based on meaningful data, our necessary to isolate the significant information. Circulating classification process must be efficient and suits our data through the web 2.0 applications such as social purpose. networks have the characteristics that make them a part of As already mentioned, the classification methods have what is called big data. To process them, we proposed to drawbacks as advantages. That is why, we opt for a adopt a classification process. combination, to elaborate a multiple classification model to When we want to treat data based on the web 2.0 exploit the strengths of the cited methods, taking into content, in order to make decisions. A simple comment or account different parameters, as shown in the Fig. 4. tweet can generate a large data stream, through feedbacks of  Accuracy: the classification process must guarantee users. Taking account of these data in decision-making is high accuracy, to ensure the relevance of our very important to harness the collective intelligence. decisions, which is a very important factor for the After a preliminary process of data streams, to centralize evaluation of the quality of the decision. those that meet our study needs, comes the classification  Facility of understanding: it is essential that phase to derive classified data according to specific classification must be a process that provides results parameters that depend on the issue in question. Finally, we which are easy to understand. It means also, that get the basis of decision-making. The framework which results must be interpreted without difficulties. presents the general process starting with the creation of the data on the web and ending with the decision-making is  Flexibility: flexibility is represented by the fact that represented in the Fig. 3. the classification can take into consideration categorical data, and not just the numerical ones, for more significant and common decisions. Fig. 4. Pillars of Big Data Classification for a Decision-making 2.0 Process Model Fig. 3. Process of the Generation of the Decision 2.0 Based on the Big Data Classification 32 VI. CONCLUSION [17] G. Dong, X. Zhang, L. Wong, and J. Li, “CAEP:classification by aggregating emerging patterns.” In Proceedings of The Second In this paper, we gave a vision on the results of a International Conference on Discovery Science (DS’99), pp. 43–55, developed study of the big data classification tools, we Japan, December 1999. presented a summary of the results concerning the [18] J. Li, G. Dong, K. Ramamohanarao, and L. Wong, “DeEPS: a new instance-based lazy discovery and classification system.” Machine techniques that we can use to treat data coming from web Learning, 54(2):99–124, 2004. 2.0, to ensure the decision-making mission. Then, we [19] W. Li, J. Han, and J. Pei, “CMAR: accurate and efficient presented a general framework of the entire process and classification based on multiple class-association rules.” In mentioned the criteria to take into consideration when Proceedings of The 2001 IEEE International Conference on Data Mining (ICDM’01), pp. 369–376, San Jose, CA, November 2001. choosing the classification method. [20] B.L.W.H.Y. Ma, “Integrating classification and association rule To exploit the strengths of the cited methods, we opt for mining.” In Proceedings of the Fourth ACM SIGKDD International a combination, to develop a multiple classification model, Conference on Knowledge Discovery and Data Mining, pp. 80–86, so that we can ensure three pillars of big data classification New York, NY, August 1998. [21] G. Dong and J. Li. “Efficient mining of emerging patterns: for a decision-making 2.0 process, which are accuracy, discovering trends and differences.” In S. Chaudhui and D. facility of understanding and flexibility. Madigan, editors, Proceedings of the Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 43–52. ACM Press, San Diego, CA, 1999. ACKNOWLEDGMENT [22] Y. Wang, High-order pattern discovery and analysis of discrete- The authors of this paper would like to thank our valued data sets. PhD thesis, University of Waterloo, Waterloo, Ontario, Canada, 1997. Research Team, Information Technology and Modeling [23] R. Agrawal, and R. Srikant, “Fast algorithms for mining association Systems Research Unit, and more generally, the Computer rules.” In Proc. 20th int. conf. very large data bases, VLDB (Vol. Science, Operational Research and Applied Statistics 1215, pp. 487-499, 1994. Laboratory, from the Faculty of Science, Abdelmalek Essaadi University of Tetuan, Morocco, for their great support. REFERENCES [1] T. O'reilly, “What is Web 2.0: Design patterns and business models for the next generation of software.” Communications & strategies, (1), 17, 2007. [2] T. O’Reilly, and J. Musser, Web 2.0 principles and best practices. O’Reilly Radar, 2006. [3] R. Kumar, J. Novak, and A. Tomkins, “Structure and evolution of online social networks.” In Link mining: models, algorithms, and applications, Springer New York 2010, pp. 337-357. [4] T. O'Reilly, What is web 2.0. O'Reilly Media, Inc, 2009. [5] P. Zikopoulos, and C. Eaton, Understanding big data. Analytics for enterprise class hadoop and streaming data, 2012. [6] E. Letouzé, “Big data for development: challenges & opportunities”. UN Global Pulse, 47, 2012. [7] R. B. Cattell, "The description of personality: basic traits resolved into clusters." Journal of Abnormal and Social Psychology 38: 476– 506, 1943. [8] V. Estivill-Castro, "Why so many clustering algorithms — a position paper." ACM SIGKDD Explorations Newsletter 4 (1): 65–75, 2002. [9] L. Rokach, Data mining with decision trees: theory and applications. World Scientific Pub Co Inc. ISBN 978-9812771711, 2008. [10] S. B. Kotsiantis, “Decision trees: a recent overview.” Artificial Intelligence Review, 39(4), 261-283, 2013. [11] J. R. Quinlan, “Induction of decision trees.” Machine learning, 1(1), 81-106,1986. [12] V. N. Vapnik, The nature of statistical learning. Springer-Verlag New York, 1995. [13] I. Steinwart, and A. Christmann, Support vector machines. Springer Science & Business Media, 2008. [14] C. W. Hsu, and C. J. Lin, “A comparison of methods for multiclass support vector machines.” Neural Networks, IEEE Transactions on, 13(2), 415-425, 2002. [15] Y. Wang, and A. K. C. Wong, “From association to classification: Inference using weight of evidence.” IEEE Trans. On Knowledge and Data Engineering, 15(3):764–767, 2003. [16] X. Yin, and J. Han, “CPAR: Classification based on predictive association rules.” In Proceedings 2003 SIAM International Conference on Data Mining(SDM’03), San Francisco, CA, May 2003, pp. 331– 335. 33