Available online www.jocpr.com Journal of Chemical and Pharmaceutical Research, 2015, 7(3):1388-1392 Research Article ISSN : 0975-7384 CODEN(USA) : JCPRC5 E-commerce recommendation system on cloud computing Xing Guo-zheng, Jiang Yu-yan, Li Chuang-xun and Wu Chao School of Management Science and Engineering, Anhui University of Technology, Ma anshan ABSTRACT Cloud computing is an emerging business model of computation, and has been paid widespread attention and application because of its clear commercial pattern. This paper gives detailed introduction about three aspects: cloud computing platform, the consumer behavior analysis of electronic commerce enterprise in cloud computing platform, and how to construct a recommendation engine commercial system. Through research on the system architecture, the platform development, consumer behavior analysis, electronic commerce application etc, we can construct a new intelligent and e-commerce personalized recommendation system. By providing dynamic and real-time personalized service, it achieves personalized and intelligent commerce recommendation engine. Key words: cloud computing; consumer behavior analysis; recommendation system INTRODUCTION With the development of Internet technology, there are more and more extensive applications of electronic commerce. But with the continuous growth of Internet data and content, the bottleneck of the development of e-commerce has become increasingly prominent [1]. Despite being an important tool of e-commerce, the development of the recommendation engine is limited. As a new network application architecture, cloud computing s development is booming. Cloud computing is proposed and developed, and lay the foundation for the massive data processing and the application in the cluster by the electronic commerce recommendation engine. Distributed system can effectively solve the I/O bottleneck problem and massive data storage of high performance computing system [2]. Its design characteristic of multi-machine environment stimulates the creation of the mass data storage. By distributed file system, mass data in the Internet can be stored in a plurality of nodes. Map Reduce [3] is a distributed computing model. Due to the excellent performance of it and the distributed storage system, it has been widely used in scientific computing, Internet service, and massive data processing. 1. Concept of Cloud Computing Based on the IT technology and the network technology, the electronic commerce activity must be equipped with the related technology capability of the construction, maintenance and management electronic commerce website. With the rapid development of e-commerce enterprises, some technology problems emerged, such as mass data storage, data mining, data integration, information security and information processing etc, which is a great challenge to the enterprises. 1.1 Definition of Cloud Computing Cloud computing system is a distributed computing system, which is pay-per-view and provides all kinds of service for the consumers. The essence of the system is the dynamic deployment, dynamic allocation / reallocation and real-time monitoring of the virtualization computing and storage resource. Therefore it can provide some services such as the computing service which is satisfied with Qos, data storage service and platform service [4]. It shields the underlying heterogeneous software and hardware resource for consumers. 1388
1.2 Distributed File System The distributed File System (DFS) adopts master-slave architecture. A distributed file system is consisted of unique Name Node (master node) and a plurality of Data Node (child node). The external performance of DFS is an ordinary film system. The user can store and fetch any files by importing the corresponding names. It is divided into different data blocks that are stored in the data nodes with its name [5]. 1.3 Distributed Computing The origin of Map Reduce are two core operations in functional programming models: Map and Reduce. Map is the mapping of a set of data to another set of data. The mapping rules are specified by a Map function. Reduce is the specification of a set of data, which is done through a Reduce function. Map is a process of the data separation, while Reduce is a data merging process. 2. Electronic Commerce Recommendation Engine 2.1 Recommendation Engine 1) With special information filtering [6] (IF, Information Filtering) technology, the recommendation engine recommend the different content (such as movies, music, books, news, images, Webpage etc.) to the consumer who may be interested [7]. The recommendation engine is able to do so, typically through comparing the consumer s personal preference with the specific reference, and then trying to predict consumer s preference degree on some un-scored items. The reference we select may be extracted from the project itself based on social or community environment according to the consumer s location. 2) E-Commence Recommendation System provides commodity information and purchase advice for clients, and then help customers to complete purchase. The function of the intelligent recommendation system can be summarized as: transforming the browsers of the electronic commerce website into buyers, it can improve the cross selling capability of the electronic commerce website, and the customer s loyalty to the electronic commerce website. 2.2 Recommendation Engine Status 1) With the expanding scale of e-commerce websites, the general large-scale e-commerce system has millions of products and user. Growth in the number of users and products will lead to reduction of the performance of the algorithm operation. The recommendation algorithms based on web are facing the serious scalability problem. 2) The e-commerce recommendation system is combined with the electronic commerce application service system, and the viewer's patience is limited, so there exists a direct connection between the real-time and the effect of the recommendation system. But in large e-commerce recommendation system, because the amount of data accumulates through many years and grows rapidly, it is more and more difficult to guarantee the real-time of recommendation system. It is a new challenge to the recommendation counting and the recommendation system architecture. 3) In Data storage, the e-commerce recommendation system framework of the current mainstream as shown in Figure1.From Figure 1, we can see that the recommendation engine used by e-commerce enterprises at present basically adopts the B/S architecture, In general, and is deployed in one or several Web application servers to mine and deal with the user input, the sever log and the statistical data. 4) Fig.1 Mainstream architecture of e-commerce recommendation system But with the data expansion, single server has been unable to meet the storage requirements. At the same time, as the number of the client increasing, the amount of concurrence user in the unit time continues to increase. This situation will consequentially lead to great pressure to the server and lengthen the response time to the client. Although the 1389
above bottleneck can be solved by increasing the amount of diskette to expand the storage or even adopting diskette array to achieve load balancing, there still exist some problems such as I/O bottleneck in reading and writing, too concentrated hardware device, too high hardware cost, too large network bandwidth consumption, etc. 3. E-Commerce Recommendation System Design Cloud Computing Environment 3.1 Design Of Recommendation System Architecture In Cloud Computing EnvironmentAiming at the characteristics of data storage form and distributed algorithm in cloud computing environment, this paper makes a detailed design of the recommendation system architecture according to the cloud computing. The basic structure is shown in Figure 2, including the persistence layer, the control layer and the presentation layer. 3.2 Fig.2 Architecture of recommendation system based on cloud computing The persistence layer, which is the distributed storage system of the cloud computing platform, is consisted of the distributed file system and the distributed database system. As there is a large number of users on the Internet and each user has its corresponding log data, the distributed storage system is the foundation of e-commerce recommendation system, and makes important basic function to solve problems such as the user behavior analysis and the association rules generated by the recommendation system. The control layer, which can succeed in building the resource mode and index by the distributed computing the Map Reduce program, contains a large number of complex intelligent recommendation algorithms and information processing. Ordinary users and the system users can interact through the development presentation layer and the recommendation engine. This architecture can shield all the underlying data storage structure and the business processing program for users, and then provide fast and efficient service. 3.3 Design Of Distributed Storage System With the continuous growth of the Internet data and content, the amount of data becomes very large on the Internet so that it is difficult for users to find the information they need, through a single server processing or a small cluster processing. The massive data storage has become the key factor to restrict the development of recommendation engine. Recommendation engines are data-intensive applications, whose system capability greatly depends on the underlying file system. If simply relying on the basic function of the file system provided by the operating system, the recommendation engine system will not be able to obtain the ideal performance. With high throughput and high I/O wideband propagation characteristics, the distributed file system can construct a global storage system though organizing the HD with multiple nodes to provide a polymeric storage capacity and I/O broadband, and is easy to expand with the expansion of the system. This article has carried on the detailed design on the distributed storage system. The data is divided into unit block which can be stored in any one disk of the colony. At the same time, a block is copied into multiple copies and distributed in different disks, which ensures the fault tolerance capability and the parallel copy capability. It is completely transparent to read and write files for users. The user operates the files in distributed system like in a personal computer. It is conducive to read and write files in the recommendation system. The architecture of distributed storage system proposed by this paper is shown in figure 3. 1390
Fig. 3 Architecture of distributed file system 3.4 Design Process Of Intelligent Recommendation With the increasing amount of Web data, how to find the useful pattern from the mass data has become an important task of the recommendation engine. The Map Reduce programming model is proposed to provide a powerful tool for mass data processing method. The key problem of massive data mining is the parallelization of data mining, a parallel algorithm. Cloud computing uses MapReduce and other new computing model, which means that the existing data mining algorithm and parallelization strategies cannot be applied directly to the cloud computing platform and should be transformed properly. There are many excellent data mining algorithms which have been transplanted into cloud computing platform. For example, the [8] PFP Growth algorithm, K-means[9] algorithm, decision tree [10] algorithm. In this paper, we propose a new intelligent business recommendation process model combining the personalized recommendation method with cloud computing technology. The model can effectively solve the problem that the intelligent recommendation method can't be combined with the actual cloud computing platform, and make full use of various computing resources in cloud computing environment to provide accurate personalized service for the user. The basic structure of the model is shown in figure 4. Fig. 4 Model of intelligent recommendation process The main function of source data is to collect and arrange the log data of the client. The data preprocessing part mainly deals with the rough initial data by eliminating noise and dimension reduction. The module plays a vital role in the recommendation system, and is the main factor that influences the accuracy of the recommendation system. It mainly is consisted of data cleaning, user identification, sequence identification, session identification. The pattern discovery mainly analyzes data with intelligent recommendation algorithms and discovers the hidden modes to 1391
produce association rules. Mode analysis can use a variety of mode analysis tools to obtain the final recommendation results, which should be provided for the user. CONCLUSION With powerful storage, operation and safety features, and ideal resource allocation and sharing mode, Cloud computing lays a good foundation for the development of electronic commerce recommendation engines. Therefore, a new commerce recommendation model arises. But the electronic commerce recommendation system based on cloud computing is still in the initial stage of exploration and application. For the practical application, there are still many imperfections of the new operation mode. With cloud computing technology gradually being perfected and matured, Commerce Recommendation System Based on cloud computing will get a rapid development. Acknowledgment We would like to thank the anonymous reviewers for their helpful comments. This work was supported by the National Natural Science Fund (No.71172219) and the Natural Science Foundation of Anhui Provincial Education Department under Grant No.KJ2011Z039, No.KJ2013A053 REFERENCES [1] Weihua Wu.The research on electronic commerce development mode in the cloud computing environment, 30 (5), 147-151, 2011. [2] Shoubin Dong, Tiezhu Zhao. Performance analysis of distributed file system for search engine 39 (4), 7-14, 2011. [3] Dean J. Ghemawat S. MapReduce: simplified data processing on large clusters 51 (1), 107-113, 2008. [4] Jianxun Zhang, ZhiminGu and Chao zheng. The summarize of cloud computing development 27 (2): 430-433, 2010. [5] Leran Lin and Delong Chen. computer knowledge and technology 5 (33), 9429-9430, 2009. [6] Shardanand U and Maes P. Social information filtering: algorithms for automating "word of mouth" ACM Press/Addison-Wesley Publishing Co:pp 107-114, 1995. [7] Li Yu and Lu Liu. The research on e-commerce recommendation personalized system computer integrated manufacturing 10 (10), 1036-1313, 2004. [8] Li H, Wang Y, Zhang D, et al. Pfp: parallel FP-growth for query recommendation ACM: 107-114, 2008. [9] Lei Xiaofeng, Xie Kunqing and Lin fan. An efficient clustering about local optimality based on K-Means algorithm 19 (7), 1683-1692, 2008. [10] Safavian S R and Landgrebe D. A survey of decision tree classifier methodology 21 (3), 660-674, 1991. 1392