User Behavior Mining in Software as a Service Environment

User Behavior Mining in Softw ar as a proceeds EnvironmentAbstract computer softw be as a Service (SaaS) provides bundle exercise vendors a Web based delivery ride to serve outstanding number of clients with multi-tenancy based infrastructure and application sharing architecture. With the growing of the SaaS business, data digging in the environment becomes achallenging sphere. In this paper, we suggest a new mensural along with a few existing metrics for guest abstract in a Software as a Service environment.Keywords Software as a Service, SaaS, Customer Behavior analysis, Data mining in SaaS EnvironmentI. IntroductionWith the rapid development of Internet Technology and the application software product usage, SaaS (Software as a Service) as a complete innovative model of software application delivery model is attracting much and more(prenominal) guests to use SaaS for reducing the software purchase and maintenance costs as it dope provide on-demand application software , and the drug users arse adjust the functions provided by redevelopments to meet changes in demand. SaaS is gaining speed with the appreciable increase in the number of vendors moving into this space1. The SaaS model is different from a regular nettsite model. In a regular website model, users of the software directly interact with the software application. But in the case of a SaaS model, users interact with the application through the service provider. The difference between a regular website model and a SaaS model can be shown in figure 1.Figure 1II. MotivationSoftware as a Service (SaaS) is being adopted by more and more software application vendors and enterprises 2.SaaS is beneficial for the clients in such a way that, a guest can unsubscribe from the services whenever he wants which makes it a take exception to manage node relationships. One of the characteristics of the SaaS business model is that one SaaS service needs to serve a large number of customers, among wh ich considerable portion are customers for whom services are offered on rill basis. As there is competition in the market, both trial and paying customers may move their business to other(a) service provider based on their requirements. It is essential for a service provider to retain the customers from migrating to some other service provider. Previous studies show that a lowly increase in retention rate would lead to a considerable increment in the new present nurse of the customers. To withstand the competition in the market, a service provider should satisfy the customers by understanding their current behavior and predicting their next move like if they are having any problems in apply the services, how much are the customers satisfied based on the seriousness and activeness of the customers.III. Related WorksA lot of work has been through in the area of analyzing the customers behavior on website model. Various methodologies are stated by various authors on various pro cesses in mining the web. In 3 Sindhu P Menon and Nagaratna P Hegde, analyze the views and methodologies stated by various authors on various processes in web mining.In 4 R. Suguna and D. Sharmila listed out work do by various authors in the web usage mining area.In 5 the authors Jiehui Ju. Et.al, gives a quick survey on SaaS. It covers key technologies in SaaS, difference between Application Software Provider and Software as a Service Provider, SaaS architecture and SaaS maturity model and the advantages that SaaS offers to small businesses.In 6, the authors Espadas et. al, presents the analysis of the impart of a set of requirements and proposes guidelines to be applied for application deployment in Software as a Service (SaaS) Environment.In 7, the authors Ning Duan, et. al, proposed an algorithm and two metrics which work with the coactionism among the users of a customer in a Software as a Service environment.IV. Problem DefinitionIn a SaaS Environment, an effective relation ship with the customer depends on how much the status of all(prenominal) customer is understood. In order to understand the status of a customer, it is necessary to study the behavior of ehte customer form time to time. It is necessary to predict the customers seriousness and activeness in exploitation the service. This prediction may help the service providers in improving their business strategies. In a business to customer website model, the mining is done based on selected metrics like visit frequency, average depth, average stay time etc. In the case of SaaS model, there is another level of users who actually use the service. So, regular user behavior metrics may not yield accurate results in the case of SaaS model. If individual customers users behavior is studied, then the difference between the customers may be identified.A lot of research is done on user behavior analysis in regular website model but those methods used for user behavior analysis may not guarantee accurate predictions. So an extra parameter or metric is to be considered. As in the SaaS model, a tenant is the direct customer of the service provider and the actual users of the service are the users of the customers, one way to study the behavior of the customers may be by summing up the individual users metrics of a customer to evaluate the customers behavior. But this way ignore the individual differences of the behaviors of the users of a customer. In addition to these regular web usage mining metrics if collaboration among the users is also considered in the analysis of customer behavior, it may yield better results than just using the regular metrics. But previous works done in user behavior analysis in SaaS uses only collaboration metrics in the analysis which ignores almost half of the analysis data.The experiment done aims at using collaboration metrics along with another metric which works with the data not considered in the collaboration metric calculation so that all the avai lable data is considered in user behavior prediction.V. ExperimentThe experiment is done in two phases, namely Data Collection descriptor and Data Processing Phase. In the Data collection phase, the necessary data (like server log files, dealings history, etc) are collected. In the second phase i.e. in Data Processing phase, the actual analysis takes place. This phase is further divided into individual modules like preprocessing, traffic pattern discovery, and pattern analysis.Preprocessing is a process of refining the break log data and feat history removing noise in data (if any) and populating database for further use in next modules. It includes data cleaning, user identification, session identification, transaction identification.Pattern Discovery is the process of discovering the usage patterns from the cleaned raw log data. As in this experiment, it is not regular usage patterns that are to be considered, collaboration patterns are to be considered. Regular usage pattern s are the sequences of activities that are performed by the users individually. But, collaboration patterns are those that are performed by users by interaction. collaboration patterns are not the transaction patterns rather they are the patterns of users that collaborate to perform a transaction.Definition of Collaboration Collaboration is defined to happen when different users of a customer work on the same business object during a certain period of time. For example, in a Human Resource management SaaS service, the vacation request is submitted by a regular employee user of a customer and then is approved/rejected by manager user of the same customer. Here two users of a customer are involved in the process of granting a leave. This is called collaboration.After the raw data is cleansed, the data used for collaboration discovery may contain dilate of the transactions performed by the users of any tenant with tenant id(tid), user id (uid), transaction id (transaction_id) (may als o be called business object id), date, time, service id (sid). In this table more than one user may be involved in the completing of a transaction.Algorithm Collaboration User Set IdentificationInput hedge 1 that consists of the transaction detailsOutput Collaboration_ put over with collaboration transaction detailsInitially Collaboration_Table is emptyGet first record from Table 1 enrol details into Collaboration_TableWhile end of table 1 not reachedGet next record from table 1Search for transaction_id in collaboration_TableIf found, update collaboration user set and no_of_usersElse Add details to collaboration_table as new recordTable 1 Sample table showing the contents of Table 1Table 2 Sample Collaboration TablePattern analysis plays vital role in the experiment. This module deals with the behavior analysis based on the collaboration patterns extracted higher up.From 7, there are two case of collaboration. They are random collaboration and repeated collaboration by certain gro up of users. The first type of collaboration can indicate the activeness of the customer no matter which users are involved in the collaboration process. It can be called as officious Collaboration Index (ACI). The second type of collaboration can be described by the usage patterns among the users of a customer. It can be called Patterned Collaboration Index (PCI). A advanced ACI value tells that a customer is actively using the SaaS service and if such a customer is still a trial customer, it probably shall be the high priority target to get it converted into paying customer. A high PCI value tells that a tenant is seriously using the SaaS service with relatively strong loyalty, cross-selling or up-selling opportunity can be explored for such a customer. The formula to calculate ACI and PCI are as followsThe AppCNorm is the normalizing grammatical constituent indicating collaboration characteristic of SaaS service. While some SaaS service are rich with collaborations and others may not be. In order to balance the difference among different SaaS services, this normalization factor is employed.Where Pni denotes the collaboration pattern i of customer n, N is the total number of customers, and m is the total number of patterns in customer n. supp(pni) is the support value of pattern Pni, and len(Pni) is the length of the pattern.These collaboration metrics works only with the collaboration data and neglects the remaining data which is almost half of the data. Hence another metric can be added along with the above metrics which considers the non-collaboration transactions. As the new metric is for non-collaboration transactions of a tenant, it can be called Average Usage Index (AUI). This can be calculated using the formulaThis AUI increases the accuracy of prediction of activeness of the customer along with ACI.VI. RESULTSFor this experiment, the data created is for 100 customers of a Software as a Service provider who is providing 6 different components of a n application as different services. Among these 100 customers, first 50 are taken as paid customers and the other 50 are taken as trial customers.Table 3 Summary of transactionsTable 4 Sample pattern listTable 5 Sample Calculated MetricsFrom the above calculated values, we can observe that though T0 is a paid customer, less ACI and PCI values indicate that this customer is not using the services to the full and then revenue generated from this particular customer is not appreciable. Rather, this customer may be planning to unsubscribe from the service and hence is an important target for the service provider to retain the customer. In the case of T45, it has high ACI value, high AUI value indicating active usage of the services and high PCI indicating that this customer is completely migrating his business onto the SaaS service generating the service provider more revenue. Among the sample trial calculated values, customer T50 is active and serious and hence, there is a high proba bility for this customer to convert into paid customer. On the other hand, customer T89 is not very active and is not serious indicating that he may be facing technical difficulties in using the services and hence should be helped with or is thinking to unsubscribe from the services.Table 6 Summary of Calculated metricsFrom the above table, for any tenant to be considered active in using the services, minimum ACI and AUI values needed are 1 and 1 respectively and minimum PCI value needed is 2.VII. ConclusionThe metrics ACI and PCI are introduced in previous works done by Ning Daun, et. al in 7 which works with collaboration data and passing the non collaboration data. In our work, a new metric is introduced AUI which considers the non collaboration data also in customer behavior analysis. Still further, frequent pattern analysis can be applied on this non collaboration data to get usage patterns and so the analysis can be further improved.VIII. References1 Wei Sun, Xing Zhang, Chan g Jie Gou, Pei Sun, Hui Su, IBM China search Lab, Beiing 100094, Software as a Service Configuration and Customization Perspective IEEE Congress on Services Part II, IEEE 2008.2 E. Knorr, Software as a Service The succeeding(prenominal) Big Thing, http//www. infoworld.com/article/06/03/20/76103_12FEsaas_1.html3 Sindhu P Menon, Nagaratna P Hegde, Requisite for Web Usage Mining A Survey, Special Issue of planetary Journal of Computer Science Informatics 2231-5292, Vol-II, Issue-1, 2, pp. 209-215.4 R. Suguna, D. Sharmila An Overview of Web Usage Mining, International Conference of Computer Applications (0975 8887), Vol. 39, No, 13, February 2012, pp. 11 13.5 Jiehui, et. al, Research on Key Technologu=ies in SaaS, International Conference on Intelligent Computing and Cognitive Informatics, 2010, pp. 384-387.6 Espadas et. al, Application Development over Software-as-a-Service platforms, The Third International Conference on Software technology Advances, 2008, pp. 87-104.7 Ning D uan, et. al, Tenant Behavior analysis in Software as a Service Environment Service Operations, Logistics and Informatics (SOLI), 2011 IEEE International Conference, pp 132-137, July 2011.