Popular Features. New Releases. Description This book offers new developments and high-quality studies into the area of web mining in E-commerce and E-service. It includes chapters on semantic web mining, web performance mining, web mining for social network analysis and for P2P services.
Back cover copy Web mining applications in E-commerce and E-services is a new research direction in the area of web mining.
- Steven Paul Scher: Essays on Literature and Music, 1967-2004 (Word and Music Studies, 5).
- Lost at Sea: The Jon Ronson Mysteries?
- Bestselling Series;
- Web Mining Applications in E-Commerce and E-Services.
Among all of the possible applications in web research, e-commerce and e-services have been identified as important domains for Web-mining techniques. Web-mining techniques also play an important role in e-commerce and eservices, proving to be useful tools for understanding how ecommerce and e-service Web sites and services are used. This book therefore collects new developments and high quality researches for the readers of this book to understand the topics of web mining applications in e-commerce and e-services as well as the state-of-the-arts in this area.
The chapters in this book include web usage mining and user browsing behavior analysis, semantic web mining, web performance mining, web mining for users' need understanding, web mining for social network analysis and web mining for P2P services.
One of the web mining directions is web usage mining, which focuses on techniques for identification of user navigation behavior within a website. This paper focuses on web usage mining for revenue maximization. However, other requests still need to be considered as they can also lead to final purchase; any frustration due to long response time can potentially result in the user leaving the site and loss of revenue, therefore good web mining can help understand user navigation behavior, improve workload management, and ultimately increase company revenue.
The idea of our approach takes the following into consideration: a only final purchase requests result in revenue contribution; b any other request can potentially lead to final purchase, depending on the likelihood of user navigation sequence starting from current request and leading to final purchase; c service differentiation and priority assignment are based first on aggregated confidence and then on average support of the composite association rules.
The paper is organized as follows. Both internet server resource management and web data mining have attracted extensive research in the literature. Internet resource management can be achieved through resource provisioning Urgaonkar et al. It is common practice to combine more than one of the techniques to achieve better results.
Khojasteh et al. It is common practice to employ both proactive and reactive approaches for long and short term resource provisioning Urgaonkar et al. In the early work, Xue et al. The algorithm works well only when the hosted applications have contrasting workload demands. Urgaonkar et al. Ashraf et al. The work in Khojasteh et al. In Wu et al. The authors in Hjort et al. Web usage mining has been used for a variety of purposes, including customer behavior studies Hjort et al.
It classifies users from their previous navigation patterns, and prioritize the subsequent requests in server overloading situations.localiser un portable samsung gratuit
E-commerce - Wikipedia
The proposed work in this paper uses a data mining technique to identify common navigation patterns from existing customers, then applies developed algorithms to requests from all existing or new users, who can potentially contribute to company revenue. The service differentiation is not applied at user level but at request level so that critical requests from less favorable users are not easily rejected. This paper uses first order Markov Chain for transition probabilities without consideration of previous navigation activities; this is mainly to avoid performance overheads in back tracking the activities.
It is important that in server overloading situations, any extra activities should be minimized. The initial weights of the navigation model are assigned based on realistic values and are regularly updated using statistics in server logs thereafter. A single association rule has confidence , that represents the likelihood of the rule and is denoted as C r , and support that represents number of traversals to the link and is denoted as S r.
Mining user navigation patterns in such a system can be viewed as a generation of association rules. The confidence of a composite association rule C r is defined as the product of the confidences of all the corresponding single rules, that is,. The support of rule r , denoted as S r represents the average number of times the links of r were traversed over the average number of times all of the links in the graph were traversed as described below:.
This paper mines user navigation patterns by using composite association rules to identify the subset of user navigation patterns that have higher probability of leading to final purchase. Calculation of the support and confidence values is conducted for each of the rules. It is assumed that a user session, which contributes to company revenue, consists of at least browse B , add to Cart C , logon O , and purchase P. From the rules in Table 1 , it can be seen that there are some direct links between some nodes and the final purchase O node, therefore it is assumed that there were previous navigation steps.
It is also observed that when two rules have same requests in the same direction, the longer rule will have smaller confidence value.
Using web mining in e-commerce applications
However, the support values can increase e. Once the association rules are identified, they can be used for algorithm development for AC and service differentiation. As we know that only final purchase requests result in revenue contribution, any other requests can potentially lead to final purchase, depending on the likelihood of the navigation sequence that starts from current request and ends at final purchase. Admission and service differentiation in this paper are only based on current and its next requests, which can be easily obtained from HTTP request headers.
Given any request and its next request, the probability of the request reaching the final purchase stage can be obtained from Table 2 , AC and differentiation decision can be made based on that. The aggregated confidence and support values of the composite association rules are recomputed based on updated weight values of the navigation model.
Since they i. When a server is overloaded, quality of service QoS will inevitably become poor. In the worst case, the service will become unavailable. It works by rejecting some less important requests to maintain the overall response time to an acceptable level.
ISPs may give their customers compensation for the rejected requests, depending on the contracted SLAs. From Table 1 , we can calculate the aggregated confidence and average support values for all direct edges that can potential lead to final purchase in the navigation model. In this paper, an AC algorithm is developed based on the derived association rules. Algorithm 1. Association rules based admission control. Initialization : assign initial threshold confidence C t. Obtain aggregated confidence C r from Table 2. Algorithm 2. Admission control and priority scheduling algorithm.
Get aggregated confidence C r and average support S r ;. The AC algorithm described in last section ensures the number of requests in the system remains at an appropriate level. However, the algorithm does not differentiate admitted requests, that is, all of them are processed in a FIFO order. In this paper, priority assignment is based on aggregated confidence and average support values as listed in Table 2.
When two requests have identical aggregate confidence values, the one with higher average support value will be assigned higher priority. All incoming requests will be placed in a priority queue based on their priorities and requests with highest priorities will be processed earlier. To maintain the overall response time at an appropriate level, AC is also applied by rejecting requests at the end of priority queue see Algorithm.
Web Mining – Online Web Data Mining Services, Solutions
The combined approach ensures that the most important requests will get the best service i. The simulation generates 1, random web users, each of which will have a sequence of web requests. The navigation sequences are based on the transition probabilities calculated from the navigation model in this paper. The performance comparison is based on the total revenue units generated from the approaches.
The association rules in Table 2 are used to set request priorities—based on current request and the next request, the higher the aggregation confidence, the higher the priority of the next request; when aggregated confidence is the same, aggregated support will be used for determining the priority of the next request.
In our simulation, when a request reaches final purchase, it results in one unit of revenue; when a customer leaves the cite without reaching the final purchase stage, or a request is rejected, no revenue will be generated. In the simulation, the probabilities in the navigation model are fixed, however, they can be slightly different overtime, therefore the model should be updated periodically based on statistics in the web logs.