A Research On Key Methods Of Personalized Web QoS Prediction Based On Invoked Charactors

With the rapid application of Web service technology, there are more and more Web services with similar functions appear on the Internet. But the quality of these services is not steady for dynamic network environment, different users’input and the status of web server. If obtain the accurate QoS information, we can discover or select the services from the large amount of services which can full satisfy the users’ requirements, and the users will have a better experience about the services. So QoS prediction becomes one of the key points in service computing field.Most QoS computation method is based on statistic at present. This kind of method is simple to realize but lack of consideration about personalized requirements. There are a few of personalized QoS prediction methods which usually are based on reputation, trust and collaborative filtering method. They can solve personalized requirements at a certain degree and they will provide different prediction values for different users. But there are some problems in them. First, they only do the prediction based on historic QoS information without adequate consideration about users’s requirements just like users’input data and other environment. This will lead to inaccurate prediction. Second, they lack of the sparse data processing when historic information is less. In addition, when the historic information is too large, the efficiency of the prediction is low and can’t provide the prediction for users in time. The QoS prediction methods are all scare of consideration about the problem above. It effect the users’satisfy degree, service discovery and selection.Aim to the problems mentioned above, this dissertation propose the personalized Web QoS prediction method based on users’invoked characters. It utilize the collaborative filtering method and data mining technology, from the angle of users, digging the invoked character pattern from historic execution information and monitoring information. QoS keep steady under these invoked patterns. So do prediction based on these patterns and collaborative filtering. This dissertation does the research about four facets followed.First, to solve the personalized prediction problem, this dissertation proposes the Web QoS prediction method based on collaborative filtering. This method, from the angle of users, analysis the users’personalized requirements and service environment characters. It does the prediction by the historic information of similar users. It considers the sparse data problem in collaborative filtering method and utilizes the information of similar services to fill the missing values in the matrix. Not only the invoke characters are considered, but also users’input character. The missing values’ filling make ready for prediction based on collaborative filtering.Second, in order to get the service used character pattern, this dissertation proposes the notion of service used character pattern. The pattern is the characters set which is the character of users usually invoke pattern. It also proposed a users’invoked pattern digging method based on DBSCAN. The method analyse the characters of services historic information and utilize the density based data spatial clustering method to digging the service using character pattern. Services quality keep relative stable under these patterns. Then the using character pattern can reduce the prediction time and increase the efficiency when the data is large. It is also the base of prediction based on users’using character pattern.Third, this dissertation proposes a QoS predict method based on users’using character pattern. It use the using character pattern obtained from last step and match the user’invoke character with character patterns. Especially when the target service has no historic invoked information under the matched pattern, this method adopts the similar patters’QoS which has experience on the target service. Aim to the pattern match question, this dissertation proposes a pattern match method based on gray correlation technology. It will calculate the correlate degree of the invoke character and QoS. So the Most Influential character will found and do the pattern match again by the sort of influential of the invoke character. This method improves the efficiency of the QoS prediction.Fourth, in order to validate the QoS prediction method performance, this dissertation design and implement the Web Service recommend tool oriented the composite services. The tool is embedded in Internetware environment sensing and evolving system. It response for create the initial instance and the candidate services. It utilizes the QoS prediction method mentioned above and obtained an accurate and high-efficiency prediction.