Font Size: a A A

Research On Personalized Web Service Recommendation Based On Unknown QoS Prediction

Posted on:2016-10-15Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y MaFull Text:PDF
GTID:1108330482460399Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
An important research field of services computing is Web service recommendation which concern two key problems:1) predictions of unknown Web service QoS values and 2) user preference oriented overall QoS estimation. The first problem originates from that unfamiliar Web services are often recommended to users, which means that recommenders have no knowledge of user-dependent QoS values resulting that predictions of unknown QoS values are necessary for Web service recommendations. Users’ various concerns on multiple QoS properties bring the second problem, and hence recommenders have to determine the suitable recommendations via overall estimations over multiple QoS properties, as users.Widely attentions have been paid to this two problems and many achievements have been got, but current QoS prediction accuracies and rationalities of overall QoS estimation are still not satisfied. Therefore, this paper have studied this two problems thoroughly, by which the following achievements are got.(1) For QoS predictions without spatio-temporal information, most existing QoS prediction methods are inspired by CF ideas which originate from subjective data prediction. However, QoS data related to Web services is objective, and some significant differences between subjective and objective data which may bring errors to the QoS prediction are ignored by existing methods. In allusion to this problem, this paper presents some important characteristics of QoS data by lots of experiments, based on which substantial improvements of traditional CF are made, allowing the unknown QoS values can be predicted accurately without spatio-temporal properties.(2) For the QoS predictions in high dimensional spatio-temporal environments, most existing methods concern characteristic of one special dimension with neglecting the integral structure of high dimensional QoS data, which brings the flaws that inaccurate predictions, lack of expansibility and difficult of employing multiple dimensions. In allusion to this problem, this paper proposed a QoS prediction approach considering all QoS dimensions wholly and uniformly to predict multi-dimensional QoS accurately and easily. Proposed approach models the multi-dimensional QoS data as a tensor first, then finds out its component matrixes by decomposing the QoS tensor. These component matrixes allow us to reconstruct the tensor accurately. Finally, the reconstructed tensor tells us the prediction of the unknown QoS values. Proposed approach predicts Web service QoS data more accurately than other existing approaches.(3) Most existing approaches cannot perform overall QoS evaluations accurately and easily. This paper proposed an overall QoS prediction method based on user preference learning. The proposed method establishes an equations model used for user preference learning, by using historical QoS values as independent variables, by using user preferences as parameters of independent variables and by using historical user ratings as dependent variables. The proposed method allows user preferences to be obtained accurately and easily, thereby enabling the overall QoS to be evaluated accurately.(4) In allusion to the cold start problem, i.e., QoS cannot be predicted precisely and user preferences cannot be learned accurately, which are caused by the lack of historical data, a Web service recommendation via network neighbor was proposed. This method solved two core problems of cold start, i.e., recommending Web services to new users and recommending new Web service to users, by finding network neighbors via Internet distance measure, by determine Web service candidates via network neighbors, by QoS predictions via network neighbors and by overall QoS estimations considering subjective and objective weight.
Keywords/Search Tags:Web service recommendation, QoS, user preference, collaborative filtering, spatio-temporal information, tensor, cold start
PDF Full Text Request
Related items