A Cloud Service Selection Method Based on Trust and User Preference Clustering

Yubiao Wang, Junhao Wen, Wei Zhou, Bamei Tao, Quanwang Wu, Zhiyong TaoIEEE Access 2019 · OpenReview

Abstract

In order to achieve universal and personalized cloud service choices in a social network environment, this paper proposes a cloud service selection method based on trust and user preference clustering (CSSM-TUPC). The method performs a comprehensive trust evaluation used to select cloud services. An improved condensed hierarchical clustering method based on user preference similarity is introduced to enhance recommendation trust accuracy. A cloud model-based approach measures user similarities, dividing users into domains to calculate intra-domain and extra-domain recommendation trust. Simulation experiments verify the accuracy and superiority of the clustering algorithm, showing improved transaction success rates and higher user satisfaction.

Introduction

As the global cloud market expands, users face a "paradox of choice" across Infrastructure (IaaS), Platform (PaaS), and Software (SaaS) offerings. Traditional selection methods often fail to account for the subjective nature of user preferences, trust uncertainty in dynamic environments, and the impact of malicious feedback. To address these challenges, the CSSM-TUPC method integrates multi-dimensional trust evaluation with a sophisticated user-clustering algorithm.

Methodology and System Architecture

The CSSM-TUPC framework is built on the principle that effective recommendations should come from users with similar preferences. The system architecture manages the flow from user requests to optimal service selection through a multi-layered trust management framework.

System architecture of the proposed cloud service selection method.

The method utilizes a Cloud Model to represent evaluations using three parameters: Expected Value, Entropy, and Hyper-entropy, capturing the inherent "fuzziness" of human judgment. Users are grouped using an Improved Condensed Hierarchical Clustering algorithm, which is optimized for speed and efficiency compared to traditional methods. Trust is then calculated as a combination of:

  • Direct Trust: Based on personal interaction history, weighted by time decay and transaction value.
  • Recommended Trust: Derived from both intra-domain (similar preference) and extra-domain feedback.

The Selection Process

The operational flow of the selection method ensures that services are evaluated rigorously before being called by the user.

Flowchart of the cloud service selection process.

As illustrated in the flowchart, the process begins with a user request (CSU). The system performs hierarchical clustering and identifies candidate services from the registration center. Comprehensive trust (CST) is calculated by merging direct and recommendation trust. Only if the CST meets a predefined threshold ($\varphi$) is the service call executed. Subsequent customer feedback is then fed back into the trust management engine to update the service's reputation.

Experimental Validation and Results

The researchers evaluated CSSM-TUPC using the MovieLens dataset for clustering accuracy, Epinions.com for social network interaction success, and CloudSim for performance in simulated cloud environments.

Key findings include:

  • Superior Accuracy: The method achieved lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) than traditional collaborative filtering models like BP-CF and Cosine-based approaches.
  • High Efficiency: The improved clustering algorithm significantly reduced processing time, completing a 120-sample test in 3.39 seconds compared to 60.1 seconds for the traditional version.
  • Robustness: The system maintained higher transaction success rates than competing models like PeerTrust, even in environments with malicious nodes, thanks to its penalty factors and preference-based filtering.

Conclusion

The CSSM-TUPC method offers a significant advancement in cloud service selection by prioritizing user preference similarity and multi-dimensional trust. While the model handles individual malicious nodes well, future research will focus on mitigating sophisticated collusion attacks and adapting the framework for mobile social networks.

Summary updated

Abstract

In order to achieve universal and personalized cloud service choices in a social network environment, we propose a cloud service selection method based on trust and user preference clustering. The method performs a comprehensive trust evaluation, which is used to evaluate and select cloud services. Meanwhile, we propose an improved condensed hierarchical clustering method based on user preference similarity to further improve the accuracy of recommendation trust. A cloud model-based approach is used to measure similarities between users, and then a hierarchical clustering method is used to divide users into different domains according to user similarity. The final recommendation trust will be obtained, which includes the intra-domain recommendation trust and the extra-domain recommendation trust. The comprehensive trust of cloud services, which consists of direct trust and recommended trust. Simulation experiments verify the accuracy and superiority of the clustering algorithm. Experimental results show that the cloud service selection method improves the transaction success rate and enables users to select more satisfactory cloud services.
BibTeX
@inproceedings{wang2019cloud,
  author = {Yubiao Wang and Junhao Wen and Wei Zhou and Bamei Tao and Quanwang Wu and Zhiyong Tao},
  title = {{A Cloud Service Selection Method Based on Trust and User Preference Clustering}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=aOTiDHrD77},
}
Figure