Elsevier

Neurocomputing

Volume 465, 20 November 2021, Pages 476-489
Neurocomputing

Ensemble deep neural network based quality of service prediction for cloud service recommendation

https://doi.org/10.1016/j.neucom.2021.08.110Get rights and content

Highlights

  • This study proposes novel approach for Quality of Service prediction for Cloud Service Recommendation.

  • A unique ensemble of Deep Learning models to learn user and service correlation is presented.

  • This approach combines neighbourhood based method and Neural Network models.

  • The proposed approach has better performance than other state-of-the-art approaches.

Abstract

Applications of Cloud Services are increasing day by day, and so is the difficulty of choosing the best-suited service for a customer. Quality of Service (QoS) parameters can be used for quality assurance and evaluation; further, a service can be recommended based on these QoS parameters’ values. Recommendation systems are getting much attention lately. It has a crucial role in almost all the major commercial platforms and many improvements are being made to make the recommendations more precise and closer to the user’s requirements. Conventional Machine Learning algorithms and statistical analysis methods, presently are not that efficient in learning the complex correlation between data elements. Lately, Deep Learning models have proven to be practical and precise in areas like natural language processing, image processing, data mining, & data interpretation. However, there are not many examples of complete Deep Learning applications for cloud service recommendation systems, though some works partially use Deep Learning. We propose the Ensemble of Deep Neural Networks (EDNN) method, which is of the hybrid type, i.e., the fusion of neighborhood-based and neural network model-based methods. The output obtained from both the models are combined using another different neural network model. Our approach for predicting QoS values is simple and different from previous works, and the results show that it outperforms other classical methods marginally.

Introduction

Cloud services are being extensively used in many fields and are actively supported by major industrial companies. To understand these services we should be able to quantify and measure the service parameters. Quality of Service (QoS) is typically used to characterize these cloud provider’s non-functional characteristics. With the increasing demand and proliferation of cloud services in the market, QoS has become an important point for marketing and differentiating functionally similar cloud services. To identify appropriate cloud services, users have to spend a significant amount of time and effort in searching and choosing the services. It’s a tedious, and less efficient method. Further, the suggestion of services based on users’ interests, historical records, or similar data about users is found to be appealing to customers and suppliers. Thus, it is important to obtain unbiased and precise client-side QoS values for different cloud services.

For different cloud service users, client-side cloud service assessments are needed [4], [21]. Customer-side cloud service research is practically difficult, primarily because of the following reasons:

  • Typically, users are not cloud service analysis specialists and aren’t familiar with most technical terminologies. Thus, it is complicated and unreliable for them to collect and process any data.

  • There is a specific cost involved with each service for its invocation, so it may not be feasible for a service user to use all of them.

  • Evaluating all cloud service applications is time-consuming and inefficient for service users, as many cloud services are available.

Thus, despite adequate tests on the client-side, it is impossible to obtain reliable QoS values for cloud services.

There are several methods that have been used to evaluate cloud services based on which these can be recommended. Recently, a considerable focus is on Collaborative filtering-based (CF) service recommendation methods [32], [41]. Based on the implementation, it is divided into two categories: Memory/Neighborhood-based and Model-based approaches. Memory-based CF approaches generally follow the following steps; Firstly, these methods find out similar users and similar services, called neighbor users and services, by computing the PCC similarity score among users and services. Then, they predict the unknown QoS values for a particular user and particular service by processing a sequence of steps with information retrieved in the last step.

The Model-based Collaborative Filtering approach needs to build a capable model based on training data and machine learning algorithms, integrating historical data from related customers to forecast the target service’s quality. Lately, the Matrix Factorization technique has seen an enormous increase in successful use cases of model-based prediction approaches.

However, predictive precision of CF-based approaches relies on neighbor selection efficiency, but in data scantiness scenario, neighbor selection quality is restricted because of less QoS related data availability. While model-based CF approaches do not depend upon neighbor selection, they directly learn latent features from service invocation; hence, under higher data sparsity conditions, model-based approaches have better prediction accuracy than memory-based models.

One of the widely used and efficient model-based approaches is Matrix Factorization (MF). As the name suggests, factorization is the core part of the this technique. A large matrix with a large dimension is converted into two smaller matrices with low dimensions called feature matrices. These matrices are multiplied and values are compared with the original matrix. Weights are updated according to the chosen algorithm (like stochastic gradient descent) until the difference in values is minimized. In our case, the user’s choices for a cloud service can depend on several factors, which can be considered as latent features in smaller matrices. MF works better than memory-based methods, even in cold start situations. But on the other side, Matrix Factorization has its features described manually, which limits its scalability. MF models are more general, and they aim to reduce some defined error function throughout the training. This method does not perform well with the increase in the size of the dataset. Hence, these models cannot learn the proper correlation between entities and treat them equally, which leads to a decline in prediction accuracy.

Lately, Deep Learning has made a great impact on fields like Image processing, Computer vision, Speech and Voice recognition. Some of the reasons why Deep Learning will be more suitable for recommendation systems than conventional methods are as follows: Nonlinear Transformation, Representation Learning, Sequence Modelling, Flexibility [48]. Deep learning can learn many complex correlations among several entities. A service user’s behavior may be non-linearly dependent on other users, which can be effectively learned using the Deep Learning model. Automatic Feature design makes it scalable and effortless and it has the support of some popular frameworks like Tensorflow, PyTorch, Keras, etc. Hence Deep Learning is a better option for service recommendation. The main contributions of this paper are as follows:

  • 1.

    A Neural Network based Ensemble of Deep Learning models (EDNN) to learn user and service correlation for QoS prediction in Cloud Services Recommendation.

  • 2.

    We have used Multi-layer Perceptron and have given a novel approach for weight prediction and balancing between User prediction and Service prediction using Deep Learning.

  • 3.

    We have carried out all the experiments on the WS-dream dataset; detailed analysis and results have been presented, which shows our prediction results to be accurate.

The remaining part of this paper is organized into several sections; Section 2 has a description of the related works done in the field of recommender systems for cloud services. Section 3 presents the approach and required details of implementation for our approach. Section 4 discusses the complexity and cost of operation. Section 5 explains the experimental requirements and configurations. Section 6 consists of results and analysis, and Section 7 concludes our work.

Section snippets

Literature survey

Cloud Services Recommendation has been under research for the past few years. Overview of the present state-of-the-art methods and their approach tells us about the progress of research and allows us to extend them further for a better solution to work in real-world scenarios.

Neighborhood based neural network approach

In the past few years, Deep Neural Networks has made a more significant contribution to diverse areas of intelligence, such as computer vision, image processing, data mining, speech recognition, etc. The discovery of deep and complex hidden features is what makes Deep Learning more optimum for compounded problems. However, the use cases of Deep learning in the area of service recommendation are new. By the literature survey, it is known that, nearly all the past pioneering methods are very

Complexity analysis

In this section, the Computational Complexity of various operations performed by our proposed method is discussed. Let us fix the shape of our dataset is m x n, where m is the number of users and n is the number of services.

Data description

To assess the prediction performance of our approach, we have used a dataset of real-world web service invocations, WS-Dream, collected by Zheng et al. [54] and Zheng and Lyu [57]. This dataset has two sub-datasets, the first one contains static Response Time and Throughput values, and the other one contains a similar set of data but in time-series fashion. For our experiment, we have used the first dataset with Response Time as the quality of service parameter. As shown in Table 4, the dataset

Result and analysis

Our approach is better in learning complex correlation from conventional Collaborative Filtering and Pure Matrix Factorization based approaches because it makes use of a Deep Neural Network-based model to learn both the user features and service features separately. Further, balance the prediction using a separate model that uses Deep Neural Network to automatically assign proper weights to the output of both user model and service model to predict the final result. Table 7, Table 8 proves our

Conclusion

Quality of Service (QoS) values for cloud service recommendation using Ensemble Deep Neural Network Models (EDNN) is predicted. The results obtained are better than conventional Collaborative Filtering approaches such as memory based, model based approaches which include Matrix Factorization, other hybrid models and a CNN-based model. We tested our model with different level of data sparsity and against different number of similar neighbors. Experimental timeline starts from establishing setup

CRediT authorship contribution statement

Parth Sahu: Visualization, Investigation, Conceptualization, Methodology, Software, Writing - original draft, Writing - review & editing, Validation. S. Raghavan: Conceptualization, Writing - original draft, Writing - review & editing, Validation. K. Chandrasekaran: Supervision, Writing - review & editing.

Declaration of Competing Interest

The authors certify that they have NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus;membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed

Acknowledgements

aaaa

Parth Sahu is currently pursuing his Master of Technology in Computer Science at the National Institute of Technology Karnataka Surathkal, Mangalore, India. He completed his B.E. in Information Technology from Shri Govindram Seksariya Institute of Technology, Indore, M.P., India. His current research interests are Machine Learning, Blockchain, Cloud Computing, Internet of Things (IoT).

References (57)

  • F. Chen et al.

    User-qos-based web service clustering for qos prediction

  • S. Chen et al.

    A cluster feature based approach for qos prediction in web service recommendation

  • X. Chen et al.

    Regionknn: A scalable hybrid collaborative filtering algorithm for personalized web service recommendation

  • X. Chen et al.

    Web service recommendation via exploiting location and qos information

    IEEE Trans. Parallel Distrib. Syst.

    (2013)
  • Z. Chen et al.

    A user dependent web service qos collaborative prediction approach using neighborhood regularized matrix factorization

  • C. Desrosiers, G. Karypis, A comprehensive survey of neighborhood-based recommendation methods. In F. Ricci, L. Rokach,...
  • P. He et al.

    A hierarchical matrix factorization approach for location-based web service qos prediction

  • J.L. Herlocker, J.A. Konstan, A. Borchers, J. Riedl, An algorithmic framework for performing collaborative filtering,...
  • W.K.Z.Y. Jin, e. a., Y., Neighborhood-aware web service quality prediction using deep learning. J. Wireless Commun....
  • S. Kumar et al.

    Missing qos-values predictions using neural networks for cloud computing environments

  • D.D. Lee, H.S.Seung, Algorithms for non-negative matrix factorization, 2001, pp....
  • X. Luo et al.

    Web service qos prediction based on adaptive dynamic programming using fuzzy neural networks for cloud services

    IEEE Access

    (2015)
  • X. Luo et al.

    Incorporation of efficient second-order solvers into latent factor models for accurate prediction of missing qos data

    IEEE Trans. Cybern.

    (2017)
  • X. Luo et al.

    Generating highly accurate predictions for missing qos data via aggregating nonnegative latent factor models

    IEEE Trans. Neural Networks Learn. Syst.

    (2015)
  • H. Ma, I. King, M.R. Lyu, Effective missing data prediction for collaborative filtering. In Proceedings of the 30th...
  • L. Mei et al.

    An adaptive service selection approach to service composition

  • A. Mnih, R.R. Salakhutdinov, Probabilistic matrix factorization. In Advances in neural information processing systems,...
  • T.D. Pham et al.

    Classification of short time series in early parkinson s disease with deep learning of fuzzy recurrence plots

    IEEE/CAA J. Autom. Sin.

    (2019)
  • Cited by (11)

    • Stable and efficient resource management using deep neural network on cloud computing

      2023, Neurocomputing
      Citation Excerpt :

      The remainder of this paper is organized as follows: related works on resource autoscaling are described in Section 2; the ProHPA scheme proposed in this study is explained in Section 3, and the overall architecture of ProHPA is presented in Section 4; ProHPA is implemented in Section 5, and the performance of ProHPA is evaluated in Section 6; lastly, the research findings on ProHPA are summarized and future research directions are proposed in Section 7. Various studies have been conducted to efficiently manage resources in cloud computing [11–19]. In this study, several universal methods for managing container resources using horizontal, vertical and hybrid autoscaling techniques are described.

    • AutoTrust: A privacy-enhanced trust-based intrusion detection approach for internet of smart things

      2022, Future Generation Computer Systems
      Citation Excerpt :

      The application layer is divided into different sectors, i.e., buildings (commercial or industrial), energy (supply/demand, oil/gas), consumer and homes (infrastructure, convenience, and entertainment), health and life science (care and research), industrial (resource automation, fluid/process, distribution), transportation (trans systems, vehicular, and non-vehicular), retail (stores, hospitality, and specialty), security (surveillance, equipment, tracking, public infrastructure, and emergency services), and information technology (public or enterprise). All the sectors of application layer are location-based, and provide services with respect to the location A requirement of real-time response at the IoT side exhibits the necessity for QoS assurance at the cloud side [13]. IoT requires efficient power consumption of applications [14], and the cloud needs a mechanism that maintains the energy resources efficiently [15].

    • Deep learning based web service recommendation methods: A survey

      2023, Journal of Intelligent and Fuzzy Systems
    View all citing articles on Scopus

    Parth Sahu is currently pursuing his Master of Technology in Computer Science at the National Institute of Technology Karnataka Surathkal, Mangalore, India. He completed his B.E. in Information Technology from Shri Govindram Seksariya Institute of Technology, Indore, M.P., India. His current research interests are Machine Learning, Blockchain, Cloud Computing, Internet of Things (IoT).

    S. Raghavan is currently doing his Doctoral Research at National Institute of Technology Karnataka Surathkal, Mangalore, India. He completed his B. Tech from SASTRA University, Thanjavur, India and M. Tech from National Institute of Technology Karnataka Surathkal, India. His area of interest include Cloud Computing, Distributed Computing, Membrane Computing and Machine Learning.

    K. Chandrasekaran is currently Professor in the Department of Computer Science & Engineering, National Institute of Technology Karnataka, India, having 31 years of experience. He has more than 300 research papers published by various reputed and peer-reviewed International journals, and conferences. He has received best paper awards and best teacher awards. He serves as a member of various reputed professional societies including IEEE (Senior Member), ACM (Senior Member), CSI (Life Member), ISTE (Life Member) and Association of British Scholars (ABS). He is also a member in IEEE Computer Society’s Cloud Computing STC. He is in the Editorial Team of IEEE Transactions on Cloud Computing, one of the recent and reputed journals of IEEE publication. He has coordinated many sponsored projects, and, some consultancy projects. He has organized numerous events such as International conferences, International Symposium, workshops and several academic short term programs at NITK. He was a visiting fellow at LMU Leeds, UK in 1995, Visiting Professor at AIT, Bangkok in 2007, and Visitor at UF, USA in 2008 and a Visitor at Univ. of Melbourne, CLOUDS LAB in 2012. He had also worked as Visiting (Professor) at DoMS, IIT Madras during Feb-Dec. 2010. His areas of interest - research include: Computer Networks and Distributed Computing, Federated Cloud Computing, Big Data Management, Internet of Things, Cyber Security, Enterprise Computing & Information Systems Management, and Knowledge Management.

    View full text