DLTSR: A DEEP LEARNING FRAMEWORK FOR RECOMMENDATIONS OF LONG-TAIL WEB SERVICES
ABSTRACT
With the growing popularity of web services, more and more developers are composing multiple services into mashups.Developers show an increasing interest in non-popular services (i.e., long-tail ones), however, there are very scarce studies trying toaddress the long-tail web service recommendation problem. The major challenges for recommending long-tail services accuratelyinclude severe sparsity of historical usage data and unsatisfactory quality of description content. In this paper, we propose to build adeep learning framework to address these challenges and perform accurate long-tail recommendations. To tackle the problem ofunsatisfactory quality of description content, we use stacked denoising autoencoders (SDAE) to perform feature extraction.Additionally, we impose the usage records in hot services as a regularization of the encoding output of SDAE, to provide feedback tocontent extraction. To address the sparsity of historical usage data, we learn the patterns of developers’ preference instead of modelingindividual services. Our experimental results on a real-world dataset demonstrate that, with such joint autoencoder based featurerepresentation and content-usage learning framework, the proposed algorithm outperforms the state-of-the-art baselines significantly
PROPOSED SYSTEM:
The major challenges to perform high-performance longtailweb service recommendations include:_ Severe sparsity of historical usage data. The sparsityof historical usage data on the long-tail side stronglylimits the applicability of traditional latent factorbased collaborative filtering methods. In addition,previous works show that tightlycoupling content extraction with the usage data canallow usage data to provide feedback, and guidethe content extraction to achieve better performance.However, considering the severe sparsity of usagerecords in the long-tail side, tightly coupling thelong-tail ratings will be of little help._ Unsatisfactory quality of developer-provided servicedescriptions, including textual descriptions, tags, categoryinformation, protocols and so on. As the usagedata is extremely sparse, we have to rely more onthe description, however, different developers canuse different terminologies, and may put insufficient,ambiguous, or even incorrect descriptive content.Themain contributions of this paper are as follows._ We propose DLTSR, a deep learning framework toaddress a gradually emergent challenge in web serviceeconomy, i.e., the long-tail web service recommendations._ We use SDAE as a foundation. The transferredknowledge from usage in the hot service side, andthe modeling of the developers’ preference are incorporatedtightly with the SDAE part to boost theperformance of long-tail service recommendations._ Experiments on real-world dataset from ProgrammableWebshow that DLTSR gains an improvementof 4.1% on Recall@5, 12.0% on Recall@50, and7.1% on Recall@250, compared with the state-of-theartbaseline (modified) TCDR1.
EXISTING SYSTEM:
As our work touches on a couple of different areas, in thissection, we describe several representative related worksand differentiate them with our approach.6.1 Web Service RecommendationsExisting personalized service recommendation approachescan be divided into two categories, i.e., QoS-based servicesrecommendations and functionality-based services recommendations.The method proposed in this paper belongs tofunctionality-based approaches.6.1.1 QoS-based web service recommendationsFor this kind of algorithms, researchers assume that themashup developers are aware of functionally which categoriesof services should be involved, and focus on the nonfunctionalproperties of services such as reliability, availabilityand response time. A the works, introducedthe non-negative tensor factorization to perform temporalawaremissing QoS prediction with the triadic relationsof user-service-time model. fused the neighborhoodbasedand model-based collaborative filtering approachesto achieve a higher prediction accuracy. proposed aQoS prediction by considering temporal information andemploying the random walk algorithm to overcome thesparsity. unified the modeling of multi-dimensionalQoS data via tensor and applied tensor decomposition topredict missing QoS value.
CONCLUSION AND FUTURE WORK
As long-tail services are playing an increasingly importantrole in web API economy, how to recommend long-tail webservices effectively is becoming a key issue. However, veryscarce work has focused this problem, and traditional webservice recommendation methods perform poorly on thelong-tail side.In this paper, we propose a deep learning frameworkwhich is specifically designed for this problem. To tacklethe problem of unsatisfactory quality of description givenby service developers and mashup queries, we use the deeplearning model SDAE as the basic component to learn robustand effective representations. Moreover, the knowledgefrom usages in the hot service side is transferred and imposedas a regularization on the output of SDAE to guide thelearning of representations. To make the best use of the fewlong-tail historical ratings, a special mechanism is designedto model developers’ preference. Experiments demonstratethat our method gains a significant improvement comparedwith the state-of-the-art baseline methods.In the future, we plan to incorporate more information,such as QoS, user profiles and social connection betweenservices into DLTSR to promote the accuracy of recommendations.We also plan to investigate more sophisticated deeplearning models such as convolutional neural networks orrecurrent neural networks, and activations such as ReLU orPReLU to further boost the performance
REFERENCES
[1] Y. Wei and M. B. Blake, “Service-oriented computing and cloudcomputing: challenges and opportunities,” IEEE Internet Computing,vol. 14, no. 6, p. 72, 2010.
[2] C. Schroth and T. Janner, “Web 2.0 and soa: Converging conceptsenabling the internet of services,” IT professional, vol. 9, no. 3, pp.36–41, 2007.
[3] J. Zhang, W. Tan, J. Alexander, I. Foster, and R. Madduri,“Recommend-as-you-go: a novel approach supporting servicesorientedscientific workflow reuse,” in Services Computing (SCC),2011 IEEE International Conference on. IEEE, 2011, pp. 48–55.
[4] X. Liu and I. Fulia, “Incorporating user, topic, and service relatedlatent factors into web service recommendation,” in Web Services(ICWS), 2015 IEEE International Conference on. IEEE, 2015, pp.185–192.
[5] Y. Zhong, Y. Fan, K. Huang, W. Tan, and J. Zhang, “Time-awareservice recommendation for mashup creation,” IEEE Transactionson Services Computing, vol. 8, no. 3, pp. 356–368, 2015.
[6] S. M. McNee, J. Riedl, and J. A. Konstan, “Being accurate is notenough: how accuracy metrics have hurt recommender systems,”in CHI’06 extended abstracts on Human factors in computing systems.ACM, 2006, pp. 1097–1101.
[7] H. Yin, B. Cui, J. Li, J. Yao, and C. Chen, “Challenging the longtail recommendation,” Proceedings of the VLDB Endowment, vol. 5,no. 9, pp. 896–907, 2012.
[8] X. Zhao, Z. Niu, and W. Chen, “Opinion-based collaborativefiltering to solve popularity bias in recommender systems,” inInternational Conference on Database and Expert Systems Applications.Springer, 2013, pp. 426–433.
[9] Y. Hu, Q. Peng, and X. Hu, “A time-aware and data sparsity tolerantapproach for web service recommendation,” in Web Services(ICWS), 2014 IEEE International Conference on. IEEE, 2014, pp.33–40.
[10] Y. Ma, S. Wang, F. Yang, and R. N. Chang, “Predicting qos valuesvia multi-dimensional qos data for web service recommendations,”in Web Services (ICWS), 2015 IEEE International Conferenceon. IEEE, 2015, pp. 249–256.