Wenping Zhang and Raymond Y.K. Lau Department of Information Systems
City University of Hong Kong Hong Kong SAR
E-mail: {wzhang23, raylau}@cityu.edu.hk
Abstract: With the rapid growth of Web of Things (WoT) and corresponding Web services, there is a pressing need to develop an effective computational method for services discovery and recommendation. Despite personalized services discovery methods have been studied before, few attempts have been made to explore ontological user profiling and probabilistic language modeling approach for Web service contextualization and ranking. This paper makes a novel contribution in terms of developing an ontology-based user profiling method to improve the Web service discovery and recommendations. In particular, a novel probabilistic language model is developed to conduct Web service contextualization and ranking. Our preliminary experimental results reveal that the proposed service personalization approach outperforms a classical baseline method.
Keywords: Services Discovery, Ontological User Profiling, Language Modeling.
1. INTRODUCTION
The fundamental problems of Web services discovery are about the representations of the semantics pertaining to service queries and resources, and the prediction of the relevance of the target resources (services) with respect to a query [17]. With the rapid growth of WoT [5] and the Semantic Web [4], personalized services discovery and recommendation has become a hot research topic [8]. For the paradigm of the semantic Web, ontology has been playing a key role in formal knowledge representation to facilitate human and computer interactions [2,4].
Ontology refers to a formal specification of conceptualization; it may take the simple form of a taxonomy of concepts (i.e., light-weight ontology), or the more comprehensive representation of comprising a taxonomy as well as the axioms and constraints which characterize some prominent features of the real-world (i.e., heavy-weight ontology) [10]. Domain ontology is one kind of ontology which is used to represent the knowledge for a particular type of application domain, and it can be expressed by using formal semantic markup languages such as OWL [11].
Although ontology has been playing a key role in Semantic Web, leveraging ontology to enhance Web services personalization is a relative new topic. This paper illustrates a novel design and development of an ontology-based personalized service discovery and selection model. In particular, ontological user profiling is applied to capture users’ possibly changing service requirements, and probabilistic language modeling is exploited to develop an effective mechanism for service query contextualization based on both current and past search service invocation history. As user queries are usually short e.g., around 2 words long on average [7],
112
query personalization and contextualization is essential for effective Web service discovery. Our preliminary experiments have shown that the proposed ontological user profiling and probabilistic language modeling methods are promising.
2. SYSTEM ARCHITECTURE
The system architecture of the Ontology-based Web Service Discovery and Recommendation System (OWSDR) is depicted in Figure 1. Users interact with their intelligent client stubs that will in-turn look up the most relevant Web services from the external service registries such as UDDI. The Query Processing and Logging module of OWSDR is responsible for accepting users’
queries and managing the query and service invocation histories. After a query is accepted to OWSDR, the Query Contextualization module will refine the original query by referring to the specific user profile. Probabilistic language model is then applied to infer the most relevant service context for the query according to the user’s ontological profile.
Figure 1. General System Architecture of OWSDR
After query contextualization, the refined query is sent to the external service registries for Web services selection. Potentially matching services are then returned to the Service Recommendation module of OWSDR for service ranking with respect to the contextualized query. The Service Invocation module selects the most relevant service in the ranked list and communicates with the external service provider. On the other hand, the User Profiling module is responsible for user profile creation and revision. An ontological user profile is first instantiated based on the ODP taxonomy. With reference to the query and service invocation history, the ontological user profile is updated to reflect the user’s most recent interests. Our prototype system was developed using Java (J2SE v 1.4.2), Java Server Pages (JSP) 2.1, and Servlet 2.5 and operated under Apache Tomcat 6.0.
3. SYSTEM EVALUATION
113 service descriptions extracted from xmethods.net. As a result, our dataset consists of 550 service descriptions (documents) for 10 ODP categories. For instance, for the weather category, the corresponding retrieved web service description is “Provide weather forecast for a given place”.
The performance evaluation metric used was top-n precision, where n=50 was applied in our experiment.
Figure 2. Web Service Descriptions at xmethods.net
For each service domain (ODP category), we applied five relevant service descriptions to train the user profiling module of the OWSDR system and establish the corresponding user profile.
The keywords of the category such as “exchange” was taken as the raw query and passed to the contextualization module for expansion. The service recommendation module was then invoked to rank the remaining service descriptions. The top fifty service descriptions ranked by the ranking module was used to evaluate the performance (e.g., precision and recall) of the OWSDR system. A baseline system was also developed based on the vector space model and the cosine similarity measure [18]:
114
of the ith term in the service description d respectively. The set {1, 2,...., } tT
t t
T represents the set of terms (i.e., the vocabulary) of our entire dataset.
For the baseline system, user profiling was conducted based on the vector space model and query contextualization was performed based on each user vector. In addition, profile revision was based on the classical Rocchio learning method [18] and the baseline system applied the cosine similarity function to rank the service descriptions with respect to each service category. Our preliminary experimental results are reported in Table 1. The average improvement in terms of top-n precision over the ten service categories is +21.77%.
Category Baseline OWSDR Improvement
book 0.58 0.68 +17.24%
health 0.52 0.62 +19.23%
business 0.54 0.66 +22.22%
education 0.48 0.58 +20.83%
weather 0.56 0.64 +14.29%
exchange 0.44 0.56 +27.27%
football 0.46 0.58 +26.09%
game 0.48 0.54 +12.50%
stock 0.54 0.60 +11.11%
software 0.62 0.72 +16.13%
Average 0.51 0.62 +21.77%
Table 1. Comparative Performance of OWSDR
The main reason of such a significant performance improvement brought by OWSDR is due to the dynamic user profiling and service query contextualization processes. In particular, the semantics of service queries and service descriptions are taken into account while Web services are evaluated. Moreover, with a personalized and contextualized service query, more accurate service matching is performed by the system. The end result is that the overall effectiveness of Web service discovery and recommendation is improved.
4. CONCLUSIONS AND FUTURE WORK
With the rapid proliferation of WoT and semantic Web services, intelligent and personalized services discovery and recommendation methods are desirable for practical deployment of Web services in real-world application contexts. Even though personalized services discovery has been examined by researchers before, few attempts have been made to exploit ontological user profiling and probabilistic language modeling for service contextualization and service ranking.
This paper makes a novel theoretical contribution in the sense that domain ontology and probabilistic language modeling have been applied to design an effective computational
115
methodology to enhance service discovery and recommendation in the real-world contexts. Our preliminary experimental results show that the proposed method is promising and it outperforms a classical baseline method by 21.77% in terms of top-n precision. Our future work will further examine and evaluate the dynamic user profiling method based on incremental users’ relevance feedback. A large-scale usability study against the proposed system will also be conducted.
ACKNOWLEDGMENT
The work reported in this paper has been funded in part by the Strategic Research Grants of City University of Hong Kong (Project No. 7004120, 7003002, 7008138) and the Shenzhen Municipal Science and Technology R&D Funding - Basic Research Program (Project No.
JCYJ20130401145617281).
References
[1] Abbar, S., et al. A Personalized Access Model: Concepts and Services for Content Delivery Platforms. Proceedings of the 10th International Conference on Information Integration and Web-based Applications & Services. 2008.
[2] Berners-Lee, T., Hendler, J., and Lassila, O. The Semantic Web. Scientific American, 284(5):34-43, 2001.
[3] Bolchini, C., Quintarelli, E., and Rossato, R. Relational Data Tailoring Through View Composition. Proceedings of the 26th International Conference on Conceptual Modeling, 2007.
[4] Burstein, M. and McDermott, D. Ontology Translation for Interoperability among Semantic Web Services. AI Magazine, 26(1): 71-82, 2005.
[5] He, J., Zhang, Y., Huang, G., and Cao, J. A smart Web service based on the context of things.
ACM Transactions on Internet Technology, 11(3):Article 13, 2012.
[6] Hsieh, P.-H. and Yuan, S.-T. Dynamic Semantic Location Modeling in Mobile Enterprise Applications.Proceedings of the 5th International Conference on Electronic Commerce.
2003.
[7] Jansen, B. and Spink, A. How are we searching the World Wide Web? A comparison of nine search engine transaction logs. Information Processing and Management, 42(1):248-263, 2006.
[8] Jiang, X. and Tan, A-H. Learning and inferencing in user ontology for personalized semantic web services. Proceedings of the 15th international conference on World Wide Web, pp. 1067-1068, 2006.
[9] Keahey, K. and Freeman, T. Contextualization: Providing One-Click Virtual Clusters.
Proceedings of the Fourth IEEE International Conference on eScience. 2008.
[10] Lau, R.Y.K., Song, D., Li, Y., Cheung, C.H., Hao, J.X. Towards A Fuzzy Domain Ontology Extraction Method for Adaptive e-Learning. IEEE Transactions on Knowledge and Data Engineering, 21(6): 800-813, 2009.
[11] Lau, R.Y.K., Lai, C.L., and Li, Y. Fuzzy Ontology Mining and Semantic Information Granulation for Effective Information Retrieval Decision Making. International Journal of Computational Intelligence Systems, 4(1):54-65, 2011.
[12] Licker, P.S., Getting Advice From a Computer: Five Models and A Simulator SIGMIS Database, 22(3):1-13, 1991.
[13] Liu, X. and Croft, B. Cluster-based retrieval using language models. Proceedings of the 27th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pp.186-193, 2004.
[14] Milic-Frayling, N., et al. On the Design and Evaluation of Web Augmented Mobile Applications.Proceedings of the 9th International Conference on Human Computer Interaction with Mobile Devices and Services. 2007.
[15] Mrissa, M., et al., A Context-Based Mediation Approach to Compose Semantic Web Services. ACM Transactions on Internet Technology. 8(1):1-23, 2007.
116
[16] Pon-Barry, H., et al. Contextualizing Learning in a Reflective Conversational Tutor.
Proceedings of the IEEE International Conference on Advanced Learning Technologies.
2004.
[17] Rong, W., Liu, K., and Liang, L. Towards Personalized Ranking in Web Service Selection.
Proceedings of the 2008 IEEE International Conference on e-Business Engineering, pp. 165-172, 2008.
[18] Salton, G. and McGill, M. Introduction to Modern Information Retrieval. McGraw-Hill, New York, 1983.