I am currently working on a research Web service framework 2SAP based on a theoretic model on Web service interfaces to enable two-way full duplex interactions. This framework addresses special issues in telecommunication systems, such as session management and asynchronous messaging control, and has been used to transform various conventional enterprise applications and telecommunication systems into Web service endpoints.

 

Since 2004, I have been working with Wu Chou to promote Avaya’s contributions to ECMA International in the CSTA Web service area. Our contribution, WS-Session, has been approved as an ECMA standard and ETSI Technical Specification. It is submitted to ISO/IEC JTC 1 for adoption under the fast-track procedure.

 

Another on-going project is developing our multimodal human-machine dialogue platform MTIP based on XML and Web service technologies. In the past, we have integrated our platform with several enterprise CRM systems to enhance the self-service capabilities through automated agents. The recent advance in this project is a synchronization component (XM-Flow), part of our three-tier dialogue architecture, which uses SMIL and EMMA to synchronize multimodal interactions.

 

I also have been working on improving our Natural Language Call Routing algorithms with various statistical techniques. Most recent advancement is a Generic Linear Classifier (GLC) framework for MCE and Boosting training that significantly increases the classification accuracy. 

 

 

All publications cited herein are coauthored with colleagues.

 

In Web service area, recent papers are as follows.

Two-way Web Service: from Interface Design to Interface Verification, Proceedings of ICWS 2005, Volume 2, page 525-532, July 2005, Orlando Florida. This paper proposed a theoretic framework for Web service interface derivation and verification based on reversal and generalization transformations to address the issues in two-way Web service development.

 

Web Service Enablement of Communication Services, Proceedings of ICWS 2005, Volume 2, page 393-400, July 2005, Orlando, Florida. This paper proposed two ways of event subscription, sink-source and source-sink, to solve the timing issues in event control.

 

WSIP-Web Service SIP Endpoint for Converged Multimedia/Multimodal Communication over IP. This paper illustrates how you can combine Web service and SIP into a component with dual roles to bridge the chasm between two worlds.

 

In dialogue platform research, some representative papers are:

Extensible Three-tier XML Dialogue System Architecture for Multimodal Interaction and Automated Agent Services, Proceedings of 9^th IASTED Conference on Internet and Multimedia Systems, and Applications, August 2005, shows a novel XML based three-tier dialogue platform architecture and its application in developing automated agents for call center systems.

 

A Distributed Multimodel Dialogue System Architecture Based on Dialogue System and Web Convergence, the 7^th International Conference on Spoken Language Processing, Vol 4, page 2253-2256, Denver, Colorado, September 16-20, 2002, proposed a distributed architecture that supports cooperative multimodal dialogue flows using hybrid constructs.

 

For NLP and text classification, here are some selected papers.

Minimum Classification Error (MCE) Framework for Generalized Linear Classifier and Its Application to Text Categorization/Retrieval, Proceedings of IEEE International Conference on Machine Learning and Applications (ICMLA’04), page 26—33, December, 2004, proposes a framework that unifies a family of linear classifiers, including Naïve Bayse, LSI, and TF-IDF classifiers for MCE training. It also illustrates how this framework can facilitate combination of MCE with AdaBoost to further improve classification accuracy by escaping local minimum.

 

An Information Theoretic Approach for Using Word Cluster Information in Natural Language Call Routing, page 2829-2832, Eurospeech 2003, Geneva, Switzerland, 9/1-9/4, 2003, proposed several methods to inject word class information into classifier training with an Information-Gain based feature selection algorithm.