Service Discovery - Dynamic Matchmaking among Web-based Heterogeneous Software Agents by Aditya Kalyanpur

Proposal: My aim is to analyze current state-of-the-art web service description and discovery mechanisms such as UDDI and DAML-S Matchmaker, determine their limitations, and propose and implement a solution that overcomes them.

Systems Analyzed: UDDI - Universal Description, Discovery and Integration is the first cross-industry effort driven by platform providers, software developers, marketplace operators and business leaders that comprehensively addresses the problems limiting the growth of service-centric computing. Its aim is to define an open framework to help heteregenous software agents 1) discover each other, 2) define how they interact over the internet and 3) share information in a global registry

Main Limitation: No semantics included. Search is keyword-based. Thus if a user is searching for 'stock quotes', and a service provides 'financial news', no match occurs.

DAML-S Matchmaker: A Semantic Matchmaker that attempts to overcome the limitations of UDDI by ensuring that web services are described semantically using DAML-S, thereby allowing description logic reasoners and inference engines to work with the dataset, improving the efficieny of the search. It defines a language called LARKS (Language for Advertising and Request for Knowledge Sharing) and uses a matchmaking process that performs both syntactic and semantic matching, and in addition allows the specification of concepts (local ontologies) via ITL, a concept language. The matching process uses 5 filters: context matching, profile comparison, similarity matching, signature matching and constraint matching.

Main Limitations:

(related to LARKS)

1. service description requires adherence to ontology terms which leads to limited expressivity..for e.g. if a car-rental service provides 'vehicle information' and wants to be as general as possible (say, by specifying 'vehicle' is a term present in 50 different ontologies), describing this service using LARKS is impossible

2. if a service consists of various other web services, expressing its description in LARKS is not easy

(related to the match-making process)

1. cannot perform planning..i.e. finding a set of services to perform a task when a single service is not possible

2. matches inputs/outputs in request and advertisement, and both i/o shud adhere to the same ontology which is a serious limitation...(maybe ontomerge or any other tool can be used to broaden the search by determining equivalent ontologies)

3. structure of services not taken into acct..e.g. if the user wants to find a service that doesnt require credit card authentication, its not possible currently (maybe additional filter needed)

4. no history feature present - (maybe the system can maintain a history of service compositions and use it to aid future searches)

5. no multi-lingual support