The Research On Some Key Techniques Of Multi-agent System On The Web

Along with the fast and fierce evolution of the Internet, the need for open applications become more and more great, because Internet itself is a distributed calculating environment of high degree openness. How to develop this kind of software that can make the best of the Internet is always a great challenge in computer science since 1990s. Multi-agent system (MAS) has great potential to be the mainstream distributed computing mode on Internet. But, to make this to be reality, on one hand, we must continually develop the key technologies of MAS, such as communicating, Ontology, content language, information discovery, coordinate, trust management and so on. On the other hand, we must combine the MAS technologies and the Web technology to promote the application of MAS on the Internet. To facilitate the application of MAS on Internet and Web, this thesis addresses several key technical problems, which covers agent capability matching, trust management, integration of agent and XML document and issues concerning transparent migration of MAS from P2P to B/S. Major contributions of this thesis include:1) advances a new algorithm A^H to compute all possible paths, the shortest (longest) path and the (immediate) ancestor or descendent relation between two nodes on a simple directed acyclic network. Firstly, A^H save some path information for each node as add-ons and build hash index on these information, then A^H use these information to reduce time complexity of all computing above.2) enhances the efficiency of Ontology based agent capability matching by using A^H algorithm. Domain Ontology can be modeled a simple directed acyclic network, Subsumption relation between two concepts can be computed by the (immediate) ancestor or descendent algorithm in A^H, Association degree between two concepts can be computed by the shortest path algorithm in A^H. Comparing to the reasoning mechanism in concept language and description logic, A^H improves the efficiency of semantic matching between agent capabilities.3) improves the extended path expressions based XML querying theory framework. Agents on the Web need an ability to access XML documents. One important way to access an XML document is querying. To make full use of the orderliness of XML documents while querying, this thesis improves the extended path expressions based XML querying theory Framework. On one hand, extended path expressions are made simpler and more flexible by introducing wildcard into them, on the other hand, the time complexity of the algorithm is also reduced by putting forward a new kind of automata, namely automata with a stop set.4) advances a role delegation supported logic D1LP-R based on delegation logic(D1LP), and utilizes D1LP-R to realize trust management in MAS. In D1LP-R, a role refers to some access rights, different roles can form a treestructure, access rights corresponding to a node in the role tree contain all access rights corresponding to all descend nodes. If an agent has the right to delegate a role, then it has the right to delegate all sub-roles. By this method, this thesis realizes big granularity access right delegating and partial access right delegating.5) brings up and realizes a method which can transparently migrate P2P MAS to B/S MAS. This method realizes this by establishing a runtime environment for all agents in MAS on a server of high performance. When a user login in the MAS through a browser, the runtime environment will instantiate the responding agent and then migrate its graphic user interface to the browser as a applet by using improved RAWT technology. By this method, a user can interact with the responding agent through the browser.