1. UML based expert system generation Using Enterprise Architect to model and generate a web-based expert system

2. Aims of research  Using Unified Modelling Language (UML) to model expert systems (ES)  Developing a system that can consume and analyze that model to generate an ES  Using the web as a deployment platform

3. Reasons for research  ES need maintaining and scaling  ES model should remain accurate  As the system grows, it’s model should still be understandable  ES benefit from portability

4. Technologies used  Enterprise Architect  UML model development  XML exporting of model  Visual Studio 2010  Shell development using ASP.NET and C#  Web based user interface

5. Similar research  Dunstan’s ES generator  Uses XML to output web-based ES related to university courses  XML file contains course information  Prolog interpreter processes the data and outputs web pages  WITS – web-based training ES for SME’s  Uses JavaScript to provide functionality  Easy to use interface

6. What does this research add?  XML data generation methods  consistency and validity of ES model  accessible web-based ES not reliant on client side scripting  mechanism for both developers and users to access the system

7. ES Modelling  Relevant UML elements  Activity Diagrams  Activities, objects, decisions, initials and finals  Edges used to link elements with optional guards  Sample ES  Housing Benefit  Income, dependants, housing need define level of housing benefit

8. KBS and ES Shell components  Knowledge base  The uml model  Development environment  EA  Inference engine .net shell  User interface  Web page  Explanation facilities  Web page

9. Mapping UML to ES  Critical factors are determined through activity diagrams  Decisions within the diagrams become rules – for example ‘fixed abode’fixed abode  Forward chaining is used to test all decisions against current knowledge base – leading to a goal state.  Uses a blind search

10. Model development  A Class model defines the system (HousingBenefit)  Class variables define critical factors  Nested composite activity diagrams define how critical factors are determined

11. Activity Diagrams  Instances of the class are added to diagrams  Run state of a variable in that instance is set accordingly  Ensures consistency/ integrity of data  Where variables must only hold specific values, enumerations are used  (demo)demo

12. ES Generator Classes  ExpertSystem class  Dictionary containing all the critical factors  Source XML data  Processes the top level activities and outputs the results  CriticalFactor class  Processes the composite activities  Determines the result of each critical factor  Nodes  Process individual elements of an activity diagram

13. Processing  Loops  Top level loop successively finds all top level activities starting from the first initial node  Sub loop processes discrete elements on each composite activity diagram  (es generator activities)es generator activities

14. Determining critical factors  Object node finds the variable  ‘extendedProperties’ subnode of the current node runstate="@VAR;Variable=HousingNeed;Value=true;Op==;@ENDVAR;“  Split on ‘;’  Split on ‘=‘  Get the variable name and value  Set the parent _criticalFactorResult to the value

15. Handling decisions  Determine the type of decision  Critical factor  Direct input  If response already exists, get it, find the corresponding guard then find the target node and continue  If not, halt programme execution, output the question and wait for a response

16. Output  Once all critical factors have been determined, the goal state is determined  Then results are output  This includes output of determining factors  Thereby providing simple explanation

17. Limitations  Only simple input can be handled  Inputs are not validated  Not like a traditional ES shell  Developers must have knowledge of UML  Changes to the system must take place at the correct location  UML based XML data is hard-coded

18. Advantages  ES model can scale whilst still being logically organised/ understandable  The model is a perfect representation of the implementation  Being web-based it is portable  The shell allows any ES to be developed and deployed on the web  Provided the developer knows UML and has Enterprise Architect.