NetWeaver™ is a knowledge base construction, maintenance, documentation, and debugging tool written at Penn State University to provide an efficient knowledge engineering environment for cross platform knowledge based systems development. NetWeaver™ is written in C++ and reads and writes platform-independent ASCII scripts that can be read by an embedded inference engine in stand-alone applications. NetWeaver™ renders knowledge dependency networks with a fully editable graphic representation so that the networks appear just like they would on the white board. All the typical Macintosh editing functions are available (i.e., Cut , Copy & Paste) and can be used on single nodes or whole sections of a network. Because the inference engine is built in, networks can be evaluated real-time with nodes changing color to indicate their changing trueness levels. This ability to peer into the logical workings of a knowledge network has really helped domain experts better understand what they "know". On some projects, NetWeaver™ alone is a suitable substitute for stand-alone knowledge based systems.
IcKEE™ (Iconic Knowledge Engineering Environment, Penn State University), also written in C++, performs as the inference engine within NetWeaver™. As an embedded tool, it is available to run under the Macintosh OS, HyperCard, MSDOS, Windows, UNIX and XWindows. IcKEE™ is completely object oriented and not just in the typical programming sense. Each logical operator (node: ANDs, ORs, NOTs, etc.) is handled as an object in memory which knows its connections, holds its own data, performs its own calculations, and communicates via messages to other, connected nodes. IcKEE™ evolved to handle not only standard Boolean connections but also mathematical calculation connections and fuzzy logic. Fuzzy logic is used to handle missing data, to evaluate between competing goals, and for the classical use of determining a variable's membership in a given class. Due to its modular nature, IcKEE™ can be (and has been) extended to embrace other methodologies as the needs arise. An interesting example is the Sequential OR (SOR) node. It was developed as a device to better handle alternative decision scenarios were there is a definite hierarchy of quality level associated with each possible data gathering method. In other words, the SOR node functions as a data route selector; it provides a method for selecting the best choice of paths within the scope of currently available data.
IcKEE™ reads knowledge bases stored in LISP-like ASCII scripts that describe the construction of the knowledge networks. These scripts are editable in a normal text processing environment or, preferably, with NetWeaver™. As the scripts are read, IcKEE™ constructs and links the objects in memory to represent the knowledge base. Knowledge base functions are extremely efficient as all objects are in memory, their routines are all compiled and no on-the-fly interpretation is required. Efficiency is further enhanced by firing only nodes when they have changed. When data changes, the associated variable passes a "dirty" message up through the network. When a dirty message reaches the top of a network, an "evaluate" message is propagated back down through the network. As each node receives the message, it checks itself to see if it is "dirty". If it is "dirty", it fires its evaluation routine and returns its new value to the requesting node once the evaluation routine is done. If the node is not dirty, the message stops propagating and the node returns its current value without performing its evaluation routine. Once the dirty nodes have been resolved adequately (as indicated by the receipt of evaluation results) goal results can be obtained.
NetWeaver™ was developed under cooperative agreements with the USDA Forest Service, Soil Conservation Service, and the USDI National Park Service. NetWeaver™ is currently only available on the Mac, but Windows and XWindows are possible in the near future.
For more information on NetWeaver™ or for a copy of the User's Manual, contact either Michael C. Saunders (the principal investigator ) or Bruce J. Miller (the author of the software):
