Delta learning system for using expert advice to revise diagnostic expert system fault hierarchies

摘要

A delta learning system takes as an initial fault hierarchy (KB. sub.0) and a set of annotated session transcripts, and is given a specified set of revision operators where each operator within a group maps a fault hierarchy (KB) to a slightly different or revised fault hierarchy (&thgr; .sub.i (KB)). The revised fault hierarchy (&thgr;.sub. i (KB)) is called a neighbor of the fault hierarchy (KB), and a set of all neighbors (N(KB)) is considered the fault hierarchy neighborhood. The system uses the revision operators to hill climb from the initial fault hierarchy (KB.sub. 0), through successive hierarchies (KB.sub.1 . . . KB. sub.m), with successively higher empirical accuracies over the annotated session transcripts. The final hierarchy (KB.sub.m), is a local optimum in the space defined by the revision operators. At each stage, to go from a fault hierarchy (KB.sub.i) to its neighbor (KN.sub.i+ 1), the accuracy of the fault hierarchy (KB.sub.i) is evaluated over the annotated session transcripts, and the accuracy of each fault hierarchy (KB*) belonging to the set of all neighbors (N(KB.sub.i)) is also evaluated. If any fault hierarchy (KB*) is found to be more accurate than the fault hierarchy (KB. sub.i), then this fault hierarchy (KB*) becomes the new standard labeled KB.sub.i+1.