Mitigating operator-induced unavailability by matching imprecise queries

In addition to equipment faults, human error is now recognized as a major cause of computer system unavailability. This paper considers one aspect of human error in critical situations-the inability of operators to retrieve and understand documentation needed for system diagnosis and repair. When technical information vital to recovery is missing, difficult to locate or inaccessible, downtime is lengthened, costs rise, and productivity falls. Finding the right information at the right time is complicated by the ambiguities of natural-language queries when seeking documentation or maintenance information. While the human information processor has the means for resolving ambiguities in language, computers do not. Hence, a key issue in downtime problem resolution is imprecision in human vocabulary. The vocabulary problem can be addressed through statistical mapping of user queries into databases of frequently-asked questions. This technique has been validated empirically, and shown to be effective in achieving correct mappings in 99% of cases tested; it is substantially better than keyword mapping, especially as syntactic and lexical differences grow. When information seeking is accelerated by this technique, downtime can be reduced