Research abstract for semantic anomaly detection in dynamic data feeds with incomplete specifications

Everyday software must be dependable enough for its intended use. Because this software is not usually mission-critical, it may be cost-effective to detect improper behavior and notify the user or take remedial action. Detecting improper behavior requires a model of proper behavior. Unfortunately, specifications of everyday software are often incomplete and imprecise. The situation is exacerbated when the software incorporates third-party elements such as commercial-off-the-shelf software components, databases, or dynamic data feeds from online data sources. We want to make the use of dynamic data feeds more dependable. We are specifically interested in semantic problems with these feeds-cases in which the data feed is responsive, it delivers well-formed results, but the results are inconsistent, out of range, incorrect, or otherwise unreasonable. We focus on a particular facet of dependability: availability or readiness for usage, and change the fault model from the traditional "fail-silent" (crash failures) to "semantic". We investigate anomaly detection as a step towards increasing the semantic availability of dynamic data feeds.