IBM wants to use Twitter to help troubleshoot your data center

Social data is finding use in most of everything from selling pet supplies to high-frequency trading these days, but IBM thinks the list is missing a key item: fixing infrastructure. Its plan for changing that is disclosed in a fresh patent application that reveals an analytic approach to turning Internet whining into useful technical feedback for system administrators.

The logic behind the invention is sound: if something goes wrong with a service, or even gives the remote impression that something has gone wrong, people will complain. And in this day and age they’re as likely as not to take their protests to the social sphere, where that valuable (or at least plentiful) real-time input becomes accessible for analysis.

The engine detailed in IBM’s filing first parses the messages into a machine-readable form that can be cross-referenced with logs about the workload in question. If someone is complaining about latency, the built-in machine learning algorithms will look for abnormal network activity, while performance issues would similarly result in a query to the relevant part of the environment.

Once a correlation has been found, the engine will set a confidence score for the relationship and repeat the process for every data point, including the relationships themselves, until coming up with the most complete possible depiction of the problem. It will then go back to the original social interactions and search for certain pre-defined phrases to assess the perceived severity.

The guidance mentioned in that last box can consist of documentation about the affected process or processes, technical notes or even articles pulled off the web that IBM’s engine thinks could help solve the issue. All of that is wrapped into an alert that also contains the relevant logs and an estimate of how long it’ll take you to solve the problem.

IBM sees the system finding use mainly in helping administrators detect emerging issues affecting too few users to register otherwise and take remediative action before the problem spreads. But the same reach that makes it possible to detect even small sentiment ripples in the social sphere also creates an elevated risk of false positives that will have to be addressed before those concept drawings can turn into reality.