Network Problem Diagnostics using Typographic Error Correction

Detecting and correcting network and service availability issues is an essential part of the network administrator's daily duty. One of the causes of errors can be the user herself providing incorrect input. The present work describes a new diagnostic method that detects incorrectly inserted inputs observed in network-related data, e.g., network traffic, log files. The proposed method aims to detect incorrect words in domains, login names, or email addresses. First, we describe how to detect possible incorrect words. For each such detected word, a list of correct candidates is created based on edit distance. Next, the correction method selects the best word by scoring candidates based on the probability of occurrence in the given context. The proposed method was implemented as a prototype and tested on words created using real user activities. The evaluation demonstrates that this approach can substantially reduce the time needed to identify this kind of errors.