Domain-specific Noisy Query Correction using Linguistic Network Community Detection

Noisy queries pose an important challenge for retrieving relevant search results. The importance for query correction increases with increasing use of hand-held devices and technologies such as SMS, tweets to search and access information. The task is further complicated for domain-specific search engines as the amount of query logs may be significantly smaller than general purpose search engines. In this paper, we propose to use the community detection technique from social network analysis for spelling correction of a set of noisy queries such as SMS messages. We focus on the task of identifying relevant questions from a set of Frequently Asked Questions (FAQ) for different domains for a set of incoming noisy queries. Experimental validation shows that the proposed CD-Speller method performs significantly better than Hunspell, the popular and industry-strength spelling correction tool.