Code mixing and code borrowing are the two important linguistic phenomena seen among the bilingual and multilingual speakers. The present scenario demands highly efficient methods to distinguish code borrowing from code mixing to quickly process the multilingual queries. As part of the Data Challenge organized by CODS 2017, we have to rank different words according to their borrowing likeliness. In this paper, a new relevance based metric is proposed by applying statistics based approach. By performing various experiments on the social media data corpus containing more than 2.5 lakh tweets, the effectiveness of the proposed relevance metric was studied.