Acronym Expansion: A Domain Independent Approach.

Acronyms are present in usually all documents to express information that is repetitive and well known. But acronyms can be ambiguous because there can be many expansions of the same acronym. In this paper, we propose a general system for acronym expansion that can work on any acronym given some context information it is used in. We present methods for retrieving all the possible expansions of an acronym from Wikipedia and AcronymsFinder.com. We propose to use these expansions to collect the context in which these acronym expansions are used and then score them using a deep learning technique called Doc2Vec. All these things collectively lead to achieving an accuracy of 90.9% in selecting the correct expansion for given acronym on a dataset we scraped from Wikipedia with 707 distinct acronyms and 14,876 disambiguations.