Toward a Short Text Classification Framework Based on Background Knowledge Discovery

The ubiquitous, diverse and growing impact of digital living creates a massive amount of short text a search query, a twit or a caption. Short text frequently presents itself as an arbitrary combination of semantically unconnected words. Using machine learning to classify the corpora of such texts is a challenging task. A large body of research exists in this area, but in this paper we will focus on Background Knowledge (BK) and its role in machine learning for shorttext and non-topical classification. More specifically, we present an effort to create a short text classification framework based on Background Knowledge. We propose novel Information Retrieval techniques to construct BK and demonstrate the advantages of Automatic Query Expansion (AQE) vs. basic search. We discuss other results of this research and its implications on the advancement of short text classification.