Classifying Out-of-vocabulary Terms in a Domain-Specific Social Media Corpus

Classifying Out-of-vocabulary Terms in a Domain-Specific Social Media Corpus SoHyun Park,♣ Afsaneh Fazly,♦ Annie Lee,♦ Brandon Seibel,♦ Wenjie Zi♦ and Paul Cook♣ ♣ Faculty of Computer Science, University of New Brunswick ♦ VerticalScope, Toronto Canada sohyun.park@unb.ca, {afazly,alee,bseibel,wzi}@verticalscope.com, paul.cook@unb.ca Abstract In this paper we consider the problem of out-of-vocabulary term classification in web forum text from the automotive domain. We develop a set of nine domain- and application-specific categories for out-of-vocabulary terms. We then propose a supervised approach to classify out-of-vocabulary terms according to these categories, drawing on features based on word embeddings, and linguistic knowledge of common properties of out-of-vocabulary terms. We show that the features based on word embeddings are particularly informative for this task. The categories that we predict could serve as a preliminary, automatically-generated source of lexical knowledge about out-of-vocabulary terms. Furthermore, we show that this approach can be adapted to give a semi-automated method for identifying out-of-vocabulary terms of a particular category, automotive named entities, that is of particular interest to us. Keywords: Lexical acquisition, social media text, out-of-vocabulary words 1. Domain-specific OOV Classification Out-of-vocabulary terms are more common in social me- dia text than more-conventional text types (Baldwin et al., 2013). Moreover, many domain-specific technical terms are not included in general-purpose dictionaries and lexi- cal resources. Domain-specific social media corpora are therefore particularly rife with out-of-vocabulary terms. Many natural language processing (NLP) systems for tasks including sentiment analysis and question answering rely on lexical knowledge. In the case that a text being pro- cessed contains out-of-vocabulary terms, system perfor- mance suffers because lexical knowledge is not available for these words. Much research in NLP has therefore fo- cused on lexical acquisition — automatically learning syn- tactic or semantic properties of words from corpora (Hearst, 1992; Lin, 1998; Turney and Littman, 2003, for example). In this work we focus specifically on out-of-vocabulary (OOV) terms in web forum text from the automotive do- main. The focus on the automotive domain is motivated by the business interests of VerticalScope, Inc. (the indus- trial collaborator in this research) in being able to more- intelligently analyze this text. VerticalScope, Inc. is a Canadian company that owns and operates one of the most highly visited automotive networks of online forums. The goal of this research is to automatically classify OOVs as one of a predefined set of domain- and application- specific categories — e.g., automotive named entity (NE), slang, spelling error, foreign term. This automatically-inferred coarse-grained lexical knowledge will later be used to build vocabularies focused on, or excluding, particular types of expressions in an effort to improve topic models (Blei et al., 2003, for example) of automotive web forum text to better understand its contents. This knowledge will also be leveraged in an effort to improve downstream NLP tasks, such as named entity recognition, for this specialized text type. 2. OOVs in Automotive Social Media For this study, we focused on a random sample of 665 al- phanumeric OOVs that 1.) have frequency greater than 1000 in an automotive web …

Content truncated to 500 words