Adapting Phrase-based Machine Translation to Normalise Medical Terms in Social Media Messages

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1675–1680, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Adapting Phrase-based Machine Translation to Normalise Medical Terms in Social Media Messages Nut Limsopatham and Nigel Collier Department of Theoretical and Applied Linguistics University of Cambridge Cambridge, UK {nl347,nhc30}@cam.ac.uk Abstract Previous studies have shown that health reports in social media, such as Dai- lyStrength and Twitter, have potential for monitoring health conditions (e.g. adverse drug reactions, infectious diseases) in par- ticular communities. However, in order for a machine to understand and make in- ferences on these health conditions, the ability to recognise when laymen’s terms refer to a particular medical concept (i.e. text normalisation) is required. To achieve this, we propose to adapt an existing phrase-based machine translation (MT) technique and a vector representation of words to map between a social media phrase and a medical concept. We eval- uate our proposed approach using a col- lection of phrases from tweets related to adverse drug reactions. Our experimen- tal results show that the combination of a phrase-based MT technique and the simi- larity between word vector representations outperforms the baselines that apply only either of them by up to 55%. 1 Introduction Social media, such as DailyStrength1 and Twit- ter2, is a fast growing and potentially rich source of voice of the patient data about experience in terms of benefits and side-effects of drugs and treatments (O’Connor et al., 2014). However, natural language understanding from social me- dia messages is a difficult task because of the lexical and grammatical variability of the lan- guage (Baldwin et al., 2013; O’Connor et al., 2014). Indeed, language understanding by ma- chines requires the ability to recognise when a phrase refers to a particular concept. Given a 1http://www.dailystrength.org/ 2http://twitter.com variable length phrase, an effective system should return a concept with the most similar mean- ing. Table 1 shows examples of mappings be- tween Twitter phrases and medical concepts. For example, a Twitter phrase ‘No way I’m gettin any sleep 2nite’ might be mapped to the med- ical concept ‘Insomnia’ (SNOMED:193462001), when using the SNOMED-CT ontology (Spack- man et al., 1997). The success of the mapping between social media phrases and formal medi- cal concepts would enable an automatic integra- tion between patient experiences and biomedical databases (Limsopatham and Collier, 2015). We refer to this mapping from social media phrases to medical concepts as medical term normalisation, which aims to determine the unique identifier of a medical concept that is mentioned in different forms in a free-text (Morgan et al., 2008). Existing works, e.g. (Elkin et al., 2012; Gob- bel et al., 2014; Wang et al., 2009), mostly fo- cused on identifying medical concepts in medi- cal documents. For example, Gobbel et al. (2014) proposed a na¨ıve Bayesian-based technique to map phrases from clinical notes to medical con- cepts in the SNOMED-CT ontology. Wang et al. (2009) identified medical concepts regarding adverse drug events in electronic medical records. On the other …

Content truncated to 500 words