Robust Representation Learning of Biomedical Names

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3275–3285 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 3275 Robust Representation Learning of Biomedical Names Minh C. Phan Aixin Sun Yi Tay Nanyang Technological University, Singapore phan0050@e.ntu.edu.sg; axsun@ntu.edu.sg; ytay017@e.ntu.edu.sg Abstract Biomedical concepts are often mentioned in medical documents under different name vari- ations (synonyms). This mismatch between surface forms is problematic, resulting in dif- ficulties pertaining to learning effective rep- resentations. Consequently, this has tremen- dous implications such as rendering down- stream applications inefficacious and/or poten- tially unreliable. This paper proposes a new framework for learning robust representations of biomedical names and terms. The idea behind our approach is to consider and en- code contextual meaning, conceptual mean- ing, and the similarity between synonyms dur- ing the representation learning process. Via extensive experiments, we show that our pro- posed method outperforms other baselines on a battery of retrieval, similarity and relatedness benchmarks. Moreover, our proposed method is also able to compute meaningful representa- tions for unseen names, resulting in high prac- tical utility in real-world applications. 1 Introduction Representation learning of words (Mikolov et al., 2013; Pennington et al., 2014), and/or sen- tences (Kiros et al., 2015; Hill et al., 2016; Lo- geswaran and Lee, 2018) forms the bedrock of many modern NLP applications. These tech- niques, largely relying on context information, have a huge impact on downstream applications. To this end, learning effective and useful represen- tations has been a highly fruitful area of research. Biomedical names1, however, are different from standard words and sentences. These names have both contextual and conceptual meanings. Con- textual meaning reflects the contexts where the names appear, and it is specifically granted to each 1Biomedical names refer to surface forms that represent biomedical concepts. They can be official names in biomedi- cal vocabularies or unofficial names mentioned in text. Concept (CUI) and their names Source C0343047: leiner’s disease, complement component 5 deficiency, c5d, complement 5 dysfunction, infantile seborrheic der- matitis, erythroderma desquamativum. UMLS C0154832: coats’ disease, abnormal reti- nal vascular development, unilateral retinal telangiectasis, coats telangiectasis NCBI- Disease C0019168: hepatitis b virus surface anti- gen, hepatitis-b surface antigen, hbs ag, hbsag, hepatitis b surface antigen BC5CDR- Chemical Table 1: Example of biomedical concepts and their names taken from one vocabulary (UMLS (Li et al., 2016)) and two annotated datasets (NCBI- Disease (Do˘gan et al., 2014) and BC5CDR-Chemical- Chemical (Li et al., 2016)). The concepts are listed by concept unique identifiers (CUI) defined in UMLS. name. Names of a broad and popular concept of- ten have slightly different contextual meanings. On the other hand, conceptual meaning maps to the definitions/contexts of the names’ associated concepts, i.e., CUIs as shown in Table 1. As such, names of the same concepts share the common conceptual meanings, although they can own dif- ferent contextual information. As illustrated in Table 1, biomedical concepts appear in the text under various names. Repre- sentations of the names are also expected to …

Content truncated to 500 words