Medical Concept Mention Identification in Social Media Posts Using a Small Number of Sample References

Proceedings of Recent Advances in Natural Language Processing, pages 777–784 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_084 777 Medical Concept Mention Identification in Social Media Posts using a Small Number of Sample References Vasudevan Nedumpozhimana1, Sneha Rautmare2, Meegan Gower2, Maja Popovi´c3, Nishtha Jain2,5, Patricia Buffini3, John Kelleher1,4 ADAPT Centre 1Technological University Dublin, 2Trinity College Dublin, 3Dublin City University, 4Maynooth University, (5now at Spoke.ai, Berlin, Germany) name.surname@adaptcentre.ie Abstract Identification of mentions of medical concepts in social media text can provide useful informa- tion for caseload prediction of diseases like Covid-19 and Measles. We propose a sim- ple model for the automatic identification of the medical concept mentions in the social media text. We validate the effectiveness of the proposed model on Twitter, Reddit, and News/Media datasets. 1 Introduction Caseload information of diseases like Covid-19 and Measles are likely reflected in social media posts in the form of mentions of relevant medical concepts. For example, increase in the mentions of medical concepts like fever, headache, cough, loss of smell etc. in social media text is a potential indication of increasing covid caseloads. Therefore models which identify the mentions of medical concepts from social media text can provide useful features for the caseload prediction of such diseases. State-of-the-art natural language processing tech- niques mostly rely on huge pre-trained language models and such models can be utilized for iden- tifying the mentions of medical concepts in social media texts. In this work, we propose a simple and effective model to automatically identify the presence of 24 selected medical concepts in social media text by using a small number of reference texts and a pre-trained language model. 2 Related Works The basis for the medical concept mention identi- fication method carried out in this work is the re- search on medical concept normalization. In the lit- erature, the medical concept normalization problem is addressed by using different approaches. Tradi- tionally lexicon-based string-matching approaches and rule-based approaches are used for medical concept normalization. For example, Aronson and Lang (2010) used a knowledge-intensive approach for concept normalization which is based on sym- bolic language processing. Leaman et al. (2013) approached the medical concept normalization problem by learning the similarity between mentions and concept names. Limsopatham and Collier (2015) approached this medical concept normalization as a phrase-based machine translation problem and they translated social media phrases into formal medical concepts. In another approach, Limsopatham and Collier (2016) used simple deep-learning-based models like CNN, and RNN with pre-trained LM and im- proved the performance of the medical concept nor- malization. Lee et al. (2017) further improved this performance by refining the dataset and leveraging the neural embeddings of health-related text. Bornet et al. (2023) showed that language mod- els can learn the semantics of medical concepts. They found that subword information is crucial for learning medical concept representation and global word co-occurance information is more use- ful for downstream tasks using these representa- tions. This suggests the suitability of language models that have both subword information and global co-occurrence information for …

Content truncated to 500 words