Togedemaru at SemEval-2023 Task 8: Causal Medical Claim Identification and Extraction from Social Media Posts

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 913–921 July 13-14, 2023 ©2023 Association for Computational Linguistics Abstract 1 The “Causal Medical Claim Identification and 2 Extraction from Social Media Posts” task at SemEval 3 2023 competition focuses on identifying and validating 4 medical claims in English, by posing two subtasks on 5 causal claim identification and PIO (Population, 6 Intervention, Outcome) frame extraction. In the context 7 of SemEval, we present a method for sentence 8 classification in four categories (claim, experience, 9 experience_based_claim or a question) based on 10 BioBERT model with a MLP layer. The website from 11 which the dataset was gathered, Reddit, is a social news 12 and content discussion site. The evaluation results show 13 the effectiveness of the solution of this study (83.68%). 14 1 Introduction 15 One of the most brainstorming tasks facing natural 16 language processing (NLP) is the information 17 extraction (Khurana, D. et al., 2023; Landosi, M.Y. 18 et al., 2023) with important applicability in the 19 medical field (Zhang, T. et al., 2021). In fact, an 20 essential step for various medical decision-making 21 processes is the identification and automatic 22 verification of specific claims from unstructured 23 user-generated text data (Doppalaudi, S. et al., 24 2022). It refers to the idea of cause and effect. The 25 causal relation extraction has been mostly treated 26 as a subtask of relation extraction, being helpful in 27 the context of personalized healthcare (Khetan, V. 28 et al., 2022). Extracted causal information from 29 clinical notes, as a crucial step within medical 30 1 SemEval-2023 Task 8 overview is available at: https://causalclaims.github.io/ decision-making, can and have to be combined 31 (Yang, J. et al., 2022). The aim of this paper is 32 focused on identification of causal claims1, 33 experience, etc., in a provided multi (or single) 34 sentence text snippet. The rest of the paper is 35 organized as follows: section 2 briefly presents 36 studies related to claim identification, section 3 37 provides information about the system designed to 38 classify text sentences, section 4 describes the 39 experimental setups. Section 5 resumes the results 40 of the conducted experiments, with their 41 interpretations, followed by section 6 with the 42 conclusions. 43 44 2 Background 45 This topic has attracted significant attention in 46 recent years, evidenced by increasing number of 47 workshops (e.g., Workshop on Curative Power of 48 MEdical Data - MEDA 2017; 2018; 2020, 49 Workshop on Events and Stories in the News 2018) 50 (Cohen, K.B et al., 2020., Gifu, D. et al., 2019, 51 Tommaso, C. et al., 2018). 52 Competitions such as SemEval-2023 Task 8: 53 Causal Medical Claim Identification and 54 Extraction from Social Media Posts are attractive, 55 especially since the problem of labeled data is 56 somewhat solved, since the automatic 57 identification and automatic verification of medical 58 claims depends on them. 59 Togedemaru at SemEval-2023 Task 8: Causal Medical Claim Identification and …

Content truncated to 500 words