YNU-HPCC at SemEval-2023 Task7: Multi-evidence Natural Language Inference for Clinical Trial Data Based a BioBERT Model

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 664–670 July 13-14, 2023 ©2023 Association for Computational Linguistics YNU-HPCC at SemEval-2023 Task7: Multi-evidence Natural Language Inference for Clinical Trial Data based on a BioBERT Model Chao Feng, Jin Wang and Xuejie Zhang School of Information Science and Engineering Yunnan University Kunming, China fengchao@mail.ynu.edu.cn, {wangjin, xjzhang}@ynu.edu.cn Abstract This paper describes the system for the YNU- HPCC team in subtask 1 of the SemEval- 2023 Task 7: Multi-evidence Natural Language Inference for Clinical Trial Data (NLI4CT). This task requires judging the textual entail- ment relationship between the given CTR and the statement annotated by the expert an- notator. This system is based on the fine- tuned Bi-directional Encoder Representation from Transformers for Biomedical Text Min- ing (BioBERT) model with supervised con- trastive learning and back translation. Super- vised contrastive learning is to enhance the clas- sification, and back translation is to enhance the training data. Our system achieved rela- tively good results on the competition’s official leaderboard. The code of this paper is available at https://github.com/facanhe/SemEval-2023- Task7. 1 Introduction With the massive increase of clinical trial reports (CTRs) publications, there are more than 10000 clinical trial reports for breast cancer at present. Therefore, it is not feasible for clinical practition- ers to timely understand all current literature, so as to provide personalized evidence-based nursing (DeYoung et al., 2020). In this context, the ap- plication of natural language inference (NLI) can promote the development of large-scale interpre- tation and retrieval of medical evidence, and the successful development of the system can greatly promote us to combine the latest evidence to sup- port personalized care (Sutton et al., 2020). To this end, the purpose of this task is to use clinical trial data to conduct multiple evidence natural language inference. The task is based on a group of clinical trial reports of breast cancer, statements, explanations, and labels annotated by domain expert annotators. Each clinical trial report data of patients consists of four sections: (1) Eligibility criteria: a series of conditions that allow patients to participate in the clinical trial. (2) Intervention: information about the type, dose, frequency, and duration of the treat- ment studied. (3) Results: number of participants in the trial, result measurement, unit, and results. (4) Adverse Events: physical signs and symptoms of patients observed during the clinical trial. An example of this task is shown in Table 1. The SemEval-2023 shared task7 consists of two subtasks (Jullien et al., 2023): • subtask 1: judgment of the reasoning relation- ship between the clinical trial report and the statement (contradiction or entailment); • subtask 2: given a set of clinical trial reporting premises and statements, extract the support- ing facts from the premises; The main difficulty of subtask 1 is textual entail- ment. Textual entailment describes the reasoning relationship between two texts, in which text 1 is the premise and text 2 is the hypothesis. If the premise can infer the hypothesis, it is an entail- ment, otherwise, it …

Content truncated to 500 words