SSNSheerinKavitha at SemEval-2023 Task 7: Semantic Rule Based Label Prediction Using TF-IDF and BM25 Techniques

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 950–957 July 13-14, 2023 ©2023 Association for Computational Linguistics SSNSheerinKavitha at SemEval-2023 Task 7: Semantic rule based label prediction using TF-IDF and BM25 techniques Sheerin Sitara Noor Mohamed and Kavitha Srinivasan Department of Computer Science and Engineering Sri Sivasubramaniya Nadar College of Engineering, Kalavakkam - 603110, India. sheerinsitaran@ssn.edu.in, kavithas@ssn.edu.in Abstract The advancement in the healthcare sector as- sures improved diagnosis and supports appro- priate decision making in medical domain. The medical domain data can be either radiology images or clinical data. The clinical data plays a major role in the healthcare sector by prevent- ing and treating the health problem based on the evidence learned from the trials. This paper is related to multi-evidence natural language inference for clinical trial data analysis and its solution for the given subtasks (SemEval 2023 Task 7 - NLI4CT). In subtask 1 of NLI4CT, the inference relationship (entailment or con- tradiction) between the Clinical Trial Reports (CTRs) statement pairs with respect to the Clin- ical Trial Data (CTD) statement are determined. In subtask 2 of NLI4CT, predicted label (in- ference relationship) are defined and justified using set of supporting facts extracted from the premises. The objective of this work is to derive the conclusion from premises (CTRs statement pairs) and extracting the supporting premises using proposed Semantic Rule based Clinical Data Analysis (SRCDA) approach. From the results, the proposed model attained an high- est F1-score of 0.667 and 0.716 for subtasks 1 and 2 respectively. The novelty of this pro- posed approach includes, creation of External Knowledge Base (EKB) along with its suitable semantic rules based on the input statements. 1 Introduction The Natural Language Processing (NLP) is one of the fastest growing sub-domain of Artificial In- telligence (AI) for medical domain (Basu et al., 2020). The contributions of NLP for medical do- main are transcribing medical documents, clinical entity resolver and end-to-end drug discovery. To improvise the existing applications, research fo- rums are conducting tasks for current requirements. One among them is SemEval 2023 and the task is multi-evidence Natural Language Inference (NLI) for clinical trial data (Jullien et al., 2023) and anal- ysis. The motivation of this task is to help the clinical practitioner to stay updated on all current literature by providing an opportunity to support the large-scale interpretation and retrieval of medi- cal evidence. To achieve this, NLI is applied on the clinical trial dataset (breast cancer Clinical Trial Reports (CTRs) and Clinical Trial Data (CTD)) to infer the medical evidence based on the relationship between premises and hypothesis. The clinical trial data analysis is divided into two subtasks namely, textual entailment and evidence retrieval. In textual entailment (subtask 1), the hypothesis is derived from CTRs premises of four sections namely, intervention, eligibility, results and adverse events with respect to the CTD statement. Under each section there are three to eighty-two premises. In evidence retrieval (subtask 2), the anticipated premises which play a significant role in deriving the …

Content truncated to 500 words