JUST-KM at SemEval-2023 Task 7: Multi-evidence Natural Language Inference using Role-based Double Roberta-Large

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 447–452 July 13-14, 2023 ©2023 Association for Computational Linguistics JUST-KM at SemEval-2023 Task 7: Multi-evidence Natural Language Inference using Role-based Double Roberta-Large Kefah Alissa and Malak Abdulah Jordan University of Science and Technology Irbid, Jordan {kfalissa20, mabdullah}@just.edu.jo Abstract In recent years, there has been a vast increase in the available clinical data. Variant Deep learning techniques are used to enhance the retrieval and interpretation of these data. This task deployed Natural language inference (NLI) in Clinical Trial Reports (CTRs) to provide individualized care that is supported by evi- dence. A collection of breast cancer clinical trial records, statements, annotations, and la- bels from experienced domain experts. NLI presents a chance to advance the widespread un- derstanding and retrieval of medical evidence, leading to significant improvements in connect- ing the most recent evidence to personalized care. The primary objective is to identify the inference relationship (entailment or contradic- tion) between pairs of clinical trial records and statements. In this research, we used different transformer-based models, and the proposed model, "Role-based Double Roberta-Large," achieved the best result on the testing dataset with F1-score equal to 67.0% 1 Introduction Breast cancer clinical trial reports (CTRs) summa- rize the results of clinical trials conducted to test new treatments or interventions for breast cancer. These reports typically include information on the study design, patient population, treatment regi- mens, outcomes (such as response rates and side effects), and overall conclusions. These reports provide a comprehensive and transparent account of the trial results to the scientific community and the general public to inform future research and clinical practice. The results of breast cancer clin- ical trials play a critical role in the development of new treatments and in improving outcomes for patients with breast cancer(McDonald et al., 2016). The number of published (CTRs) has grown sig- nificantly in recent years, with over 10,000 CTRs for breast cancer alone. As a result, it has become increasingly more work for clinical practitioners to keep up with all the current literature and provide patient care based on the latest evidence (DeYoung et al., 2020). Natural Language Inference (NLI) has the po- tential to enhance the analysis and interpretation of CTRs by providing a more automated and consis- tent approach to extracting information and draw- ing inferences from these reports(Agrawal et al., 2019). Multi-evidence Natural Language Infer- ence (NLI) is a task in natural language processing that involves determining the relationship between two pieces of text, specifically whether the sec- ond text is an entailment, contradiction, or neutral concerning the first text. In multi-evidence NLI, multiple pieces of evidence are used to make the determination instead of relying on a single text. This can be useful when the relationship between the two texts needs to be clarified or additional context is needed to make an accurate determina- tion(Storks et al., 2019). NLI has been approached using various methods, including symbolic logic, knowledge bases, and neural networks. Recently, it’s become …

Content truncated to 500 words