Enhancing health literacy accessibility: an exploration of computational approaches to 
lay summarization

Roca Beltran, Patrícia

Enhancing health literacy accessibility: an exploration of computational approaches to lay summarization

License

Citation

Roca Beltran, P. Enhancing health literacy accessibility: an exploration of computational approaches to lay summarization. 2024. handle: http://hdl.handle.net/10230/61501

This citation was generated automatically.

Abstract

As demonstrated by the global Covid-19 pandemic, it is crucial that everyone has access to health literacy in order to make informed decisions in day-to-day life. However, the inherent complexity of biomedical texts and the lack of prior knowledge among non-specialists restricts access to this crucial information that significantly impacts society. To improve the accessibility of biomedical information to the general population, this thesis aims to develop a computational system for generating simplified summaries of biomedical articles automatically. To achieve this goal, we use a dataset which has been provided by the organizers of the shared task BioLaySumm 2024, one of the tasks offered in the BioNLP Workshop of the Association for Computational Linguistics (ACL). This dataset consists of pairs of articles and their simplified summaries. Using these datasets, we build the systems based on transformer networks using two different strategies for content selection. The basic structure of the systems consists of a pre-trained language model, called BART, which we fine-tune for abstractive lay summary generation. In contrast, to select important parts in an extractive manner by choosing relevant fragments of the original text, we use an algorithm called TextRank. We experimented with combinations of the two models to compare the performances.