The Women in AI speaker series, a collaboration between the Schwartz Reisman Institute for Technology and Society and Deloitte, welcomes Sophia Ananiadou, a professor of computer science at the University of Manchester, and director of UK National Centre for Text Mining, which provides tools, resources, systems and infrastructure for biomedicine.

Ananiadou’s research focuses on natural language processing (NLP) and text mining for biomedical contexts. In this session, she will explore how NLP techniques can be used to summarize research to increase the speed and reliability of knowledge discovery, and will discuss current trends in biomedical text summarization, the use of pre-trained language models (PLMs), benchmarks, evaluation measures, and challenges faced in both extractive and abstractive methods.

Abstract:

Making sense of the growing literature across a wide range of research domains requires methods that will increase the speed and reliability of knowledge discovery. Moreover, due to the proliferation of scientific databases and ontologies, discovery of previously unknown knowledge demands that scientists engage with many resources, covering different levels and views of (multiple) domain spaces in context (e.g., degree of confidence in a finding). With the availability of large pre-trained transformer language models, neural natural language processing (NLP) models have been deployed for several downstream tasks, including information extraction and summarization. Information extraction (e.g., named entity recognition and inter-sentence relation extraction) can capture implicit relations. Moreover, events which encapsulate n-ary relationships (e.g., interactions between any number of concepts) are extracted with end-to-end neural methods, including capturing richer contextual information such as certainty and polarity.

Text summarization techniques are used to support users in accessing information efficiently, by retaining only the most important semantic information contained within documents. Text summarization is important in a variety of scenarios, including systematic reviews (synthesis), and evidence-based medicine. In this talk, I will discuss current trends in biomedical text summarization, the use of pre-trained language models (PLMs), benchmarks, evaluation measures, and challenges faced in both extractive and abstractive methods, including recent approaches such as hybrid unsupervised summarization methods using salience, the incorporation of fine-grained medical knowledge into PLMs to extractive summarization, and long document summarization using local and global semantics.