Using Discourse Structure to Differentiate Focus Entities from Background Entities in Scientific Literature

Using Discourse Structure of Scientific Literature to Differentiate Focus from Background Entities in Pathogen Characterisation Antonio Jimeno Yepes1,2, Ameer Albahem1, and Karin Verspoor2,1 1School of Computing and Information Systems, University of Melbourne 2School of Computing Technologies, RMIT University Melbourne, Victoria, Australia antonio.jimeno@gmail.com, ameer.albahem@gmail.com, karin.verspoor@rmit.edu.au Abstract In the task of pathogen characterisation, we aim to discriminate mentions of biological pathogens that are actively studied in the research presented in scientific publications. These are the pathogens that are the focus of direct experimentation in the research, rather than those that are referred to for context or as playing secondary roles. This task is an in- stance of the more general problem of identi- fying focus entities in scientific literature, in which key entities of interest must be discrim- inated from other potentially relevant entities of the same type mentioned in the articles. In this paper, we explore the hypothesis that fo- cus pathogens can be differentiated from other, non-actively studied, pathogens mentioned in articles through analysis of the patterns of men- tions across different segments of a scientific paper, that is, using the discourse structure of the paper. We provide an indicative case study with the help of a small data set of PubMed ab- stracts that have been annotated with actively mentioned pathogens. 1 Introduction Global monitoring of repositories of potentially harmful biological materials is an important com- ponent of ensuring the health and safety of our populations. In this context, we are building an information extraction system to identify informa- tion related to experimentation with potentially dan- gerous biological pathogens – e.g. viruses, bacte- ria, and biological toxins – as well as to detect facilities that may serve as repositories of harmful pathogens. This system will systematically scan open access data sets for evidence of research on those pathogens, thereby supporting gathering of information from public resources for biosecurity purposes (Jarrad et al., 2015). A key requirement for automated characterisa- tion of research on pathogens using text-based in- formation sources, including the scientific litera- ture, is to identify pathogens that are actively stud- ied. An actively studied pathogen is defined as an organism that is subjected to direct physical experi- mentation in the research. Recognition of potentially relevant entities is rel- atively advanced using biomedical named entity recognition tools that detect biological nomencla- ture such as the names of biological organisms (e.g. as studied in the context of BioCreative (Smith et al., 2008)). However, differentiating mentions of actively studied organisms from other, background or incidental mentions of organisms poses a deeper natural language processing challenge. In the con- text of chemical patents, it has been suggested that only ~10% of chemical mentions play a major role within the patent (Akhondi et al., 2019). It is insuf- ficient to simply detect a mention of a potentially relevant pathogen name; it must also be decided whether that pathogen is a focus of the experiments. The main goal of our pathogen characterisation task is therefore to enable filtering out pathogens that are mentioned …

Content truncated to 500 words