Automatic Prediction of Evidence-based Recommendations via Sentence-level Polarity Classification

International Joint Conference on Natural Language Processing, pages 712–718, Nagoya, Japan, 14-18 October 2013. Automatic Prediction of Evidence-based Recommendations via Sentence-level Polarity Classification Abeed Sarker , Diego Moll´a-Aliod Centre for Language Technology Macquarie University Sydney, NSW 2109 {abeed.sarker, diego.molla-aliod}@mq.edu.au C´ecile Paris CSIRO – ICT Centre Sydney, NSW 2122 cecile.paris@csiro.au Abstract We propose a supervised classification ap- proach for automatically determining the polarities of medical sentences. Our po- larity classification approach is context- sensitive, meaning that the same sentence may have differing polarities depending on the context. Using a set of carefully se- lected features, we achieve 84.7% accu- racy, which is significantly better than cur- rent state-of-the-art for the polarity clas- sification task. Our analyses and exper- iments on a specialised corpus indicate that automatic polarity classification of key sentences can be utilised to generate evidence-based recommendations. 1 Introduction Evidence Based Medicine is a practice that re- quires practitioners to rely on the best available medical evidence when answering clinical queries. While this practice improves patient care in the long run, it poses a massive problem of informa- tion overload to practitioners because of the large volume of medical text available electronically (e.g., MEDLINE1 indexes over 22 million arti- cles). Research has shown that the act of searching for, appraising, and synthesising evidence from multiple documents generally requires more time than practitioners can devote (Ely et al., 1999). As a result, practitioners would benefit from au- tomatic systems that help perform these tasks and generate bottom-line recommendations. In this paper, we take the first steps towards the generation of bottom-line, evidence-based sum- maries. Our analyses reveal that the polarities of key sentences in medical documents can be utilised to determine final recommendations asso- ciated with a query. Key sentences refer to the 1http://www.ncbi.nlm.nih.gov/pubmed most important sentences in a medical abstract that are associated with a posed query. In our work, we use the sentences extracted by a domain-specific, query-focused text summariser. Consider the fol- lowing sentence for example: A significant body of evidence supports the use of long-acting bronchodilators and inhaled corticosteroids in reducing exacerbations in patients with moderate to severe COPD. The sentence is taken from a medical abstract, and clearly recommends the use of bronchodila- tors and inhaled corticosteroids, which are the context interventions in this case. In other words, it has a positive polarity for this task. Since pos- itively polarised key sentences generally represent the recommendations, we attempt to automatically identify the polarities of medical sentences as the first step towards generating bottom-line recom- mendations. We show that sentence-level polarity classification is a useful approach for generating evidence-based recommendations. We model the problem of sentence polarity classification as a bi- nary classification problem, and we present a su- pervised machine learning approach to automati- cally classify the polarities of key sentences. Our classification approach is context dependent, i.e., the same sentence can have differing polarities de- pending on the context. 2 Related Work Research work most closely related to ours is that by Niu et al. …

Content truncated to 500 words