A Statistical Relational Learning Approach to Identifying Evidence Based Medicine Categories

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 579–589, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics A Statistical Relational Learning Approach to Identifying Evidence Based Medicine Categories Mathias Verbeke♦ Vincent Van Asch♣ Roser Morante♣ Paolo Frasconi♠ Walter Daelemans♣ Luc De Raedt♦ ♦ Department of Computer Science, Katholieke Universiteit Leuven, Belgium {mathias.verbeke, luc.deraedt}@cs.kuleuven.be ♣ Department of Linguistics, Universiteit Antwerpen, Belgium {roser.morante, vincent.vanasch, walter.daelemans}@ua.ac.be ♠ Dipartimento di Sistemi e Informatica, Universit`a degli Studi di Firenze, Italy p-f@dsi.unifi.it Abstract Evidence-based medicine is an approach whereby clinical decisions are supported by the best available findings gained from scien- tific research. This requires efficient access to such evidence. To this end, abstracts in evidence-based medicine can be labeled using a set of predefined medical categories, the so- called PICO criteria. This paper presents an approach to automatically annotate sentences in medical abstracts with these labels. Since both structural and sequential information are important for this classification task, we use kLog, a new language for statistical relational learning with kernels. Our results show a clear improvement with respect to state-of-the-art systems. 1 Introduction Evidence-based medicine (EBM) or evidence-based practice (EBP) combines clinical expertise, the pref- erences and values of the patient and the best available evidence to make good patient care deci- sions. Clinical research findings are systematically reviewed, appraised and used to improve the patient care, for which efficient access to such evidence is required. In order to facilitate the search process, medical documents are labeled using a set of prede- fined medical categories, the PICO criteria. PICO is an acronym for the mnemonic concepts that are used to construct queries when searching for scientific ev- idence in the EBM process. The need to automatize the annotation process has initiated research into au- tomatic approaches to annotate sentences in medical documents with the PICO labels. As indicated by Kim et al. (2011), both the struc- tural information of the words in the sentence, and that of the sentences in the document are important features for this task. Furthermore, sequential infor- mation can leverage the dependencies between dif- ferent sentences in the text. Therefore we propose an approach using kLog (Frasconi et al., 2012) to tackle this problem. kLog is a new language for statistical relational learning with kernels, that is embedded in Prolog, and builds upon and links together concepts from database theory, logic programming and learn- ing from interpretations. Learning from interpreta- tions is a logical and relational learning setting (De Raedt et al., 2008) in which the examples are inter- pretations, that is, sets of tuples that are true in the examples. In a sense, each example can be viewed as a small relational database. kLog is able to trans- form relational into graph-based representations and apply kernel methods to extract an extended high- dimensional feature space. The choice for kLog was motivated by previous results (Verbeke et al., 2012), where we showed that a statistical relational learning …

Content truncated to 500 words