Learning local and global contexts using a convolutional recurrent network model for relation classification in biomedical text

Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 311–321, Vancouver, Canada, August 3 - August 4, 2017. c⃝2017 Association for Computational Linguistics Learning local and global contexts using a convolutional recurrent network model for relation classification in biomedical text Desh Raj and Sunil Kumar Sahu and Ashish Anand Department of Computer Science and Engineering Indian Institute of Technology Guwahati Guwahati, India Abstract The task of relation classification in the biomedical domain is complex due to the presence of samples obtained from het- erogeneous sources such as research ar- ticles, discharge summaries, or electronic health records. It is also a constraint for classifiers which employ manual fea- ture engineering. In this paper, we pro- pose a convolutional recurrent neural net- work (CRNN) architecture that combines RNNs and CNNs in sequence to solve this problem. The rationale behind our ap- proach is that CNNs can effectively iden- tify coarse-grained local features in a sen- tence, while RNNs are more suited for long-term dependencies. We compare our CRNN model with several baselines on two biomedical datasets, namely the i2b2- 2010 clinical relation extraction challenge dataset, and the SemEval-2013 DDI ex- traction dataset. We also evaluate an at- tentive pooling technique and report its performance in comparison with the con- ventional max pooling method. Our re- sults indicate that the proposed model achieves state-of-the-art performance on both datasets.1 1 Introduction Relation classification is the task of identifying the semantic relation present between a given pair of entities in a piece of text. Since most search queries are some forms of binary fac- toids (Agichtein et al., 2005), modern question- answering systems rely heavily upon relation clas- sification as a preprocessing step (Fleischman 1The code for the can be found at: https://github.com/desh2608/ crnn-relation-classification. et al., 2003; Lee et al., 2007). Accurate relation classification also facilitates discourse processing and precise sentence interpretations. Hence, this task has witnessed a great deal of attention over the last decade (Mintz et al., 2009; Surdeanu et al., 2012). In the biomedical domain, in particular, extract- ing such tuples from data may be essential for identifying protein and drug interactions, symp- toms and causes of diseases, among others. Fur- ther, since clinical data tends to be obtained from multiple (and diverse) information sources such as journal articles, discharge summaries, and elec- tronic patient records, relation classification be- comes a more challenging task. To identify relations between entities, a vari- ety of lexical, syntactic, or pragmatic cues may be exploited, which results in a challenging vari- ability in the type of features used for classifi- cation purpose. Due to this variability, a num- ber of approaches have been suggested, some of which rely on features extracted from POS tag- ging, morphological analysis, dependency pars- ing, and world knowledge (Kambhatla, 2004; San- tos et al., 2015; Suchanek et al., 2006; Mooney and Bunescu, 2005; Bunescu and Mooney, 2005). Deep learning architectures have recently gathered much interest because of their ability to conve- niently extract relevant features without the …

Content truncated to 500 words