MIT at SemEval-2017 Task 10: Relation Extraction with Convolutional Neural Networks

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 978–984, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics MIT at SemEval-2017 Task 10: Relation Extraction with Convolutional Neural Networks Ji Young Lee∗ MIT jjylee@mit.edu Franck Dernoncourt∗ MIT francky@mit.edu Peter Szolovits MIT psz@mit.edu Abstract Over 50 million scholarly articles have been published: they constitute a unique repository of knowledge. In particular, one may infer from them relations be- tween scientific concepts. Artificial neu- ral networks have recently been explored for relation extraction. In this work, we continue this line of work and present a system based on a convolutional neural network to extract relations. Our model ranked first in the SemEval-2017 task 10 (ScienceIE) for relation extraction in sci- entific articles (subtask C). 1 Introduction and related work The number of articles published every year keeps increasing (Druss and Marcus, 2005; Larsen and Von Ins, 2010), with well over 50 million schol- arly articles published so far (Jinha, 2010). While this repository of human knowledge contains in- valuable information, it has become increasingly difficult to take advantage of all available infor- mation due to its sheer amount. One challenge is that the knowledge present in scholarly articles is mostly unstructured. One approach to organize this knowledge is to clas- sify each sentence (Kim et al., 2011; Amini et al., 2012; Hassanzadeh et al., 2014; Dernon- court et al., 2016). Another approach is to extract entities and relations between them, which is the focus of the ScienceIE shared task at SemEval- 2017 (Augenstein et al., 2017). Relation extraction can be seen as a process comprising two steps that can be done jointly (Li and Ji, 2014) or separately: first, entities of in- terest need to be identified, and second, the rela- tion among the entities has to be determined. In ∗ These authors contributed equally to this work. this work, we concentrate on the second step (of- ten referred to as relation extraction or classifica- tion) and on binary relations, i.e. relations be- tween two entities. Extracted relations can be used for a variety of tasks such as question-answering systems (Ravichandran and Hovy, 2002), ontol- ogy extension (Schutz and Buitelaar, 2005), and clinical trials (Frunza and Inkpen, 2011). In this paper, we describe the system that we submitted for the ScienceIE shared task. Our sys- tem is based on convolutional neural networks and ranked first for relation extraction (subtask C). Existing systems for relation extraction can be classified into five categories (Zettlemoyer, 2013): systems based on hand-built patterns (Yangar- ber and Grishman, 1998), bootstrapping meth- ods (Brin, 1998), unsupervised methods (Gonza- lez and Turmo, 2009), distant supervision (Snow et al., 2004), and supervised methods. We focus on supervised methods, as the ScienceIE shared task provides a labeled training set. Supervised methods for relation extrac- tion commonly employ support vector ma- chines (Uzuner et al., 2010, 2011; Minard et al., 2011; GuoDong et al., 2005), na¨ıve Bayes (Za- yaraz and Kumara, 2015), maximum entropy (Sun and Grishman, …

Content truncated to 500 words