GU IRLAB at SemEval-2018 Task 7: Tree-LSTMs for Scientific Relation Classification

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 831–835 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics GU IRLAB at SemEval-2018 Task 7: Tree-LSTMs for Scientific Relation Classification Sean MacAvaney, Luca Soldaini, Arman Cohan, and Nazli Goharian Information Retrieval Lab Department of Computer Science Georgetown University {firstname}@ir.cs.georgetown.edu Abstract SemEval 2018 Task 7 focuses on relation ex- traction and classification in scientific litera- ture. In this work, we present our tree-based LSTM network for this shared task. Our ap- proach placed 9th (of 28) for subtask 1.1 (rela- tion classification), and 5th (of 20) for subtask 1.2 (relation classification with noisy entities). We also provide an ablation study of features included as input to the network. 1 Introduction Information Extraction (IE) has applications in a variety of domains, including in scientific litera- ture. Extracted entities and relations from scien- tific articles could be used for a variety of tasks, in- cluding abstractive summarization, identification of articles that make similar or contrastive claims, and filtering based on article topics. While onto- logical resources can be leveraged for entity ex- traction (G´abor et al., 2016), relation extraction and classification still remains a challenging task. Relations are particularly valuable because (unlike simple entity occurrences) relations between enti- ties capture lexical semantics. SemEval 2018 Task 7 (Semantic Relation Extraction and Classification in Scientific Papers) encourages research in rela- tion extraction in scientific literature by providing common training and evaluation datasets (G´abor et al., 2018). In this work, we describe our ap- proach using a tree-structured recursive neural net- work, and provide an analysis of its performance. There has been considerable previous work with scientific literature due to its availability and inter- est to the research community. A previous shared task (SemEval 2017 Task 10) investigated the ex- traction of both keyphrases (entities) and relations in scientific literature (Augenstein et al., 2017). However, the relation set for this shared task was limited to just synonym and hypernym relation- ships. The top three approaches used for relation- only extraction included convolutional neural net- works (Lee et al., 2017a), bi-directional recur- rent neural networks with Long Short-Term Mem- ory (LSTM, Hochreiter and Schmidhuber, 1997) cells (Ammar et al., 2017), and conditional ran- dom fields (Lee et al., 2017b). There are several challenges related to scientific relation extraction. One is the extraction of the entities themselves. Luan et al. (2017) produce the best published results on the 2017 ScienceIE shared task for entity extraction using a semi- supervised approach with a bidirectional LSTM and a CRF tagger. Zheng et al. (2014) provide an unsupervised technique for entity linking scientific entities in the biomedical domain to an ontology. Contribution. Our approach employs a tree- based LSTM network using a variety of syntac- tic features to perform relation label classification. We rank 9th (of 28) when manual entities are used for training, and 5th (of 20) when noisy entities are used for training. Furthermore, we provide an ab- lation analysis of the features used by …

Content truncated to 500 words