Texterra at SemEval-2018 Task 7: Exploiting Syntactic Information for Relation Extraction and Classification in Scientific Papers

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 821–825 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics Texterra at SemEval-2018 Task 7: Exploiting Syntactic Information for Relation Extraction and Classification in Scientific Papers Andrey Sysoev and Vladimir Mayorov Ivannikov Institute for System Programming of the Russian Academy of Sciences 25 Alexander Solzhenitsyn Street Moscow, Russia sysoev@ispras.ru, vmayorov@ispras.ru Abstract In this work we evaluate applicability of en- tity pair models and neural network architec- tures for relation extraction and classification in scientific papers at SemEval-2018. We carry out experiments with representing entity pairs through sentence tokens and through short- est path in dependency tree, comparing ap- proaches based on convolutional and recurrent neural networks. With convolutional network applied to shortest path in dependency tree we managed to be ranked eighth in subtask 1.1 (”clean data”), ninth in 1.2 (”noisy data”). Similar model applied to separate parts of the shortest path was mounted to ninth (extraction track) and seventh (classification track) posi- tions in subtask 2 ranking. 1 Introduction Information extraction is an important part of nat- ural language processing. During SemEval-2018 an evaluation devoted to extraction and classifi- cation of relations in scientific papers was held (G´abor et al., 2018). The task is described as fol- lows: given abstracts of scientific articles with de- tected entities, the goal is to choose correct re- lations for provided ⟨source, target⟩ entity pairs (subtask 1 - relation classification) and to deter- mine correct relations among all entity pairs (sub- task 2 - relation extraction and classification). The target quality metric in classification is macro- average of F1-scores of every class; for extraction scenario the target metric is F1-score. Our method is based on multinomial classifica- tion of entity pairs and their sentences with neural networks. We experiment with representing entity pairs through all sentence tokens and through to- kens along the shortest path between entities in de- pendency tree (Bunescu and Mooney, 2005). We employ convolutional (CNN) (LeCun et al., 1989) and bidirectional Long Short-Term Memory (biL- STM) (Hochreiter and Schmidhuber, 1997; Tan et al., 2015) neural network based approaches to encode sentences and dependency tree paths. In this work we mainly focus on relation clas- sification, so most of analysis and experiments are carried out for this task. Slightly modified mod- els which achieve the best results on subtask 1 are adapted for solving relation extraction and classi- fication problem. The rest of the paper is organized as follows: in section 2 we describe some known approaches for relation extraction and classification. In section 3 our approach is presented in details. Section 4 out- lines results of described approach evaluation on official SemEval-2018 task 7 test set. We wrap up with some final thoughts in section 5. 2 Related Work The relation extraction and classification prob- lem has a long history. Early approaches were based on manually constructed patterns, used to detect entities in the relation under considera- tion (Blaschke and Valencia, 2001). Further approaches utilized machine …

Content truncated to 500 words