UniMa at SemEval-2018 Task 7: Semantic Relation Extraction and Classification from Scientific Publications

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 826–830 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics UniMa at SemEval-2018 Task 7: Semantic Relation Extraction and Classification from Scientific Publications Thorsten Keiper,1 Zhonghao Lyu,1 Sara Pooladzadeh,1 Yuan Xu,1 Jingyi Zhang,1 Anne Lauscher,1,2 and Simone Paolo Ponzetto1 1University of Mannheim, Germany 2Stuttgart Media University, Germany {tkeiper, zlyu, spooladz, yuaxu, jizhang}@mail.uni-mannheim.de {anne, simone}@informatik.uni-mannheim.de Abstract Large repositories of scientific literature call for the development of robust meth- ods to extract information from scholarly papers. This problem is addressed by the SemEval 2018 Task 7 on extracting and classifying relations found within scien- tific publications. In this paper, we present a feature-based and a deep learning-based approach to the task and discuss the results of the system runs that we submitted for evaluation. 1 Introduction Nowadays, the exploding amount of scientific lit- erature makes it ever more problematic for re- searchers and scholars to get focused access to the state-of-the-art in a certain field of science. There- fore, it is getting increasingly important to de- velop effective computational approaches for ex- tracting information from large scholarly corpora. SemEval 2018 Task 7 (G´abor et al., 2018) ad- dresses this problem with a shared task on extract- ing and classifying semantic relations in scientific papers. The task is divided into two subtasks: 1. Relation Classification. Given an existing rela- tion between two entities and their context, the task is to predict the label of the relation out of the set of possible classes, namely USAGE, RESULT, MODEL-FEATURE, PART WHOLE, TOPIC, COMPARISON. The task is decom- posed into two different scenarios according to the data used, namely with either manually an- notated entities (1.1) or noisy data with auto- matically extracted entities (1.2). 2. Relation Extraction and Classification. This subtask addresses the whole end-to-end pipeline of relation extraction. Given abstracts of scientific papers annotated with entities, systems are required to extract pairs of entities in a semantic relation, as well as assign a label and directionality to the extracted relation. We participated in the subtask 1.1, relation classi- fication on clean data, and subtask 2 (relation ex- traction only). Our approach relies on supervised learning using Support Vector Machines (SVMs), k-Nearest Neighbors (kNNs), and Convolutional Neural Networks (CNNs). We were ranked 16th on task 1.1 with an F1 score 44.0% and 7th on task 2 with an F1 of 28.4%. 2 Related Work A series of supervised systems for extracting keyphrases and relations in scientific publications was presented in the context of the SemEval task 10 in 2017 (Augenstein et al., 2017). In contrast to the present task formulation, the set of relations consisted only of two, namely hyponymy and syn- onymy. For this task, the best systems were those based on neural approaches. Lee et al. (2017) obtained the highest score by employing a convolutional neural network with a specific embedding layer encoding manually crafted features such as the word, the position of the word and the part-of-speech (POS) tag. …

Content truncated to 500 words