LABDA at SemEval-2017 Task 10: Relation Classification between keyphrases via Convolutional Neural Network

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 969–972, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics LABDA at SemEval-2017 Task 10: Relation Classification between keyphrases via Convolutional Neural Network V´ıctor Su´arez-Paniagua, Isabel Segura-Bedmar and Paloma Mart´ınez Computer Science Department Carlos III University of Madrid Legan´es 28911, Madrid, Spain vspaniag,isegura,pmf@inf.uc3m.es Abstract In this paper, we describe our participa- tion at the subtask of extraction of relation- ships between two identified keyphrases. This task can be very helpful in improv- ing search engines for scientific articles. Our approach is based on the use of a con- volutional neural network (CNN) trained on the training dataset. This deep learn- ing model has already achieved successful results for the extraction relationships be- tween named entities. Thus, our hypothe- sis is that this model can be also applied to extract relations between keyphrases. The official results of the task show that our architecture obtained an F1-score of 0.38% for Keyphrases Relation Classifica- tion. This performance is lower than the expected due to the generic preprocessing phase and the basic configuration of the CNN model, more complex architectures are proposed as future work to increase the classification rate. 1 Introduction Nowadays, a deluge of scientific articles is pub- lished every year, which demonstrates that we are living in an emerging knowledge era. An impor- tant drawback of this situation is that the study of a given field or problem requires reviewing an huge number of scientific publications, becoming such a very arduous task. Most search engines apply linguistic normalization (such as lemmatization or stemming) and some of them also exploit the se- mantic analysis of texts in order to detect concepts to improve their recall. The goal of the ScienceIE Task at SemEval 2017 (Augenstein et al., 2017) is the extraction of keyphrases (such as MATE- RIALS, PROCESSES and TASKS) and relation- ships between them from scientific articles. This competition provides a common evaluation frame- work to researches allowing a fair way to evalu- ate and compare their approaches. Our participa- tion focuses on the subtask of extracting relation- ships between keyphrases. In particular, these re- lationships are HYPONYM-OF (for example, ’fe- mur’ is HYPONYM-OF ’bone’), SYNONYM-OF (for example, ’ophthalmologist’ is SYNONYM- OF ’oculist’), and NONE. The detection of these relationships between keyphrases can improve the performance of current researches. In this paper, we describe the participation of the group LaBDA for participating in the subtask C (extraction of relationships between keyphrases) evaluated on the scenario 3, where the test dataset includes texts as well as the annotations of their keyphrases (boundaries and types). Our approach is based on the CNN proposed in (Kim, 2014), which was the first work to exploit a CNN for the task of sentence classification. This model was able to infer the class of each sentence, and re- turned good results without the need for external information. To this end, the model computes an output vector, which describes the whole sentence, and applies …

Content truncated to 500 words