EELECTION at SemEval-2017 Task 10: Ensemble of nEural Learners for kEyphrase ClassificaTION
Abstract
This paper describes our approach to the SemEval 2017 Task 10: "Extracting Keyphrases and Relations from Scientific Publications", specifically to Subtask (B): "Classification of identified keyphrases". We explored three different deep learning approaches: a character-level convolutional neural network (CNN), a stacked learner with an MLP meta-classifier, and an attention based Bi-LSTM. From these approaches, we created an ensemble of differently hyper-parameterized systems, achieving a micro-F1-score of 0.63 on the test data. Our approach ranks 2nd (score of 1st placed system: 0.64) out of four according to this official score. However, we erroneously trained 2 out of 3 neural nets (the stacker and the CNN) on only roughly 15% of the full data, namely, the original development set. When trained on the full data (training+development), our ensemble has a micro-F1-score of 0.69. Our code is available from https://github.com/UKPLab/semeval2017-scienceie.
Keywords
Cite
@article{arxiv.1704.02215,
title = {EELECTION at SemEval-2017 Task 10: Ensemble of nEural Learners for kEyphrase ClassificaTION},
author = {Steffen Eger and Erik-Lân Do Dinh and Ilia Kuznetsov and Masoud Kiaeeha and Iryna Gurevych},
journal= {arXiv preprint arXiv:1704.02215},
year = {2017}
}
Comments
In revision, changed to pdfTeX output