SwissCheese at SemEval-2016 Task 4: Sentiment Classification Using an Ensemble of Convolutional Neural Networks with Distant Supervision

Proceedings of SemEval-2016, pages 1124–1128, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics SwissCheese at SemEval-2016 Task 4: Sentiment Classification Using an Ensemble of Convolutional Neural Networks with Distant Supervision Jan Deriu∗ ETH Zurich Switzerland jderiu@student.ethz.ch Maurice Gonzenbach∗ ETH Zurich Switzerland mauriceg@student.ethz.ch Fatih Uzdilli Zurich University of Applied Sciences Switzerland uzdi@zhaw.ch Aurelien Lucchi ETH Zurich Switzerland alucchi@inf.ethz.ch Valeria De Luca ETH Zurich Switzerland vdeluca@vision.ee.ethz.ch Martin Jaggi ETH Zurich Switzerland jaggi@inf.ethz.ch Abstract In this paper, we propose a classifier for predicting message-level sentiments of En- glish micro-blog messages from Twitter. Our method builds upon the convolutional sen- tence embedding approach proposed by (Sev- eryn and Moschitti, 2015a; Severyn and Mos- chitti, 2015b). We leverage large amounts of data with distant supervision to train an en- semble of 2-layer convolutional neural net- works whose predictions are combined using a random forest classifier. Our approach was evaluated on the datasets of the SemEval-2016 competition (Task 4) outperforming all other approaches for the Message Polarity Classifi- cation task. 1 Introduction Sentiment analysis is a fundamental problem aiming to give a machine the ability to understand the emo- tions and opinions expressed in a written text. This is an extremely challenging task due to the complex- ity of human language, which makes use of rhetor- ical devices such as sarcasm or irony. Deep neu- ral networks have shown great promises at captur- ing salient features for these complex tasks (Mikolov et al., 2013b; Severyn and Moschitti, 2015a). Par- ticularly successful for sentiment classification were Convolutional Neural Networks (CNN) (Kim, 2014; Kalchbrenner et al., 2014; Severyn and Moschitti, 2015a; Severyn and Moschitti, 2015b; Johnson and Zhang, 2015), on which our work builds upon. ∗ These authors contributed equally to this work These networks typically have a large number of pa- rameters and are especially effective when trained on large amounts of data. In this work, we use a dis- tant supervision approach to leverage large amounts of data in order to train a 2-layer CNN 1, extend- ing the 1-layer architecture proposed by (Severyn and Moschitti, 2015a). More specifically, we train a neural network using the following three-phase procedure: i) creation of word embeddings for ini- tialization of the first layer; ii) distant supervised phase, where the network weights and word embed- dings are trained to capture aspects related to senti- ment; and iii) supervised phase, where the network is trained on the provided supervised training data. We also combine the predictions of several neural networks using a random forest meta-classifier. The proposed approach was evaluated on the datasets of the SemEval-2016 competition, Task 4 (Nakov et al., 2016)2 for which it reaches state-of-the-art re- sults. 2 System Description 2.1 Convolutional Neural Networks We combine the outputs of two 2-layer CNNs hav- ing similar architectures but differing in the choice of certain parameters (such as the number of convo- lutional filters). These two networks were also ini- tialized using different word embeddings and used slightly different training data for the distant super- vised …

Content truncated to 500 words