MITRE at SemEval-2016 Task 6: Transfer Learning for Stance Detection

TL;DR

This paper presents a transfer learning approach using recurrent neural networks and word embeddings to achieve top performance in stance detection on tweets, effectively leveraging large unlabeled datasets.

Contribution

Introduces a transfer learning method with neural networks and word embeddings for stance detection, achieving state-of-the-art results on Twitter data.

Findings

Achieved an F1 score of 67.8 in stance detection.

Utilized distant supervision and hashtag prediction for feature learning.

Demonstrated the effectiveness of transfer learning in social media analysis.

Abstract

We describe MITRE's submission to the SemEval-2016 Task 6, Detecting Stance in Tweets. This effort achieved the top score in Task A on supervised stance detection, producing an average F1 score of 67.8 when assessing whether a tweet author was in favor or against a topic. We employed a recurrent neural network initialized with features learned via distant supervision on two large unlabeled datasets. We trained embeddings of words and phrases with the word2vec skip-gram method, then used those features to learn sentence representations via a hashtag prediction auxiliary task. These sentence vectors were then fine-tuned for stance detection on several hundred labeled examples. The result was a high performing system that used transfer learning to maximize the value of the available training data.