pkudblab at SemEval-2016 Task 6 : A Specific Convolutional Neural Network System for Effective Stance Detection

摘要: In this paper, we develop a convolutional neural network for stance detection in tweets. According to the official results, our system ranks 1 on subtask B (among 9 teams) and 2 A 19 twitter test set of SemEval2016 Task 6. The main contribution work is as follows. We design ”vote scheme” prediction instead predicting when accuracy validation reaches its maximum. Besides, make some improvement specific subtasks. For A, separate datasets into five sub-datasets according their targets, train models. B, establish two-class training dataset from domain corpus, then modify softmax layer perform three-class classification. Our can be easily re-implemented optimized other related tasks.