Incorporating Hand-crafted Features in a Neural Network Model for Stance Detection on Microblog
Mumtahina Ahmed, Abu Nowshed Chy, Nihad Karim Chowdhury · 2020
Microblogs, especially twitter, has made unprecedented opportunities for users to assert their stance towards various entities, issues, and events. Analyzing user stances from tweets provide opportunities to various organizations for decision making. However, it is challenging to identify the stance of a tweet due to its short length characteristics and idiosyncratic nature. Most of the previous studies explore either neural network-based features or hand-crafted features for their learning models. In this paper, we introduce a stance detection method that incorporates both the deep semantic features and hand-crafted features in a unified neural model. We exploit several opinionated lexicons and other textual and twitter-specific characteristics to extract a rich set of hand-crafted features. We use a supervised feature selection method to devise effective features and pass them to train a multilayer perceptron (MLP) network. Besides, we employ a convolutional layer in conjunction with a BiLSTM network (Conv-BiLSTM) to extract the higher-level contextual features. Extensive experiments on the SemEval-2016 stance detection dataset demonstrate the efficiency of our method over several state-of-the-art methods.