Target Specific Stance Detection from Social Media with Multilayer Perceptron

Social media allows individual to share their judgment or standpoint towards any topic or news. People easily relies on the news that are circulated on social media. Generally, controversial political and social news can instigate people to react. So, automatic detection of stance towards any topic is very much essential that can help to detect rumor and fake news. Context identification and truthfulness of any post can improve the existing stance detection model. This research proposed a novel stance detection model that considers paragraph2vec embedding model for contextual analysis, TF-IDF for importance and relevance of each term and LIWC for emotion and psycholinguistic analysis with Multilayer Perceptron (MLP) classifier. This study considers English language standard dataset and achieved 10–35% improvement of F1-score compared to baseline model. Among various combination of feature extraction method, the proposed stance detection model portrays achievable performance.