Filtering of Impertinent Remarks using distributed expression

Abstract

This paper proposes a filtering method, using supervised learning with distributed representation of posted documents and vectors as features. In recent years, online discussion platforms have witnessed a great popularity. However, there here are many harmful contents such as unrelated spam in these discussion platforms, and violent remarks that insult and discriminate against opponents. As result, it becomes necessary to build a discussion platform that allows online users to participate safely by removing inappropriate remarks. To remove inappropriate remarks, understanding and classifying the meanings of documents is needed. Toward this end, we adopt doc2vec and ELMo to word embedding documents. In addition, we constructed a vectorized document by using document similarity calculation and deep neural networks (DNN). The experimental results show that the proposed method is able to classify with higher accuracy.