Transformer-based Summarization by Exploiting Social Information

Information from social media, e.g. user comments and tweets, provides an additional channel that enriches the content of Web documents. This paper introduces a model by exploiting relevant social information to enhance web document summarization. Different from prior studies using feature engineering, we empower our model by utilizing transformers. To do that, relevant user posts are paired with sentences for utilizing the support from social information. The paired information is fed into transformers for taking full advantage of the contextual aspect. The model is then adapted by stacking an additional convolution neural network layer for classification. Experimental results on two English datasets show that our model achieves promising results for summarizing single documents.