Text Summarization as the Potential Technology for Intelligent Internet of Things

Applying automatic text summarization technology to Internet of Things can save network cost and improve computing speed. However, the models of generated text summarization are always using sequence-to-sequence model with attention mechanism. Unluckily, this method for abstractive summarization has two main shortcomings: first, they fail to address unknown words problems, second, their generated summaries are not very readable because of repetition. In our work, our goal is to enhance the semantic coherence of summaries for original texts. In order to this end, we propose a new model that augments the traditional model in two ways. First, we apply semantic relevance to pointer-generator network to get high similarity between source texts and our summaries. Second, we change the mechanism of coverage and use it to pointer-generator network to discourage repetition. Following other works, we apply our new model to Chinese social media dataset LCSTS. Our experiments suggested that our new model outperforms current abstractive baseline systems on the dataset.