Unsupervised Extractive Summarization Based on Context Information

Unsupervised extractive summarization do not need parallel corpus pairs for training, so they are widely used in many scenarios. Many unsupervised extractive summarization methods usually lose the context information of sentences. Based on the context information, this paper proposes a model of unsupervised extractive summarization. At the same time, we tested many ways to express the context information. We also studied the relationship between sentences in the abstract. We have proved that the context information and the relationship between the sentences are very helpful to the task and we developed an unsupervised summarization system without any training.