| Automatic document summarization is the process of reducing a text document with a computer program in order to create a coherent short passage that represents the most important points of the original document.In this thesis, we propose to use machine learning approaches for extractive summarization.We select four machine learning approaches according to the large amount of investigation and the accumulation of long-term work to achieve the automatic summarization system. At the same time, we concentrate on working on the features that transform the text into the appropriate format for those algorithms. We introduce the word embedding feature and the traditional features like position, length or something else. This thesis implements many experiments to show that a proper way to combine the features can lead to a better result. In addition, these attempts laid the foundation of the cooperative training system.We try to find a way to integrate the approaches, because we believe that different method can give a different view of the same question. If there is a right way to get these approaches to work together, there will be a better result. And the unlabeled data may provide information somehow when there is not enough labeled corpus. According to the above conditions and assumptions, we find out that cooperative training matches well. So we just use the algorithms to the task. The experimental results showed that cooperative training is a very useful technique that can increase the classification accuracy by leveraging the information from the unlabeled data, and the dissimilar machine learning approaches integrate in a befitting way can improve the performance of the system. |
