Abstract Employing deep learning makes it possible to learn high-level features from raw data, resulting in more precise models. On the other hand, submodularity makes the solution scalable and provides the means to guarantee a lower bound for its performance. In this paper, a deep submodular network (DSN) is introduced, which is a deep network meeting submodularity characteristics. DSN lets modular and submodular features to participate in constructing a tailored model that fits the best with a problem. Various properties of DSN are examined and its learning method is presented. By proving that cost function used for learning process is a convex function, it is concluded that minimization can be done in polynomial time and also, by choosing a suitable learning rate and performing enough iterations, a lower empirical error can be ensured. Finally, in order to demonstrate the applicability of DSN for real-world problems, automatic multi-document summarization is considered and a summarizer called DSNSum is introduced. Then, the performance of DSNSum is compared with the state-of-the-art summarizers based on DUC 2004 and CNN/DailyMail corpora. The experimental results show that the performance of the proposed summarizer is comparable with the state-of-the-art methods.
Paper
Full text
Deep submodular network: An application to multi-document summarization
Semantic Scholar · Computer Science · 2020
Abstract
Abstract Employing deep learning makes it possible to learn high-level features from raw data, resulting in more precise models. On the other hand, submodularity makes the solution scalable and provides the means to guarantee a lower bound for its performance. In this paper, a deep submodular network (DSN) is introduced, which is a deep network meeting submodularity characteristics. DSN lets modular and submodular features to participate in constructing a tailored model that fits the best with a problem. Various properties of DSN are examined and its learning method is presented. By proving that cost function used for learning process is a convex function, it is concluded that minimization can be done in polynomial time and also, by choosing a suitable learning rate and performing enough iterations, a lower empirical error can be ensured. Finally, in order to demonstrate the applicability of DSN for real-world problems, automatic multi-document summarization is considered and a summarizer called DSNSum is introduced. Then, the performance of DSNSum is compared with the state-of-the-art summarizers based on DUC 2004 and CNN/DailyMail corpora. The experimental results show that the performance of the proposed summarizer is comparable with the state-of-the-art methods.