Framework of Automatic Text Summarization Using Reinforcement Learning

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 256–265, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Framework of Automatic Text Summarization Using Reinforcement Learning Seonggi Ryang Graduate School of Information Science and Technology University of Tokyo sryang@is.s.u-tokyo.ac.jp Takeshi Abekawa National Institute of Informatics abekawa@nii.ac.jp Abstract We present a new approach to the problem of automatic text summarization called Au- tomatic Summarization using Reinforcement Learning (ASRL) in this paper, which models the process of constructing a summary within the framework of reinforcement learning and attempts to optimize the given score function with the given feature representation of a sum- mary. We demonstrate that the method of re- inforcement learning can be adapted to auto- matic summarization problems naturally and simply, and other summarizing techniques, such as sentence compression, can be easily adapted as actions of the framework. The experimental results indicated ASRL was superior to the best performing method in DUC2004 and comparable to the state of the art ILP-style method, in terms of ROUGE scores. The results also revealed ASRL can search for sub-optimal solutions efficiently under conditions for effectively selecting fea- tures and the score function. 1 Introduction Automatic text summarization aims to automatically produce a short and well-organized summary of sin- gle or multiple documents (Mani, 2001). Automatic summarization, especially multi-document summa- rization, has been an increasingly important task in recent years, because of the exponential explosion of available information. The brief summary that the summarization system produces allows readers to quickly and easily understand the content of orig- inal documents without having to read each individ- ual document, and it should be helpful for dealing with enormous amounts of information. The extractive approach to automatic summariza- tion is a popular and well-known approach in this field, which creates a summary by directly selecting some textual units (e.g., words and sentences) from the original documents, because it is difficult to gen- uinely evaluate and guarantee the linguistic quality of the produced summary. One of the most well-known extractive ap- proaches is maximal marginal relevance (MMR), which scores each textual unit and extracts the unit that has the highest score in terms of the MMR cri- teria (Goldstein et al., 2000). Greedy MMR-style algorithms are widely used; however, they cannot take into account the whole quality of the sum- mary due to their greediness, although a summary should convey all the information in given docu- ments. Global inference algorithms for the extrac- tive approach have been researched widely in recent years (Filatova and Hatzivassiloglou, 2004; McDon- ald, 2007; Takamura and Okumura, 2009) to con- sider whether the summary is “good” as a whole. These algorithms formulate the problem as integer linear programming (ILP) to optimize the score: however, as ILP is non-deterministic polynomial- time hard (NP-hard), the time complexity is very large. Consequently, we need some more efficient algorithm for calculations. We present a new approach to the problem of au- tomatic text summarization …

Content truncated to 500 words