This paper presents a novel sentence extraction framework that takes into account the consecutiveness of important sentences using a Support Vector Machine (SVM). Generally, most extractive summarizers do not take context information into account, but do take into account the redundancy over the entire summarization. However, there must exist relationships among the extracted sentences. Actually, we can observe these relationships as consecutiveness among the sentences. We deal with this consecutiveness by using dynamic and difference features to decide if a sentence needs to be extracted or not. Since important sentences tend to be extracted consecutively, we just used the decision made for the previous sentence as the dynamic feature. We used the differences between the current and previous feature values for the difference feature, since adjacent sentences in a block of important ones should have similar feature values to each other, where as, there should be a larger difference in the feature values between an important sentence and an unimportant one. We also present a way to ensure that no redundant summarization occurs. Experimental results on a Corpus of Japanese classroom Lecture Contents (CJLC) showed that the dynamic and difference features were complementary, and our proposed method outperformed traditional methods, which did not take context information into account.
Bibliographic reference. Fujii, Yasuhisa / Yamamoto, Kazumasa / Kitaoka, Norihide / Nakagawa, Seiichi (2008): "Class lecture summarization taking into account consecutiveness of important sentences", In INTERSPEECH-2008, 2438-2441.