On-line Summarization of Time-series Documents using a Graph-based Algorithm

PACLIC 28 !470 On-line Summarization of Time-series Documents using a Graph-based Algorithm Satoko Suzuki Graduate School of Humanities and Sciences, Ochanomizu University 2-1-1 Otsuka Bunkyo-ku Tokyo,Japan suzuki.satoko@is.ocha.ac.jp Ichiro Kobayashi Graduate School of Humanities and Sciences, Ochanomizu University 2-1-1 Otsuka Bunkyo-ku Tokyo,Japan koba@is.ocha.ac.jp Abstract As enormous amount of electronic documents on the Web have been increasing, the ne- cessity of automatic summarization has also been increasing to help people grasp the essen- tial points of the documents. Many summa- rization techniques dealing with single doc- ument and multi-documents have been stud- ied. However, due to the increase of the documents which report the change of top- ics along a timeline, called time-series docu- ments, in recent years, a summarization tech- nique which generates a summary of time- series documents, called timeline summariza- tion, has been actively studied as an area of automatic summarization. There are different difficulties in summarizing time-series doc- uments from other type of automatic sum- marization. The basic approach for timeline summarization is to extract sentences which describe major events in object documents in chronological order to generate a timeline summary. However, unlike the prior studies of timeline summarization, we particularly focus on on- line summarization of time-series documents and propose an on-line graph-based timeline summarization method. With our proposed method, a summary of time-series documents can be generated at any point of time when it is required. We conduct experiments to investi- gate the ability of our proposed method, evalu- ate the results with ROUGE metrics, and show our proposed method produces a better sum- mary compared to other representative sum- marization methods. 1 Introduction The automatic summarization techniques have been required due to increasing the amount of electronic documents. The object documents handled by au- tomatic summarization are diverse from newspa- per and academic articles to the documents used in social network services such as Weblog, Twit- ter, etc. Depending on object documents, an appro- priate summarization technique is applied. Due to the increase of electronic documents updated day by day, in recent years, new techniques which summa- rize time-series documents, called timeline summa- rization, have been actively studied. There are dif- ferent difficulties in summarizing time-series docu- ments from other type of automatic summarization, because as for timeline summarization we have to summarize current information taking account of the information from the past. Besides, we have to de- cide the important information which should be in- cluded in a summary, tracking the change of topics. The basic approach for timeline summarization is to extract sentences which describe major events in ob- ject documents in chronological order to generate a timeline summary. In this study we paticularly focus on the develop- ment of an on-line method to summarize time-series documents. To achieve this, we have to deal with various problems, for example, how to combine the current information with the information from the past in order to recompile object information to be summarized, how to track topics, how to rank impor- tant information, etc. To deal …

Content truncated to 500 words