Unsupervised Text Recap Extraction for TV Series

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1797–1806, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Unsupervised Text Recap Extraction for TV Series Hongliang Yu and Shikun Zhang and Louis-Philippe Morency Language Technologies Institute Carnegie Mellon University Pittsburgh, PA, 15213, USA {yuhongliang, shikunz, morency}@cs.cmu.edu Abstract Sequences found at the beginning of TV shows help the audience absorb the essence of previous episodes, and grab their attention with upcoming plots. In this paper, we pro- pose a novel task, text recap extraction. Com- pared with conventional summarization, text recap extraction captures the duality of sum- marization and plot contingency between ad- jacent episodes. We present a new dataset, TVRecap, for text recap extraction on TV shows. We propose an unsupervised model that identifies text recaps based on plot de- scriptions. We introduce two contingency fac- tors, concept coverage and sparse reconstruc- tion, that encourage recaps to prompt the up- coming story development. We also propose a multi-view extension of our model which can incorporate dialogues and synopses. We con- duct extensive experiments on TVRecap, and conclude that our model outperforms summa- rization approaches. 1 Introduction According to a study by FX Networks, in U.S., the total number of ongoing scripted TV series hit a new high of 409 on broadcast, cable, and streaming in 20151. Such a large number indicates there are more shows than anyone can realistically watch. To attract prospective audiences as well as help current view- ers recall the key plot when airing new episodes, some TV shows add a clip montage, which is called a recap sequence, at the beginning of new episodes or seasons. Recaps not only help the audience 1http://tinyurl.com/jugyyu2 absorb the essence of previous episodes, but also grab people’s attention with upcoming plots. How- ever, creating those recaps for every newly aired episode is labor-intensive and time-consuming. To our advantage, there are many textual scripts freely available online which describe the events and ac- tions happening during the TV show episodes2. These textual scripts contain plot descriptions of the events, dialogues of the actors, and sometimes also the synopsis summarizing the whole episode. These abundant textual resources enable us to study a novel, yet challenging task: automatic text recap extraction, illustrated in Figure 1. The goal of text recap extraction is to identify seg- ments from scripts which both summarize the cur- rent episode and prompt the story development of the next episode. This unique task brings new technical challenges as it goes beyond summariz- ing prior TV episodes, by introducing a concept of plot contingency to the upcoming TV episode. It differs from conventional summarization techniques which do not consider the interconnectivity between neighboring episodes. Text recaps should capture the duality of summarization and plot contingency between neighboring episodes. To our knowledge, no dataset exists to study this research topic. In this paper, we present an unsupervised model to automatically extrapolate text recaps of TV shows from plot descriptions. Since we assume recaps should cover …

Content truncated to 500 words