Information Retrieval
Advances in Automatic Text Summarization
Advances in Automatic Text Summarization
| Hi-index | 0.00 |
In recent years, the world is filled with a large amount of information through the internet and so on. Such a situation increasingly enhances the worth of the automatic text summarization which can support a quick grasp of the text content. The automatic text summarization has so far been accomplished by extracting some important sentences from a text based on various surface cues. To be compared, we tried to devise a new method to extract the important sentences from the story according to the way in which the people comprehend it. In designing this new method, we took account of the text comprehension model, that is, how people comprehend the story text. Then we devised the procedure for transforming from a set of the propositions to the propositional network. In Experiment 1, the participants were asked to select the sentences regarded as important from five stories. Then we examined how the propositions drawn from each important sentence were connected in the propositional network of each story. As a result, we identified three distinctive connection patterns. In Experiment 2, it was examined whether those connection patterns are valid as the rules to extract the important sentences from five new stories. From the sentences extracted our system and the important sentences selected by the participants, we calculated the aggregation accuracy measures. As a result, it was found that they were clearly higher than the baselines. Moreover they were equal to or higher than ones obtained in the previous researches. This finding was replicated to the stories used in Experiment 1.