Abstractive summarizers trained on single referencesummaries may struggle to produce output that achieve multiple desirableproperties . We propose a new approach to generate multiple variants of the target summary . We then score and select admissible ones according to users’ needs . Our generator gives a precise control over the length of the summary, which is especially well-suited when space is limited . Our selectors are designed topredict the optimal summary length and put special emphasis on faithfulness to the original text . Both stages can be effectively trained, optimized andevaluated. Our experiments on benchmark summarization datasets suggest that this paradigm can achieve state-of-the-art performance. It can also be used to help users find the most important information .
Author(s) : Kaiqiang Song, Bingqing Wang, Zhe Feng, Fei LiuLinks : PDF - Abstract
Code :
Keywords : summary - length - users - approach - multiple -