文章链接:https://arxiv.org/pdf/1811.05701.pdf
会议: AAAI 2019
Abstract
先是讲述之前的工作,要没实在情节规划方面做限制,要么是只能在狭窄的领域生成故事,最后提出自己的工作。
We propose a plan-and-write hierarchical generation framework that
first plans a storyline, and then generates a story based on the
storyline.
Introduction
先是讲述之前的一些工作,而这些工作主要关注在两个方面:1、情节规划 2、基于案例推理(case-based reasoning),同时也讲述这些方法的缺点:太过依赖于人功能注释,或者领域比较局限。
提出自己的故事生成框架(Story Generation)
we propose to decompose story generation into two steps: 1) story planning which generates plots,and 2) surface realization which composes natural language text based on the plots.
Problem Formulation
## Method
*Dynamic Schema
*:通俗理解就是边生成故事片段(sentence),边规划(生成)Storyline中的next word。
Storyline Planning:动态规划的方法生成Storyline是将该任务看做成一个content-introducing generation problem (We implement the content-introducing method proposed by Yao et al. [2017],)
输入的是:context (the title and previously generated sentences are taken as context) + the previous word in storyline
Context = ctx =[t,s1:i−1]
Story Generation : 同样被看做是一个content-introducing generation problem 生成器结构和Storyline Planning一样 优化方程为:
Static Schema
先根据tiitle生成一个完整的Storyline(牺牲了了灵活性,加强了连贯性),然后再生成故事。
Storyline Planning:使用Seq2Seq模型
Story Generation:
其他就看文章内容吧