[Story Generation Study 보충학습 : Story Generation Survey Paper] Automatic Story Generation: A Survey of Approaches 리뷰

 

 

 

 

[Story Generation Study 보충학습 : Story Generation Survey Paper]
Automatic Story Generation: A Survey of Approaches 리뷰 

 

 

 

[목차]

1. Definitions 
2. Models
    2-1.Structural Models 
    2-2.Planning based Models 
    2-3.ML Models
3. KNOWLEDGE SOURCES FOR STORYTELLING
4. TOWARD INTERESTING STORIES
5. STORY EVALUATION
6. DISCUSSION (Limitations)
7. Future Works 

 

 

 

어떤 새로운 연구 분야를 접할 때, Survey 논문을 먼저 읽어보는 것을 좋아해서 이번에도 읽게 되었다...

워낙 양이 많긴 하지만, 큰 틀을 잡아놓고 공부하는 것이 이해에 도움이 되어서 정리해보고자 한다 ! 

 

 

 

Definitions 

 

• Story
• Story Event
• Plot Graph
• Story Frame

 

 

 

Models 

 

 

 

1. Structural Models 

 

1) Graph-Based Approaches

 

 

 

 

 

 

 

 

 

2) Grammar-Based Approaches

 

 

 

 

 

 

---

 

 

2. Planning based Models 

 

1) Goal-Directed Approaches

 

 

 

2) Analogy-Based Approaches

 

 

 

3) Heuristic Search Approaches

 

 

---

3. ML Models

 

 

1) Story Abstraction

 

 

 

2) Script Learning and Generation

 

 

 

 

 

3) Story Completion

 

 

 

 

4) Story Generation

 

 

 

 

 

 

 

 

 

 

KNOWLEDGE SOURCES FOR STORYTELLING

 

 

1. Types of Knowledge for Story Generation

 

• Thematic knowledge: Settings of the story are identified such as time, place, and objects. This
  knowledge should also describe the story world concepts, their properties at a certain
  point in time, and their relationships.
• Characters knowledge: A complex and extensive representation of intelligent characters including
  characters’ goals, physical state, personality, and emotional relationships.
• Plot knowledge: Includes the knowledge needed to construct the story plot, such as agents,
  events, goals, and actions, and how these components are linked.
• Linguistic knowledge: Used for representing the story in specific linguistic structures to present
  it to the reader in natural language.
• Literary knowledge: Incorporates principles of storytelling in literature designed to increase
  story interestingness.
• Feedback knowledge: Effects of story fragments on the user are collected to predict the audience’s
  response to new stories.

 

 

2. Story and Semantic Relationships Corpora

 

 

 

3. Crowdsourcing Knowledge

 

 

 

 

 

4. Commonsense Knowledge

 

 

 

 

 

 

 

TOWARD INTERESTING STORIES

 

• Story Suspense
  
  • Conflict
  • Uncertainty
  • Emotions
• Discourse
• Characters
• Dialogue
• Narrative text

 

 

 

 

STORY EVALUATION

 

 

 

 

 

 

 

DISCUSSION (Limitations)

 

• Dispersion
• Domain knowledge
  • Story corpora
  • Crowdsourcing
  • Commonsense knowledge
  • Events semantic relations
• Seq2Seq models
  • Losing consistency
  • Repetitive words
  • OOV problem
• Pre-trained language models
• Story interestingness
• Objective evaluation

 

 

 

 

Future Works 

 

• Decomposition
• Deep learning
• Hybrid systems
• Automatic evaluation
• Benchmarking

 

 

 

Reference

 

 

 

https://dl.acm.org/doi/pdf/10.1145/3453156

Automatic Story Generation: A Survey of Approaches: ACM Computing Surveys: Vol 54, No 5