Few-Shot Character Understanding in Movies as an Assessment to Meta-Learning of Theory-of-Mind

Paper | Project Page | Poster

🎞️ Dataset: ToM-in-AMC

Please download the data from Google Drive.

gdown --folder 1y9_zP4zOaP_WPntYQuXJXOdtv72POGg_ -O data
# SHA256: 551a726269cf9d6149d32b7712230111ddd4f43ac8ea461edd9456c8527711ad

and run dataset_overview.ipynb for details.

The dataset consists of ∼1,000 parsed movie scripts from IMSDb, each corresponding to a few-shot character understanding task.

🤖 Models

Our dataset evaluate the machines’ ToM in two settings:

• Transductive setting: meta-model predicts with all characters’ previous acts as examples

• Inductive setting: meta-model predicts with a mental model of characters generated by all characters’ previous acts

  💡 This setting is more stringent and has advantages in emphasizing the effects of various ToM dimensions, improving explanability and mitigating shortcuts.

Inductive Setting: ToMPro

[To Be Cleaned] See /tompro.

Transductive Setting: In-Context Learning

[To Be Cleaned] See /icl.

📝 Citation

@inproceedings{yu2024few,
  title = {Few-Shot Character Understanding in Movies as an Assessment to Meta-Learning of Theory-of-Mind},
  author = {Yu, Mo and Wang, Qiujing and Zhang, Shunchi and Sang, Yisi and Pu, Kangsheng and Wei, Zekai and Wang, Han and Xu, Liyan and Li, Jing and Yu, Yue and Zhou, Jie},
  booktitle = {Proceedings of the 41st International Conference on Machine Learning},
  year = {2024}
}