VALUEscope is a computational framework that leverages language models to quantify social norms and values within online communities. The project is grounded in social science perspectives on normative structures and utilizes a novel Return Potential Model (RPM) to dissect and analyze linguistic and stylistic expressions across various digital platforms.
valueScope/
│
├── norm_induction/ # Scripts and data for norm induction processes
│ └── data/ # Data and outputs from the norm induction process
│ └── scripts/ # Scripts related to surface important norms
├── norm_prediction/ # Scripts for training norm prediction models and using them for binary label generation
│ └── human-annotation-labels/ # JSON files of 4 topics for 450 human-annotated labels
│ └── synthetic-labels/ # Scripts and prompts for GPT synthetic label generations for classifier training
├── style_transfer/
│ └── scripts/ # Scripts related to style transfer generation
│ └── filter/ # Scripts related to style transfer filter
├── upvote_prediction/
│ └── scripts/ # Scripts for predicting upvotes based on community norms
├── .gitignore # Specifies intentionally untracked files to ignore
└── README.md # This README file
To set up the VALUEscope framework on your local machine, follow these steps:
- Clone the Repository:
git clone https://github.com/stellali7/valueScope.git
- Navigate to the repository directory:
cd valueScope
The framework includes several components that can be run separately. Instructions for running each component are provided in their respective directories.
This project is licensed under the MIT License.
If you use VALUEscope in your research, please consider citing our work:
@misc{park2024valuescope,
title={ValueScope: Unveiling Implicit Norms and Values via Return Potential Model of Social Interactions},
author={Chan Young Park and Shuyue Stella Li and Hayoung Jung and Svitlana Volkova and Tanushree Mitra and David Jurgens and Yulia Tsvetkov},
year={2024},
eprint={2407.02472},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2407.02472},
}