Repo for The Synthetic Party (Det Syntetiske Parti), made by party secretary Computer Lars.
This GitHub repository hosts an extensive collection of datasets and theoretical references, emotion and Valence-Arousal-Dominance (VAD) classifications, visualizations, analyses, and supplementary materials used to represent The Synthetic Party of Denmark.
The project challenges traditional notions of political representation by leveraging machine learning to interpret and articulate constituency sentiments. The repository of The Synthetic Party's data material should be of interest to those working in the intersection of political theory, artificial intelligence, and artistic research.
-
- Clean datasets
- Emotion and VAD Classifications
- Visual Analysis
-
- Contributing
- License
- Contact
Contains the following sub-folders with documents and files.
- Literature references: List of references for theory included in the exposition.
- Textual Basis for Scripts: Detailed written material explaining the concept and operations of The Synthetic Party.
- Video material used in the exposition scripts (excerpt).
The datasets included in this repository are comprehensive collections of data pertaining to The Synthetic Party's operation. These datasets include:
Clean Datasets
- Theory of The Synthetic Party: Insights into the philosophical and political underpinnings of The Synthetic Party.
- Training Data: The data used to finetune the AI models, including textual data from 200+ Danish micro-parties as found within their founding documents, online deliberations, websites, and media coverage.
- AI Party Program: The text-generated party program of The Synthetic Party, which is based on a finetuned GPT 3-model by OpenAI. Training data was used to finetune. Dataset is inferred from the party's publications on Medium.
- Online Deliberation Records: Transcriptions and records of online discussions and deliberations between humans and AI, taking place on The Synthetic Party's Discord-server. The text generated by chatbot is based on the model GPT-NEO-20Xb by EleutherAI.
- News Media Coverage: A curated collection of international media coverage of The Synthetic Party, highlighting its societal impact and public reception.
Emotion and VAD Classification
This section contains the above datasets classified according to various basic emotions and affective models. These classifications provide insights into the affective and emotional dimensions of the data, offering an understanding of public sentiment within discourse. The classification is executed through several models, each lending a perspective on the data:
- Text2Emotion: A PyPi Library. Uses a lexical bag-of-words approach to infer six basic emotions from text, such as happiness, sadness, anger, etc.
- BERT Danish: Large language model fine-tuned on Danish SoMe material, which classifies eight basic emotions.
- GPT 3.5: Generative large language model by OpenAI to provide affective VAD (Valence, Arousal, Dominance) analysis, map data onto the multidimensional valence-arousal-dominance spectrum, and correlate with new emotion classification.
Visual Analysis
This section presents a range of analyses performed on the datasets, including:
- Sentiment Analysis: Classifying and evaluating affective and emotional content within the political discourse.
- Thematic Analysis: Exploring recurrent themes and topics in the data.
- Comparative Analysis: Juxtaposing different data sets to uncover underlying patterns and relationships.
A series of visualizations are included to offer interpretative insights into the datasets, including but not limited to:
- Sentiment Bars: Illustrating the basic constituency sentiment.
- Emotion Bar Graphs: Visualizing the emotional spectre of dataset.
- Cubes: Displaying a multidimensional distribution of data across the valence-arousal-dominance spectrum.
- Network Diagrams: Showing the interconnections between different elements within the datasets.
- Keywords maps: Showing themes and topical debates
Visualizations are created by Matplotlib, where:
- The matplotlib.pyplot.bar function creates the bar graphs illustrating the categorical aspect of emotions.
- The mpl_toolkits.mplot3d toolkit is utilized for dimensional representation to generate 3D cubes depicting valence, arousal, and dominance.
This repository is intended for researchers, political theorists, data scientists, and artists interested in the intersection of AI, democracy, and art. To use the datasets and materials:
Clone the Repository: Use Git or checkout with SVN using the web URL: git clone https://github.com/YourRepository/thesyntheticparty.git
Explore the Contents: Navigate through the directories to find datasets, visualizations, and analyses.
Contributing
Contributions to this project are welcome. To contribute:
- Fork the Repository (click the "Fork" button located at the top right corner of the repository page)
- Create your Feature Branch (git checkout -b feature/YourFeature)
- Commit your Changes (git commit -m 'Add some YourFeature')
- Push to the Branch (git push origin feature/YourFeature)
- Open a Pull Request
License
This project is licensed under the Creative Commons - see the LICENSE.md file for details.
For inquiries related to The Synthetic Party, please contact party figurehead Leader Lars, artist-researcher Asker Bryld Staunæs or party secretary Computer Lars.