ECS 289G Project
University of California Davis
Our storytelling system employs a GPT-2 instance to generate the initial portion of an adventure story based on a dataset-provided prompt. To ensure narrative coherence and dramatic progression, the generated story segment is assigned a drama score derived from the sentiment analysis of the current scene. The drama score then informs the generation of suitable input options that affect plot intensity and narrative drive. Subsequently, a human player selects one of these input options, and the chosen input, along with the story history, is fed back into the GPT-2 instance to further develop the narrative. Through iterative cycles of AI-generated content and human input, our system facilitates the co-creation of engaging, customized stories that cater to individual preferences while maintaining a consistent narrative structure.
- GPT-2 : to generate text
- NLTK : language processing into sentences
- TextBlob : to generate sentiment score of a text
- Hugging Face transformers: to generate user choice from text
- BLEU, Perplexity : to assign a score evaluating the quality of text
Graphical representation of sentiment score value progression
Clone the repo using "git clone https://github.com/aditya-parab/Adventures-of-AI-game.git".
Run the "ecs-289g.ipynb" python notebook. Remember to add/change the initial prompt in cell number 4 before running the code.
Going throught the code, you will be able to edit and use multiple different GPT-2 models. The code is self-explanatory.