Getting Inspiration for Feature Elicitation: App Store- vs. LLM-based Approach

About the project

The code of LLM-inspired approach is in llm.py. The code of AppStore-inspired approach is in gp.py.

The description folder contains the code for creating the app description vector database.

The features we used for evaluation can be found in the evaluation folder.

Note about vector database

During our evaluation, we used Qdrant as the local vector database to retrieve relevant app descriptions in AppStore-Inspiration. However, due to the size limitations, we cannot provide the storage file for the vector database. Therefore, we have commented out the code related to Qdrant.

As an alternative, the tool uses the Google Play search engine to find relevant app descriptions. Please note that this alternative has lower performance compared to using our vector database.

Getting Started

1. Install poetry (link)

2. Install dependencies

poetry install

3. Set environ variable OPENAI_API_KEY

export OPENAI_API_KEY="your api key"

Usage

Launch the inspiration service

uvicorn server:app --port 12345

Play with piStar

Open piStar/tool/index.html. piStar is an open-source goal modelling tool (link).

Citation

If you find our work useful, please cite our paper:

@inproceedings{Wei:GettingInspirationFeature:2024,
	title = {Getting Inspiration for Feature Elicitation: App Store- vs. LLM-based Approach},
	author = {Wei, Jialiang and Courbis, Anne-Lise and Lambolais, Thomas and Xu, Binbin and Bernard, Pierre Louis and Dray, Gérard and Maalej, Walid},
	booktitle = {39th IEEE/ACM International Conference on Automated Software Engineering (ASE'24)},
	year = {2024},
	doi = {10.1145/3691620.3695591},
	publisher = {ACM}
}