Refining GPT-3 Embeddings with a Siamese Structure for Technical Post Duplicate Detection

Datasets

Please check the datasets folder for details.

Installation

To use this project, you need to have the following dependencies installed:

• Python 3.7 or higher
• Some Python libraries (Specified in requirements.txt)

Clone this repository to your local machine:

git clone https://github.com/mooselab/suppmaterial-PostDupGPT3.git

You can install the Python libraries using pip:

cd ./suppmaterial-PostDupGPT3/src
pip install -r requirements.txt

Model Training

This replication package contains a tiny sample dataset for testing the codes.

Train with Triplet loss

cd ./src
python ./train_triplet_loss.py

Train with MNR loss

cd ./src
python ./train_MNR_loss.py

DupPredictor

We re-implemented the DupPredictor. The re-implementation is under DupPredictor_ReImp folder.

A dataset use to do the grid search for parameters and CQADupStack testsets of nine domains are included.

Usage:

1. Install the requirements:

cd ./DupPredictor_ReImp
pip install -r requirements.txt

2. Search for the best parameters for composer:

python ./DupPred_param_search.py

This process involves the training of the topic model. The trained model will be saved and used in the evaluation process.

3. Predict with testsets:

python DupPredictor.py

By running it, all the nine sub-domains will be iterated.

Tools

The Tools folder contains codes used to generate the GPT-3 embeddings and print the prediction results.

Citation

Xingfang Wu, Heng Li, Nobukazu Yoshioka, Hironori Washizaki, Foutse Khomh, Refining GPT-3 Embeddings with a Siamese Structure for Technical Post Duplicate Detection, Proceedings of the 31st IEEE International Conference on Software Analysis, Evolution, and Reengineering (SANER), March 12 - 15, 2024, Rovaniemi, Finland, IEEE.

License

This project is licensed under the MIT License.