This repository is the official implementation of CoNCRA: A Convolutional Neural Network Code Retrieval Approach.
Our source code its an adaptation of: https://github.com/codekansas/keras-language-modeling
We propose a technique for semantic code search: A Convolutional Neural Network approach to code retrieval (CoNCRA). Our technique aims to find the code snippet that most closely matches the developer's intent, expressed in natural language. We evaluated our approach's efficacy on a dataset composed of questions and code snippets collected from Stack Overflow. Our preliminary results showed that our technique, which prioritizes local interactions (words nearby), improved the state-of-the-art (SOTA) by 5% on average, retrieving the most relevant code snippets in the top 3 (three) positions by almost 80% of the time.
We ran our experiments at Google colab. The notebooks and source code to run our models is available at notebooks folder.
To train the model(s) in the paper, execute the following notebooks:
- train_cnn_stack_over_flow_qa.ipynb
- train_shared_cnn_stack_over_flow_qa.ipynb
- train_unif_embedding_stack_over_flow_qa.ipynb
To evaluate my model on StaQC Dataset, run the following notebooks:
- evaluate_best_stack_over_flow_qa.ipynb
- evaluate_stack_over_flow_qa.ipynb
You can download pretrained models here:
- CoNCRA Model trained on StaQC Dataset using margin 0.05, 4000 filters and kernel size 2.
Our model achieves the following performance on :
| Model name | MRR | Top 1 Accuracy |
|---|---|---|
| CoNCRA Model | 0.701 | 57.7% |
| Unif | 0.675 | 53.9% |
| Embedding Model | 0.637 | 49.3% |