A reference implementation for Learning to Curate Context: Jointly Optimizing Retrieval and Prediction for Multimodal Social Media Popularity (AAAI 2026).
data/ # ICIP, SMPD, Instagram
data/README.md
src/config/config.yaml
src/models/
src/utils/
src/train.py
src/test.py
requirements.txt
See data/README.md for more data descriptions. Datasets download link: Google Drive.
Use the project Conda environment:
conda create -n jrpp python=3.12
conda activate jrppInstall PyTorch and required packages:
pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu126
pip install -r requirements.txtThe PyTorch version, CUDA version, and random seed may affect model performance. Please use the recommended environment for reproducibility. Verify the environment:
python -c "import torch, pandas; print(torch.__version__); print(torch.cuda.is_available())"python src/train.py --data-name ICIP --run-name jrpp
python src/train.py --data-name SMPD --run-name jrpp
python src/train.py --data-name Instagram --run-name jrpppython src/test.py --data-name ICIP --model-path results/ICIP/jrpp/JRPP_best.pt
python src/test.py --data-name SMPD --model-path results/SMPD/jrpp/JRPP_best.pt
python src/test.py --data-name Instagram --model-path results/Instagram/jrpp/JRPP_best.ptMost method parameters live in src/config/config.yaml. For training and evaluation parameters, run python src/train.py --help or python src/test.py --help.
@inproceedings{xu2026learning,
title = {Learning to Curate Context: Jointly Optimizing Retrieval and Prediction for Multimodal Social Media Popularity},
author = {Xovee Xu and Shuojun Lin and Fan Zhou and Jingkuan Song},
booktitle = {AAAI Conference on Artificial Intelligence (AAAI)},
year = {2026},
volume = {40},
number = {2},
month = {jan},
numpages = {9},
pages = {1382--1390},
publisher = {AAAI},
doi = {10.1609/aaai.v40i2.37112}
}MIT
xovee at uestc.edu.cn