CAMERED

This repo is the released code of our SIGIR 2025 work Collaboration and Controversy Among Experts: Rumor Early Detection by Tuning a Comment Generator and our extended paper Multi-View Controversy Among Multi-Expert Comments for Rumor Early Detection.

0 - Requirements

torch==2.1.0
cudatoolkit==12.5
transformers==4.36.2

1 - Train and Generate Comments

1. Prepare data from their original papers, and put them in ./data/.
2. Process Twitter and Weibo data, e.g.,

cd generation
python data_twitter_preprocess.py --input_folder "../data/twitter15/" --output_folder "../data/twitter15/"
python data_weibo_preprocess.py --input_folder "../data/weibo16/" --output_folder "../data/weibo16/"

3. Extract entities from data, and obtain their entity descriptions, e.g.,

python twitter_entity_extraction.py --input_file "../data/twitter15/train.json" --output_file "../data/twitter15/entity_description.json" --llm_model "../gpt-2/"
python weibo_entity_extraction.py --input_file "../data/weibo16/train.json" --output_file "../data/weibo16/entity_description.json" --llm_model "../chinese-t5-base/"

4. Train the generator, e.g.,

python train_generator.py

5. Generate comments, e.g.,

python generator_comments.py --input_json_path "../data/twitter16/test.json" --output_tsv_path "../data/twitter16/generated_comments.tsv" --model_path "../flan-t5-base/" --linear_layer_path "../data/twitter16/gan/final_generator/" --num_experts 10 --K 16 --threshold 0.5

2 - Train Detection Models

1. Prepare generated data, put the generated comments in ./data/[DATASET]/generated, and rename them as generated_train.json (we provide a test case in the folder)
2. Train and test detection models, e.g.,

cd detection
python main.py --dataset [DATASET] --model [MODEL]

DATASET in twitter15, twitter16, weibo16, weibo20, and MODEL in cbert, defend, casfend, kahan, cameredit

3. The evaluation metrics are saved in ./detection/json/[DATASET], and you can run ./read_json.py to automatically read the json files into an excel table

Citation

@inproceedings{wang2025collaboration,
  author       = {Bing Wang and
                  Bingrui Zhao and
                  Ximing Li and
                  Changchun Li and
                  Wanfu Gao and
                  Shengsheng Wang},
  title        = {Collaboration and Controversy Among Experts: Rumor Early Detection by Tuning a Comment Generator},
  booktitle    = {International {ACM} {SIGIR} Conference on Research and Development in Information Retrieval},
  pages        = {468--478},
  publisher    = {{ACM}},
  year         = {2025},
  doi          = {10.1145/3726302.3729928},
}