Who’s important? — SUnSET: Synergistic Understanding of Stakeholder, Events and Time for Timeline Generation

Install

Installing necessary dependency for ROUGE score calculation. For Ubuntu/Debian:

apt-get update
apt-get install libxml-parser-perl

For CentOS/RHEL/Fedora:

yum install perl-XML-Parser

Install Python environment via Conda:

conda create -n sunset_env python==3.10
conda activate sunset_env
pip install -r requirements.txt

Dataset

To download the original dataset, please refer to complementizer/news-tls.

All SET data of T17 and Crisis (after KG resolution) are available here.

Steps

With the SET data provided, you can start from Step 1 (skip the SET extraction and KG resolution steps).

If you want to start from scratch, unfold and run the following parts.

-2. SET extraction

-1. KG Construction & Resolution

1. Create vector database

python src/create_milvus_db.py \
    --model_path $EMBEDDING_MODEL_PATH \
    --client_name $DATASET.db \
    --dataset $DATASET \
    --input_path $RAW_DATA_PATH \
    --set_path $SET_DATA_PATH

2. Run Preprocessing

python src/preprocessing.py \
    --dataset $DATASET \
    --input_path $RAW_DATA_PATH \
    --output_path $INPUT_DATA_PATH

3. Run SET Matching

python src/set_matching.py \
    --model_path $EMBEDDING_MODEL_PATH \
    --dataset $DATASET \
    --gt_path $RAW_DATA_PATH \
    --scoring_path $SCORING_PATH \
    --input_path $INPUT_DATA_PATH/timeline_outputs \
    --output_path $INPUT_DATA_PATH/timeline_outputs \
    --min_common_stake "$stake" \
    --beta "$beta"

4. Run Timeline Generation & Evaluation

python src/cluster_tls_eval.py \
    --text_rank \
    --dataset $DATASET \
    --model_path $EMBEDDING_MODEL_PATH \
    --scoring_path $SCORING_PATH \
    --raw_data_path $RAW_DATA_PATH \
    --info_save_path $INPUT_DATA_PATH/info_outputs \
    --timelines_path $INPUT_DATA_PATH/timeline_outputs/$DATASET \
    --events_path $INPUT_DATA_PATH/event_outputs/$DATASET \
    --output_path $INPUT_DATA_PATH/result/$DATASET \
    --beta $beta