AbsMeetingSummarization

Abstractive Meeting Summarization

1 Dataset

This dataset is extracted from AMI meeting, the dataset folder is organized as follows:

data/
    ami-dataset.pkl  # AMIDataset instance
    ami-vocab.pkl    # Vocabulary
    train.json       # training data
    valid.json       # validation data
    test.json        # testing data

The description of records in *.json is as follows:

{
    "agents": "Who is the speaker in one sentence",
    "st_times": "Start time of one sentence",
    "end_times": "End time of one sentence",
    "article_sents": "Sentences of this meeting", 
    "article_pos_tags": "Pos tags for each token",
    "article_ner_tags": "Ner tags for each token", 
    "article_tf_tags": "Tf tags for each token", 
    "article_idf_tags": "IDF tags for each token", 
    "summary_sents": "Sentences of summary of this meeting", 
    "summary_pos_tags": "Pos tags for each token for each summary",
    "summary_ner_tags": "Ner tags for each token for each summary"
}

2. Requirements

Need install rouge in your linux environmentlink. And then

$ pip install pyrouge

2 Training

cd AbsMeetingSummarization
bash scripts/run_HAS.sh

3 Testing

Build a directory named ```ckp`` and move your saved model into it. The directory may look like:

ckp/
    models/
        your_model.pt

python test.py --ckp-path checkpoints/param_tuning/ --report_score