AspectSum

The source code of the paper SALAS: Supervised Aspect Learning Improves Abstractive Multi-document Summarization Through Aspect Information Loss.

Requirements

• Python (tested on 3.8.13)
• CUDA (tested on 11.4)
• PyTorch (tested on 1.8.0)
• Transformers (tested on 4.6.0)
• numpy (tested on 1.23.2)
• tqdm

Datasets

The MRED dataset can be downloaded from https://github.com/Shen-Chenhui/MReD The WikiAsp dataset can be downloaded from https://github.com/neulab/wikiasp

Data Processing

First, place the data in raw_data. Next, processing the data by

python 1_process_data.py

The processed files are stored in processed_data. Here are some descriptions.

• doc: original document
• doc_with_sent_aspect: aspects with each sentence
• sent_controlled_doc: [label1, label2, ... ], per-sentence label sequence for the meta-review where label1 represents the category label for 1st sentence, label2 for the 2nd sentence and so on
• seg_controlled_doc: [label1, label2, ... ], label sequence for the meta-review on segment level where label1 represents the category label for 1st segment (the sentences of the same label), label2 for the 2nd segment and so on
• summary: Summary of the document
• summary_with_seg_aspect: segment level summaries
• summary_with_sent_aspect: Sentence level summaries
• sample_id: yyyy-id, where yyyy is the year

Training and Evaluation

The training and evaluation are executed by 3_run_our_idea.py, you can replace the pretrained_model_path to use different pre-trained models.

Case Study

Our implementation of the case study is in case_study.py.