PDSum: Prototype-driven Continuous Summarization of Evolving Multi-document Sets Stream

Presented at WWW'23 [Paper]

Task: EMDS (Evolving Multi-document Sets Stream Summarization)

How to run

Packages

See requirements.txt (pip install -r requirements.txt)

Datasets

Go to link for the preprocessed data sets and put them under a "datasets" folder (The original datasets can be found from WCEP and W2E)

Sample (WCEP)

For your own dataset, please preprocess the dataset in JSON file with the following columns

• date: the publication date of a document (the default temporal context of summaries)
• sentences: [s1, s2, .., sn] - a list of sentences in a document
• sentence_tokens: [[t1, t2, ..tn],[t1,...]] - lists of tokens in each sentence
• sentence_counts: the number of sentences in a document
• Query: the identifier for a document set (e.g., type/category/query)

Usage

Refer to PDSum_run_example.ipynb and set the datasets and hyperparameters (default values are given), and run the notebook as instructed.

Hyperparameters

• dataset = 'Default' (summaries are returned at Default: every unique date (e.g., WCEP) or Custom: at every true summary date (e.g., W2E))
• max_sentences = 1 (the number of sentences in summaries)
• max_tokens = 40 (the number of tokens in summaries)
• batch = 64, epoch= 5, temp = 0.2, lr = 1e-5 (training settings)
• D_in = 1024, D_hidden = 1024, head = 2, dropout = 0 (model settings)
• N = 10 (the number of keywords)
• distill_ratio = 0.5 (the ratio of distilling knowledge from previous documents)

Cite

@inproceedings{pdsum,
  title={PDSum: Prototype-driven Continuous Summarization of Evolving Multi-document Sets Stream},
  author={Yoon, Susik and Chan, Hou Pong and Han, Jiawei},
  booktitle={Proceedings of the ACM Web Conference 2023},
  pages={1650--1661},
  year={2023}
}