SummPip

This code is for Sigir 2020 paper SummPip: Unsupervised Multi-Document Summarization with Sentence Graph Compression

Python version: this code is in Python3.6

Dataset

source data which has minimal text pre-processing

target data (for evaluation)

Test SummPip

Step1: place downloaded dataset in the folder ./dataset/multi_news/.

Step2: download the pre-trained word2vec model and place it in the folder ./word_vec/multi_news.

• If you want to run SummPip on your own dataset, you need to pre-train a W2V model yourself first with gensim.

Step3: Unsupervised Extractive Summarisation

python run_main.py

• You may want to change -nb_clusters and -nb_words to control the length of the output summary when applying SummPip on your own dataset.