ConDigSum

Official

GitHub - Junpliu/ConDigSum: Code for EMNLP 2021 paper "Topic-Aware Contrastive Learning for Abstractive Dialogue Summarization"

Junpeng Liu, Yanyan Zou, Hainan Zhang, Hongshen Chen, Zhuoye Ding, Caixia Yuan, Xiaojie Wang EMNLP 2021 Paper Conda is highly recommended to manage your Python environment. pip install --editable ./ pip install requests rouge==1.0.0 pip install transformers==4.4.0 bert-score==0.3.8 Before training ConDigSum, please download BART-Large from here, and update PRETRAIN_PATH to the path of model.pt in training scripts.

https://github.com/Junpliu/ConDigSum

Summary

To capture the various topic information of a conversation and outline salient facts for the captured topics, this work proposes two topic-aware contrastive learning objectives, namely coherence detection and sub-summary generation objectives, which are expected to implicitly model the topic change and handle information scattering challenges for the dialogue summarization task.

The proposed contrastive objectives are framed as auxiliary tasks for the primary dialogue summarization task, united via an alternative parameter updating strategy.

Architecture

Pre-Training

Experiments

Home Page Top Stories

Top stories in the U.S. and world news, politics, health, science, business, music, arts and culture. Nonprofit journalism with a mission. This is NPR.

https://www.npr.org/

Performance

GitHub - pltrdy/files2rouge: Calculating ROUGE score between two files (line-by-line)

Given two files with the same number of lines, files2rouge calculates the average ROUGE scores of each sequence (=line). Each sequence may contain multiple sentences. In this case, the end of sentence string must be passed using the --eos flag (default: "."). Running files2rouge with a wrong eos delimiter may lead to incorrect ROUGE-L score.

https://github.com/pltrdy/files2rouge

Further Readings

The Beginner's Guide to Contrastive Learning

And in case you landed here looking for a tool to label your machine learning data or train a model, check out: Let's dive in. Contrastive Learning is a Machine Learning paradigm where unlabeled data points are juxtaposed against each other to teach a model which points are similar and which are different.

https://www.v7labs.com/blog/contrastive-learning-guide