license: mit
task_categories:
- audio-classification
- text-classification
language:
- en
tags:
- toxicity
- audio
- podcasts
- diarization
- temporal_toxicity
pretty_name: Toxic Conversation Chains
Toxicity Begets Toxicity: Unraveling Conversational Chains in Political Podcasts
Accepted at ACM Multimedia 2025
Naquee Rizwan, Nayandeep Deb, Sarthak Roy, Vishwajeet Singh Solanki, Kiran Garimella, Animesh Mukherjee
[Paper] || [Arxiv] (Main content + Appendix in one PDF) || Please also follow the [GitHub] link for codes.
Abstract
Tackling toxic behavior in digital communication continues to be a pressing concern for both academics and industry professionals. While significant research has explored toxicity on platforms like social networks and discussion boards, podcasts—despite their rapid rise in popularity—remain relatively understudied in this context. This work seeks to fill that gap by curating a dataset of political podcast transcripts and analyzing them with a focus on conversational structure. Specifically, we investigate how toxicity surfaces and intensifies through sequences of replies within these dialogues, shedding light on the organic patterns by which harmful language can escalate across conversational turns. Warning: Contains potentially abusive/toxic contents.
Dataset
The top 100 toxic conversation chains and their ground truth cpd annotations, each for conservative and liberal podcast channels are present in the GitHub repository [cpd/dataset]. That folder contains:
- two annotation csv files (one each for conservatives and liberals) containing annotations of individual annotators (ex: Annotator_ND) and based on the majority voting as well (refer 'Inter_Annotator'). Further, this file also contains the cpd results as predicted by traditional CPD algorithms (refer [ruptures] library).
- two json files (one each for conservatives and liberals) containing the details of top 100 toxic conversation chains.
Hugging Face
Additionally, here we also provide this [Hugging Face] dataset with:
- audio clips (.wav files) of top 100 toxic conversation chains (for both conservatives and liberals). These files are required to run the audio prompts in [cpd/dataset/audio_prompt_cpd.py]. Note- Please accordingly update the path to folders to make the code working.
- all toxic conversation chains from both, conservative and liberal podcast channels. As stated in the paper, we define a toxic conversation chain whose anchor segment's toxicity value is greater than 0.7.
- complete diarized dataset with toxicity scores calculated using Perspective API for both conservative and liberal podcast channels.
Appendix
ACM MM 2025 did not have the provision of incorporating supplementary material. Hence, we provide it [here].
Please cite our paper
@inproceedings{10.1145/3746027.3754553,
author = {Rizwan, Naquee and Deb, Nayandeep and Roy, Sarthak and Solanki, Vishwajeet Singh and Garimella, Kiran and Mukherjee, Animesh},
title = {Toxicity Begets Toxicity: Unraveling Conversational Chains in Political Podcasts},
year = {2025},
isbn = {9798400720352},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3746027.3754553},
doi = {10.1145/3746027.3754553},
abstract = {Tackling toxic behavior in digital communication continues to be a pressing concern for both academics and industry professionals. While significant research has explored toxicity on platforms like social networks and discussion boards, podcasts-despite their rapid rise in popularity-remain relatively understudied in this context. This work seeks to fill that gap by curating a dataset of political podcast transcripts and analyzing them with a focus on conversational structure. Specifically, we investigate how toxicity surfaces and intensifies through sequences of replies within these dialogues, shedding light on the organic patterns by which harmful language can escalate across conversational turns. Warning: Contains potentially abusive/toxic contents.},
booktitle = {Proceedings of the 33rd ACM International Conference on Multimedia},
pages = {11776–11784},
numpages = {9},
keywords = {change point detection, podcasts, toxic conversation chains, toxicity begets toxicity, transcripts},
location = {Dublin, Ireland},
series = {MM '25}
}
@inproceedings{Rizwan_2025, series={MM ’25},
title={Toxicity Begets Toxicity: Unraveling Conversational Chains in Political Podcasts},
url={http://dx.doi.org/10.1145/3746027.3754553},
DOI={10.1145/3746027.3754553},
booktitle={Proceedings of the 33rd ACM International Conference on Multimedia},
publisher={ACM},
author={Rizwan, Naquee and Deb, Nayandeep and Roy, Sarthak and Solanki, Vishwajeet Singh and Garimella, Kiran and Mukherjee, Animesh},
year={2025},
month=oct, pages={11776–11784},
collection={MM ’25} }
Contact
For any questions or issues, please contact: nrizwan@kgpian.iitkgp.ac.in