PCLMM: A multimodal dataset for Patronizing and Condescending Language detection

The PCLMM dataset is sourced from Bilibili, the largest online community for young people in China. Our work aims to uncover microaggressions targeted at vulnerable groups, including discriminatory and patronizing language expressions, as well as facial expression features. PCLMM contains six main vulnerable communities :

Disabled individuals

Women

The elderly

Children

Single-parent families

Low-income groups

The PCLMM dataset contains 715 videos, totaling 21 hours of content, with an average video length of 1.80 mins and a frame rate of 30 FPS, comprising 2.3M frames. Approximately 27.4% of the videos were labeled as patronizing (label-1), aligning with the distribution of PCL data on internet platforms.

If you would like to learn more about our PCL field (a branch of toxic speech detection) and PCLMM dataset, please refer to this paper, accepted by ICASSP 2025: https://arxiv.org/abs/2409.05005

The code implementation of the paper can be found in our repository: https://github.com/dut-laowang/PCLMM

Updation

February 9, 2025 – The Annotation_Link.csv containing the original video links has been updated.