Inconsistencies between language tags and actual video content

#9
by ysdede - opened

Regarding the Turkish content in the dataset:
Based on the analysis of the first 8 million rows (see: https://huggingface.co/datasets/PleIAs/YouTube-Commons/discussions/7#66432f1f7de474fd56f9c9b6)

There are only 8 rows where the transcription_language is listed as 'tr' (Turkish).
However, there are 15,600 rows where the original_language is marked as 'tr' (Turkish). Upon further inspection, it appears that most of the videos in these rows are not actually in the Turkish language, despite being labeled as such.

This discrepancy in the language tags and the actual content of the videos highlights potential issues or inaccuracies in the dataset's metadata, particularly concerning the Turkish content.

image.png

ysdede changed discussion title from This discrepancy in the language tags and the actual content to Discrepancy in the language tags and the actual content
ysdede changed discussion title from Discrepancy in the language tags and the actual content to Inconsistencies between language tags and actual video content

Sign up or log in to comment