Proceedings of the Second Workshop on Natural Language Processing for Internet Freedom: Censorship, Disinformation, and Propaganda
Материалы Второго семинара по обработке естественного языка для свободы Интернета: цензура, дезинформация и пропаганда
2019-01-01
SCID: 54.1/mqg2g9xj
Discuss with AI
CNN localizationWeibo censorshipcensored social media postsmultimodal analysissentiment analysis
Figures from the paper
Abstract (AI)
Widespread Chinese social media applications such as Weibo are widely known for monitoring and deleting posts to conform to Chinese government requirements. In this paper, we focus on analyzing a dataset of censored and uncensored posts in Weibo. Despite previous work that only considers text content of posts, we take a multi-modal approach that takes into account both text and image content. We categorize this dataset into 14 categories that have the potential to be censored on Weibo, and seek to quantify censorship by topic. Specifically, we investigate how different factors interact to affect censorship. We also investigate how consistently and how quickly different topics are censored. To this end, we have assembled an image dataset with 18,966 images, as well as a text dataset with 994 posts from 14 categories. We then utilized deep learning, CNN localization, and NLP techniques to analyze the target dataset and extract categories, for further analysis to better understand censorship mechanisms in Weibo. We found that sentiment is the only indicator of censorship that is consistent across the variety of topics we identified. Our finding matches with recently leaked logs from Sina Weibo. We also discovered that most categories like those related to anti-government actions (e.g. protest) or categories related to politicians (e.g. Xi Jinping) are often censored, whereas some categories such as crisisrelated categories (e.g. rainstorm) are less frequently censored. We also found that censored posts across all categories are deleted in three hours on average.
Key Findings
1
Censored posts across all categories were deleted within three hours on average.
2
Posts concerning anti-government actions and politicians were censored more frequently, while crisis-related posts such as rainstorms were censored less often.
3
Sentiment was the only censorship indicator consistently associated with censorship across the identified topics, consistent with leaked Sina Weibo logs.
4
The researchers assembled datasets containing 18,966 images and 994 text posts spanning 14 potentially censored categories.
5
The study introduces a multimodal analysis of Weibo censorship using both post text and images, unlike prior text-only approaches.
Research Object
Censorship of multimodal posts on Weibo
Research Subject
Topic-dependent censorship mechanisms, including the effects of text-image content and sentiment on censorship likelihood, consistency, and deletion speed
Publication Details
Publication Date
2019-01-01
Journal
Publisher
ISSN
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest