Analysis of Censorship of Weibo

Introduction

For Internet companies with huge information, information review plays a very important role. The information review work is the guarantee of the quality of the entire website and the lifeline of the website. For example, Sina, Sohu, facebook, Alibaba, twitter, etc., which generate massive amounts of information every day, as well as various large-scale forums and blogs, need professional information auditors to handle professionally. The auditing system and standards are different. For each company, we mainly analyze the comparison of Facebook and Weibo's auditing system and the comparison between the two companies. In detail, we discuss how to judge whether the content is in violation and decide whether to delete it. Method.

Theoretical Context

The media review includes all television, print media, radio, film, cinema, text messaging, instant messaging, video games, literature and networking that have a large audience. Chinese officials can access uncensored information through an internal file system. Reporters Without Borders ranks China's publishing situation as 'very serious,' the worst of its five grades. China's online censorship policy is 'universal' by the OpenNet Initiative's global network, which is also the worst level. Freedom House ranks China's publishing as the worst grade 'not free,' calling it 'a complex combination of state control over political media in China through news content, legal restrictions on journalists, and financial incentives for self-censorship.' Realization', and the implementation of more and more 'network missing' materials were written by active bloggers.

Save your time!
We can take care of your essay

• Proper editing and formatting
• Free revision, title page, and bibliography
• Flexible prices and money-back guarantee

Place an order

Detailed Analysis of Case Study

Censorship of Weibo

Sina Weibo is China's largest and most popular Weibo platform. Registered users are said to have reached 500 million. Every Chinese netizen has registered an account. Unlike Facebook, Sina Weibo must undertake the review task. Researchers at Houston Rice University have collected millions of posts to analyze and identify the size and speed of postings on Sina Weibo examiners (or Weibo's small secretaries). The paper (PDF) was published on the preprinted website arxiv. The researchers observed that 30% of the deletions occurred within 5 to 10 minutes after posting, and 90% of the deletions occurred within 24 hours (Zhu et al., 2013). Assuming that an inspector on Sina Weibo can read 50 posts per minute on average, then scanning for 70,000 new posts per minute in Sina requires 1400 people to work at the same time (Douglas, 2017). If they work 8 hours a day, then 4,200 people need to be satisfied (Douglas, 2017).

Given the limitations of the Weibo API, researchers have primarily tracked sensitive user groups that are most likely to send sensitive posts. From July 20 to September 8, 2012, researchers used the API to search the timeline of 3,500 users at a frequency of once per minute, searching for the common timeline every four seconds (Lee and Kim, 2012). Since Sina Weibo does not support anonymous queries, they use Tor to hide IPs and create fake user accounts. They collected a total of 2.38 million user timeline posts, and the deletion rate was 12.75% (Beveridge, 2015). Considering the size of the big data set that Sina needs to deal with, the peak value of 5 to 10 minutes after posting, especially considering that the deletion cannot be handled automatically, how does Sina quickly find and delete sensitive posts? (Agrawal et al., 2004) There are six hypotheses:

• Sina Weibo has a list of monitoring keywords, and the examiner will browse the posts containing these keywords to decide whether to delete them.
• Weibo has targeted the monitoring of users who frequently send sensitive posts.
• After discovering a sensitive post, the examiner can trace all relevant reposts and delete them all at once.
• By keyword search, Weibo deletes the traced post and causes the specific keyword to appear to delete the peak in a short time.
• The examiner's work is distributed and relatively independent, and some of them may be part-time.
• The speed of deletion is related to the theme, and there is a difference in the speed of deletion according to the sensitivity of the theme. The researchers used natural language processing techniques to analyze the topic and found that some popular topic posts were deleted faster than others, such as group sex, Beijing rainstorm deaths and judicial independence.

The researchers summarized the filtering mechanism of Weibo, in which the active filtering mechanism includes: explicit filtering, Weibo notifies the poster that their post content violates the content policy (but sometimes the user is not sure what is the sensitive word blocked) Implicit filtering, Weibo needs to allow the post to go online after manually reviewing the post; pretending to post successfully, other users can't see the user's post.

Comparison between Censorship of Weibo, Facebook and Twitter

The difference between Weibo and Facebook and Twitter is:

(1) Buddy List

Sina Weibo does not have a buddy list function, and can not classify and manage buddies; Twitter has a strong list function, which can not only classify its friends, but also classify any users, and the classified list can also be subscribed separately.

(2) Interconnection and Interoperability

Sina Weibo and Sina's user system integration, but not integrated with other websites; Twitter can be integrated with FriendFeed, Facebook, LinkedIn and other systems to achieve full two-way interoperability.

(3) Openness

Sina Weibo is currently a completely closed microblogging website, does not support API, does not support RSS, does not support computer client, does not support mobile client, in short, almost nothing is supported; Twitter is an almost completely open microblogging Services, in addition to registration, almost all features provide API support, there are countless client software, support for RSS, a large number of users use the unofficial client to update Twitter, users can deeply understand that Twitter is not a website, but a service.

(4) Review Mechanism

Sina Weibo has a review mechanism, the information released by users will be monitored, and “harmful information” will be deleted. This is also the experience of Sina’s pioneering Internet pioneers. There have been many microblogging platforms in China before, but they are unable to solve the information release. The monitoring problem is forced to shut down; Twitter does not usually review the user's information, but the user who distributes the ad will delete it.

Facebook's censorship policy was born in the conference room, but the implementation was carried out by an auditor in an outsourcing company in Morocco or the Philippines.

Facebook said that auditors have plenty of time to review posts and have no performance requirements. But the auditors said they would review about 1,000 items a day (Ellison et al., 2007). Each review time is 8-10 seconds, and the video will take a little longer (Ellison et al., 2007). That is to say, in less than 10 seconds, the auditor needs to filter more than a thousand pages of rules in the brain to make judgments.

For some people, pay is tied to speed and accuracy, and many people only work for a few months. One of them said that they had little incentive to contact Facebook when they found defects in the review process. As far as the system itself is concerned, Facebook mainly allows companies that hire these auditors to self-manage. The rulebook is written for English speakers, so many of these reviewers don't speak the local language on the post, they sometimes rely on Google Translate, but Google Translate is not reliable. After all, understanding the local environment is critical to identifying inflammatory speech. An auditor said that if no one can read the language of the post, approve any post, which may have contributed to violence in certain areas, such as Myanmar. Although Facebook said that such an approach is illegal, it has little knowledge of these outsourcing companies, and it is difficult to control them because Facebook relies on these companies to expand elsewhere. Also, another obstacle to controlling inflammatory rhetoric may be Facebook itself. The company's algorithms tend to promote the most inflammatory content, sometimes the kind of content that Facebook wants to suppress. Facebook tries to simplify the problems that even legal experts are likely to entangle into one-size-fits-all rules, such as what is annoying ideas and what is dangerous rumours. Facebook believes that these documents are for training purposes only, but the auditors say they are used as a daily reference. Looking at it alone, each rule might make sense. But on the whole, it is not necessarily the case. Jonas Kaiser, an expert on cyber extremism at Harvard University, says it’s “extremely difficult” for a technology company to delineate these boundaries. “It allows social networks to make decisions, and traditionally it’s court Work.” Sana Jaffrey, who studies Indonesian politics at the University of Chicago, says the ban is a shortcut. It requires the auditor to find a banned name or logo, which is much easier than letting them judge when political opinions are dangerous. But this means that in most parts of Asia and the Middle East, Facebook prohibits tough religious groups representing important segments of society, which is equivalent to closing one side of the debate. Also, Facebook's decisions tend to favour governments that can fine or supervise them.

Facebook's main seven review system initiatives:

• (1) Stronger detection: Improve the ability to distinguish error information, establish a better technical system, and detect error messages before people report.
• (2) Simpler reporting: Improved user interface makes it easier for people to escalate to get error messages faster.
• (3) Third-party verification: Zuckerberg wrote, 'There are many fact-checking agencies, we have already contacted such institutions, and we plan to learn from more institutions.'
• (4) Warning: Label fake news tagged by third parties and the Facebook community, and display a warning when people read and share.
• (5) Relevant article quality: Improve the accuracy of “relevant reports” recommended to users.
• (6) Interrupting the fake news economy: Suspension of fake news sites using Facebook's advertising services.
• (7) Listening: Zuckerberg wrote, “We will continue to work with journalists and others in the news industry, especially through their participation, to better understand their fact-checking systems and learn from them.”

Facebook released a Community Standards Enforcement Report, the first time they disclosed the disposition of non-compliant posts and fake accounts. In the first quarter of this year, Facebook deleted a total of 2,888,700 posts containing pornographic violence, terrorism and hated speech, while closing and deleting approximately 583 million fake accounts (Kingston et al., 2018).

On the same day, Weibo also announced the “Community Management Work Announcement” in April, which is a monthly report that began in 2018 (Tse et al., 2018). It also uses a series of figures to show the determination of rectification. Tse etc. Also found that in the past month, Weibo has blocked and deleted more than 62 million Weibo and nearly 70,000 junk and machine accounts were cleaned up, which is not counting more than 8,000 accounts that involve political, pornographic, pornographic, and false information.

The volume of the two companies is huge. Facebook is the world's largest social network, with more than 2 billion active users and nearly 400 million Weibo (Chiu et al., 2012). Two months ago, Facebook’s data loss in two days was almost equivalent to two Weibos due to data breaches from Cambridge analysis.

There are some differences in the factors that drive the two companies to move. For Facebook, in addition to the big troubles of unseen privacy leaks, fake news, violence, pornography, prejudice and other harmful content on the platform have made it criticized for ten years.

Weibo was interviewed by the National Internet Information Office in January this year and was asked to rectify and disseminate the problems of illegal speculation, vulgar pornography, ethnic discrimination and other illegal and illegal information. The hot search list, hot topic list, Weibo question and answer The functions are off the assembly line and the offline time lasts for a week.

What content will be deleted? Both reports list the types of posts that are cleaned up. Facebook mainly includes terrorism propaganda, picture violence, nude and sex, hate speech, spam and fake accounts. Facebook has recently been widely criticized for hate speech, but the highest proportion of the deleted content is pornography – a total of 21 million 'nude and sex' messages were cleaned up in the first quarter of this year, the largest share of all categories. Ranked second is picture violence, with a total of 3.4 million deleted, accounting for 11% of the total (Plantin et al., 2018).

The same is true for Weibo, and posts for adult content have been deleted the most. Www.qdaily.com selected the announcement data of Weibo in the last three months. During this period, there were 4.814 million pieces of information about yellow information, accounting for 90% of the total number of posts (Glowacki et al., 2018) and the number of accounts banned due to pornographic content reached 29,134. Politically relevant information is also the focus of the post. For Facebook, this piece mainly refers to terrorism and hate speech. In the first quarter of this year, Facebook took a total of 1.9 million posts related to terrorism such as Al Qaeda and Islamic State (ISIS), up from 1.1 million in the previous quarter (Vergani, 2018).

However, there are only 2.5 million hate speeches handled by Facebook this time, the smallest of all categories (Newman, 2018). Moreover, Facebook's self-tested content accounts for only 38% of all tagged content, and more than 60% of content needs to be reported by users (Newman, 2018). Part of the reason for this is because 'Facebook's artificial intelligence system is still difficult to tell the hate in speech,' said Guy Rosen, vice president of product management at Facebook. Another reason is that hate speech on Facebook may not be obvious in itself.

In April of this year, some false information on Facebook contributed to the conflict between Sri Lankan Buddhists and Muslims. The culprit is often a rumour with inflammatory nature that may initially have nothing to do with hatred and prejudice. For the Facebook security team, similar issues have increased the difficulty of the review.

On Weibo, this piece is expressed as 'when the government is harmful information.' Weibo cleared 393,000 pieces of such information in three months and took measures to prohibit the issuance of Weibo and comments, restricted access, and account closure for 7975 accounts (Liu and Jansen, 2018). Weibo does not clearly define what 'when the government's harmful information' means. From the deleted account, the offenders come from various industries and even include @Liushenleilei, who interprets Jin Yong’s martial arts novels.

Accounts that publish false information, spam, etc. are also punished by Weibo or customs, or banned words. Counting the 51,100 accounts that were frozen due to automated behaviour, Weibo has dismissed 274,054 accounts in the last three months (Tian et al., 2018).

Facebook did not disclose the account status that was penalized for publishing these offending content but only listed the number of fake accounts that were closed. Guy Rosen wrote on the company's official blog that in the first three months of the year, the company closed about 583 million fake accounts -- a quarter of Facebook's 2.2 billion active users, the vast majority of which were registered (Valenzuela et al., 2018). It was closed for a few minutes. Among all the processed accounts, Facebook actively identified 98.5% of the accounts (Valenzuela et al., 2018).

'Artificial + Algorithm' is the basic means of content review. According to Facebook, 99.5 percent of terrorism-related posts were found by Facebook itself, not by users. Facebook attributed this to the improvement of artificial intelligence technology: 'This growth is mainly due to our increased ability to use image recognition technology to detect offending content, which can detect newly released content and detect old posts.'

Facebook's trend toward image violence, nudity and sexuality, and spam messages are similar. About 90% of the content was discovered before users reported it, and the main reason was 'technical upgrade.'

The only difference is that regarding hate speech, Facebook self-tested content accounted for only 38% of all tagged content, and more than 60% of content needs to be discovered by users (Marino et al., 2018). Moreover, only 2.5 million of the posts were processed, making people wonder if more content is still hidden in the dark (Marino et al., 2018).

According to Guy Rosen, vice president of product management at Facebook, the reason for this result is that artificial intelligence systems are still difficult to discern hate in speech. But Mark Zuckerberg is optimistic about the future. In his testimony in Congress, he expressed his plan to use AI to clear hate speech on his platform: 'I am optimistic that in five to ten years, we will have some artificial intelligence tools to gain insight into different types of content. Language nuances to be more accurate.'

In the review mechanism, Weibo also has a clear distinction with Facebook. After being reconsidered and rectified in January this year, Weibo adjusted the content review technology. Wei Wei, vice president of Weibo, said in an interview interview in February this year that Weibo introduced the 'edit manual intervention' model, using algorithm mining as the basis, giving up the pure algorithm in sorting and selection, and introducing editors to violate the relevant The content of laws and regulations, social negative energy information, and excessive entertainment information are manually intervened. Under this strategy, the manual review team of Weibo is getting bigger and bigger. According to Cao Zenghui, there are now 332 Weibo operators, considering increasing to more than 600 people, mainly responsible for “managing accounts with influence”; content editing center, now 50 people, will increase to not low In the case of 100 people, he is mainly responsible for “enhancing the manual intervention review of hot search and other hotspot areas”; the number of security auditors is now 1,100, which will increase to 2,000, mainly responsible for “auditing and dealing with violations and regulations, multi-layer 'Re-examination'; Weibo supervisors will be increased to 2,000 people, composed of Weibo public recruitment of netizens, report on the yellow information in the station, and then provide online subsidies and material rewards on a monthly basis (Mao et al., 2018).

Finally, there are 197 R&D personnel in R&D information identification technology, which will increase by about 10% in the future (Mao et al., 2018). Cao Zenghui said that the existing technologies include keyword detection systems, image detection systems, illegal information model detection, and harmful account feature libraries.

Counting all planned increments, Weibo has about 4,900 community content security personnel, and Weibo currently has 411 million active users (Lei et al., 2018). On average, each security staff needs to cover more than 80,000 users (Lei et al., 2018).

Facebook also has a human team, and their security team is divided into two teams: community operations and community integrity. The Community Integrity team is primarily responsible for building automated tools for reporting-response mechanisms.

Last fall, Facebook had more than 10,000 content reviewers and will expand to 20,000 this year (Tsay-Vogel et al., 2018). With Facebook's current 2 billion active users, each security person needs to cover 100,000 users (Chou et al., 2019). This makes Facebook spend a lot of money. In 2016, Facebook allocated approximately $150 million to its community operations, and by 2017, Facebook increased the community operations team's budget by nearly 50% to $220 million (Frattaroli et al., 2018). The Wall Street Journal quoted insiders as saying that the 2018 budget will be $770 million (Frattaroli et al., 2018).

In March of this year, the United Nations claimed that Facebook was responsible for spreading hate speech in the Rohingya minority in Myanmar. Facebook companies argue that it is difficult to stop the spread of hate speech because of the lack of reviewers who speak the local language. Guy Rosen said that Facebook would hire more local reviewers this year. Weibo did not disclose the specific funding for content review.

Conclusion

When a lot of problems break out, it is often caused by some deep-seated reasons for a long time. The fake news issue on Facebook is not just that the fake news we see is viral.

At the age of 19, Zuckerberg founded Facebook, which was originally based on a technology-controlled approach to building connections between schoolmates. Over the past 12 years, Zuckerberg has made some very strategic decisions to turn Facebook into the most powerful information 'dealer' on Earth through News Feed, but Zuckerberg The grid has not been incorporated into a media company. So far, the CEO of the technology-controlled company still emphasizes that Facebook is not a media company, but a technology platform that simply transmits information. Zuckerberg seems to have both fish and bear's paws. He wants to provide more content so that users can stay on the Facebook page, and they don't want Facebook to become an arbiter of authenticity. People are on Facebook. Still can publish anything that he wants to publish, so he subtly chooses to let the user judge the authenticity of the content, and provides the current so-called reporting system - but the reporting system does not play many roles. Therefore, the key to seeing the problem is that Facebook has not set its position first, perhaps not understanding its role and huge media influence. If you choose only to be a technology platform, don't provide media services; if you choose to provide Media services should bear the social, ethical responsibilities that the media should bear. Because only when it is recognized that it is necessary to bear the responsibility of the media, it will be discovered that the media has more complicated considerations than the current mode of operation. For example, the establishment of corresponding news rules - first of all, should establish a clear, unified standard to distinguish what is news, which content can be pushed as news (this is also the most criticized aspect of Facebook's News Feed business). The problem with Facebook's media services should be solved not only by algorithms, engineers or subcontractors. In addition to this, there is a bigger problem behind these disputes and business, which is the issue of interest orientation. As a company, Facebook is a core company in pursuit of digitally rising financial statements, regardless of the product service or algorithm design. It allows users to spend more time browsing Facebook pages for more users. Facebook - because this can attract more investment for Facebook and get more advertising revenue. As a profitable company, this is understandable; but as a news service company, you can't just consider the benefits, because news will affect users' values, emotions and many important decisions - for example, in the United States may even affect the election results. In this case, more consideration of its benefits will inevitably lead to more problems.

References

1. Agrawal, R., Kiernan, J., Srikant, R. and Xu, Y., 2004, June. Order preserving encryption for numeric data. In Proceedings of the 2004 ACM SIGMOD international conference on Management of data (pp. 563-574). ACM.
2. Beveridge, M., 2015. A Little Birdie Told Me: Twitter and the 2014 New Zealand General Election. Unpublished MA Dissertation. Massey University.
3. Chiu, C., Lin, D. and Silverman, A., 2012. China’s social-media boom. McKinsey & Company.
4. Chou, H.L., Liu, Y.L. and Chou, C., 2019. Privacy behavior profiles of underage Facebook users. Computers & Education, 128, pp.473-485.
5. Douglas, F.E., 2017. Circumvention of censorship of internet access and publication (Doctoral dissertation, University of Illinois at Urbana-Champaign).
6. Ellison, N.B., Steinfield, C. and Lampe, C., 2007. The benefits of Facebook “friends:” Social capital and college students’ use of online social network sites. Journal of Computer‐Mediated Communication, 12(4), pp.1143-1168.
7. Frattaroli, S., Schulman, E., McDonald, E.M., Omaki, E.C., Shields, W.C., Jones, V. and Brewer, W., 2018. Utilizing Facebook and Automated Telephone Calls to Increase Adoption of a Local Smoke Alarm Installation Program. Journal of public health management and practice: JPHMP.
8. Glowacki, M., Narayanan, V., Maynard, S., Hirsch, G., Kollanyi, B., Neudert, L.M., Howard, P., Lederer, T. and Barash, V., 2018. News and Political Information Consumption in Mexico: Mapping the 2018 Mexican Presidential Election on Twitter and Facebook.
9. Kingston, B., Mattson, S.A., Dymnicki, A., Spier, E., Fitzgerald, M., Shipman, K., Goodrum, S., Woodward, W., Witt, J., Hill, K.G. and Elliott, D., 2018. Office of Justice Programs Facebook logo Twitter logo.
10. Lee, S. and Kim, J., 2012, February. WarningBird: Detecting Suspicious URLs in Twitter Stream. In NDSS (Vol. 12, pp. 1-13).
11. Lei, K., Liu, Y., Zhong, S., Liu, Y., Xu, K., Shen, Y. and Yang, M., 2018. Understanding User Behavior in Sina Weibo Online Social Network: A Community Approach. IEEE ACCESS, 6, pp.13302-13316.
12. Liu, Z. and Jansen, B.J., 2018. Questioner or question: Predicting the response rate in social question and answering on Sina Weibo. Information Processing & Management, 54(2), pp.159-174.
13. Mao, K., Niu, J., Chen, H., Wang, L. and Atiquzzaman, M., 2018. Mining of marital distress from microblogging social networks: A case study on Sina Weibo. Future Generation Computer Systems, 86, pp.1481-1490.
14. Marino, C., Gini, G., Vieno, A. and Spada, M.M., 2018. The associations between problematic Facebook use, psychological distress and well-being among adolescents and young adults: A systematic review and meta-analysis. Journal of Affective Disorders, 226, pp.274-281.
15. Newman, D.K., 2018. Racist Discourse in Disguise: Exploring How White Facebook Users Expressed Ambivalence Following the Police Shooting of Philando Castile.
16. Plantin, J.C., Lagoze, C., Edwards, P.N. and Sandvig, C., 2018. Infrastructure studies meet platform studies in the age of Google and Facebook. New Media & Society, 20(1), pp.293-310.
17. Stutzman, F. and Kramer-Duffield, J., 2010, April. Friends only: examining a privacy-enhancing behavior in facebook. In Proceedings of the SIGCHI conference on human factors in computing systems (pp. 1553-1562). ACM.
18. Tian, X., Batterham, P., Song, S., Yao, X. and Yu, G., 2018. Characterizing depression issues on Sina Weibo. International journal of environmental research and public health, 15(4), p.764.
19. Tsay-Vogel, M., Shanahan, J. and Signorielli, N., 2018. Social media cultivating perceptions of privacy: A 5-year analysis of privacy attitudes and self-disclosure behaviors among Facebook users. new media & society, 20(1), pp.141-161.
20. Tse, D., Weng, X., Wan, X., Yao, L. and Zhang, W., 2018, August. Research on Information Security Perception of Sina Weibo Users Based on KISCAP Model. In 2018 17th IEEE International Conference On Trust, Security And Privacy In Computing And Communications/12th IEEE International Conference On Big Data Science And Engineering (TrustCom/BigDataSE) (pp. 1584-1588). IEEE.
21. Valenzuela, S., Correa, T. and Gil de Zúñiga, H., 2018. Ties, likes, and tweets: Using strong and weak ties to explain differences in protest participation across Facebook and Twitter use. Political Communication, 35(1), pp.117-134.
22. Vergani, M., 2018. How Is Terrorism Changing Us?: Threat Perception and Political Attitudes in the Age of Terror. Springer.
23. Zhu, T., Phipps, D., Pridgen, A., Crandall, J.R. and Wallach, D.S., 2013, August. The Velocity of Censorship: High-Fidelity Detection of Microblog Post Deletions. In USENIX Security Symposium (pp. 227-240).