Crowdsourced Content Filtering and Cascades
A lot of commercial web pages and applications exist with their own regulated content, but in more recent past decade, there have been a rise of websites and applications that, instead of generating their own content, rely on users to create their own content (e.g. YouTube, Imgur, tumblr). This has given rise to the role of the owner as intermediary instead of producer. With this change in role comes with new challenges. As a content producer, the company makes decisions on what to filter and produce. However, in an environment that encourages creativity and collaboration, this is no longer the case; they would like to encourage user content (obviously) but now have the problem of liability for what is posted on their platforms. This is an important issue for three reasons: legal responsibility, technical design aspects (how websites and applications can be designed to invite user-content), and protection of privacy, free expression, and its balance with business.
On one hand, the company shouldn’t be in charge of all the filtering because of the problem with all the content belonging to the users. On the other hand, there are legal obligations for filtering some things (child pornography, obvious copyright, etc.). However, filtering effort has a cost to it and if the company takes it upon themselves to do all the filtering, there is a tendency to over-filter things that are less problematic in the community’s POV, and this causes much user-dissatisfaction (such as video copyright claims for background music in games). The solution to this, then, is often self-governance for platform management. Self-governance is a familiar and popular idea in the realm of the Internet; people in a particular community form an attachment to the platform as well as other users. This evokes a feeling of independence as a community and so there is a difference, for users, between filtering enforced by the intermediary and filtering by users themselves. This system works because seriously illegal things like sale of copyrighted material and child pornography (which would be a huge liability) are disapproved by the vast majority and so will get filtered very easily. Small cases of gray-areas (like derivative works of copyrighted material or artistic paintings that happen to have nude children as elements…think cherubs) could be passed over by users, which are not much of a liability for the intermediary and having this content keeps users happy. Thus, users themselves become “gatekeepers” for what content is shared and what is filtered.
How does this work? One popular mechanism is “red-flagging” (a familiar feature on popular websites like YouTube), which, paired with an appropriate system can lead to effective filtering. For example, if a certain piece of content is flagged illegal by users three times, then it gets removed. If a person is deciding whether or not to add the third flag, seeing that it’s been flagged twice sort of “confirms” their decision and can influence their vote. This way, some illegal content would be more and more susceptible to flagging from this reliance of users on the votes of those before them (sounds familiar?).
There are still some small problems, though. An intermediary’s own values may dictate a want for certain content filtered, but only the highly illegal and hateful extremes actually get filtered by the users. For example, copyrighted materials are commonly shared on the internet. Another problem is that with power in the hands of the users, there is still a tendency to over-filter certain other things that go against certain user’s own values, so that there is stricter filtering than the intermediary may have intended (e.g. Facebook dealing with women angry about removals of pictures of mothers breastfeeding).
The reason this paper seemed relevant to what we are learning in class is the general topic of the wisdom of crowds and a little on information cascades. I found the overall topic of user-governed filtering interesting, but also the page or two on how it works and the motivation behind it. Having a large number of people report certain content indicates that perhaps they are onto something; it is also the case that perhaps people’s voting comes from the influence of what came before them. Consider another example: Reddit, an online forum for discussion and other shenanigans. Users can upvote or downvote other user’s comments and posts. Sometimes it is the case that a crowd mentality forms; when a certain comment starts off getting downvoted by a few people in the beginning, it may influence other people to downvote the comment as well, starting a snowball effect. It could have been that the comment was only a little controversial, or just a sarcastic joke taken seriously by the initial downvoters. When later users see this comment, they may have thought it was sort of ok, but seeing the large amount of downvotes may have caused them to conclude that this comment was perhaps not so funny afterall. This could be one of the factors for why popular image-sharing site Imgur, in one of the updates for the website design, removed the red-green vote ratio bar below pictures (the information is still accessible, but not as easily) and kept just the point value. Often on these sorts of websites, highly downvoted comments end up on the very bottom of the page or hidden; in a way, this system not only serves as a way of judging the helpfulness/comedic value of certain comments/posts relative to other comments/posts, but also an example of filtering.
Source:
“LET THE USERS BE THE FILTER? CROWDSOURCED FILTERING TO AVOID ONLINE INTERMEDIARY LIABILITY”
http://ipp.oii.ox.ac.uk/sites/ipp/files/documents/IPP2014_Hartmann.pdf