Testified: An information verification framework
Originally written: http://www.avineshwar.com/2018/07/testified-information-verification.html
You might see the word framework. Well, we do kind of need a framework at this point as a lot of people need to follow that.
Why? Free press is dead. People are selling opinions and journalism is probably thriving, at its best, in universities.
Coming back, the intention is to offer a mechanism which can help in targeting "fake news".
I believe "fake news" is a term that gained attention from an urban dictionary perspective in a sense that it does not actually just talks about the "fakeness" of any piece of media-related information but, any information in general that is being shared with a wider audience with an unclear or a hidden motive.
Now, this means verifying a piece of information that has been shared by an individual (or a bot) on a social platform (blogs are included). We want to limit such tests in such a way that only the information that should be testified, takes the route of "testification".
What we intend to test:
>> any image that does not maps only to an individual i.e. an image containing textual data and an image should be checked, an image containing only text should also be checked, an image containing a group of friend or an individual should not be checked. You get the idea.
>> any shared information that is not limited to oneself, personal, involving (and is limited to) friends/families/etc.
>> transcription of shared videos as necessary. For example, too many laughing smileys/emoticons gives us an idea that probably it is a funny video and need not be testified; on the other hand, a video with a shock smiley or anything potentially concerning from a global perspective needs to be testified.
Some points to note:
>> Keeping a score against an account from which an information is shared. So, if my score is not satisfactory from the correctness perspective of information sharing, a viewer, to some degree, can choose to make a decision about my post. This also gives me a way to keep a track of such accounts in a more granular way.
>> To any shared information (post automated assessment), assigning a metadata information like "neutral", "false", "probably false", probably true" and, "true" will help in ensuring a benign information forwarder (or "sharer" of any information; a benign forwarder is one whose account score is not alarmingly low else, a more stringent but contextually relevant measure can kick in) is aware of the probable correctness of the shared information.
Some ways of automated assessments:
>> Quora
>> Stackoverflow
>> Books (Google TalkToBooks, may be?)
>> News sources (especially, regional)
>> Historical news data
>> and so on..
Important:
>> The metadata can be expanded but should be kept as low in number as possible. There should also be non-conflicting definitions of each of the metadata terms.
>> Assigning scores to the places from where assessments are conducted is also necessary to ensure that the 3rd parties are providing reliable information.
How to measure success:
>> A program for on-board independent paid information reviewers who provide a brief feedback post assessment, per post. For example, a "probably false" could become "neutral" post assessment. This essentially means that we identified an item to improve upon in our information assessment automation.
>> Hire academics and/or people who get excited about journalism.
>> Something else?.
Finally, I do think citation makes a lot of sense in this framework too i.e. a piece of information with an un-trusted citation can be treated differently, perhaps, checked against an oracle for truthfulness or can be categorized as a false information.