 |
| Before Caffeine |
Using Caffeine to Fight Against Fake News
How to create maximum value in minimum time ? Attend a design-a-thon.
1. And the wait is over…
At 7:30 PM we finally see the problem. All the speculations about topics, structure, and the event
actually being a hack-a-thon which has been marketed as design-a-thon to pull the non- tech
crowd comes to an end.
The best problem statements that we’ve encountered, not for what it asks but how it asks for it.
2. Pick and split
Which topic to choose was the only thing in the entire event on which we were at a split. Evaluating
the pros and cons of going down the product design road with very topic, we finally arrived at
building a solution to tackle fake news.
The two main influencers for our decision were:
1. Our lack of any in-depth knowledge for cyber security and it’s current state in the world.
2. Facebook released their ‘Fake News Checker’ just a day before, hence providing us a plenty of
resources on working of fact checkers by some major tech blogs. Thanks Mark for making
the timing so right :)
Next came splitting the responsibilities. We knew that it’s going to be a long night and hence if
something break downs or we miss a deadline, there needs to be someone held responsible but
who ? We decided on voting for a team leader but the condition put forward was that who ever
becomes the leader, voluntarily / involuntarily, must be entitled to one third the prize, if we win.
Since everyone was pretty confident on their skills and strengths we decided not to follow that
road. The structure was proposed. Looking reasonable, everyone get’s to
have their own team and be a leader.
So finally we came to this split:
You’re my employee! So are You.
3. Mujhe Idea aayega!
The real brain work starts now. What to and how to do it ? Here, we would like to thank Simon
Sinek to creating such an amazing framework for product designers. Having read at several places and gathering ideas from here and there, Karanjot figured out few ways to segregate fake news from the genuine one and also identified two broad principals of calculating the authenticity of a content piece.
Traversing these circles from the Why to the What made us empathise with those people who
facing the fake news problem for real. For here emerged our market assumptions, major user
groups, a partial marketing strategy and most importantly a confident product idea!
Now, it was time to go out in the field and get our assumptions validated.
4. Making way through the green field of Users
It was already 9:45 PM and we had just finalised our questionnaire. This speed was slow but
considering the fact that we had actually gone through brain storming to find the need and
inference drawn from every question we were complacent because even if we acquire a smaller
data set, we will know what exactly to draw from it. Nothing left hazy and unaccounted due to
variability in human nature.
Questionnaire Reasoning
We distributed on this questionnaire to those people whom we knew won’t ask what are you doing
but simply fill the form and save our time. In two hours or so we received eight responses, two
more than we expected!
Finally we met our 1 am deadline with this
Our not so scribbled Scribbling
5. Reaping the green field of Users
We had what we wanted to know about users, a small set yet precise. It was time to formalise
everything we had learned about the user, his current behaviour, needs and possibly how we could
change this behaviour (if we wanted to).
All this learning, brainstorming and mess got formalised in an affinity map.
The affinity map for the world, the argument map for us
We finally met the 3 am deadline despite taking a lot of time for bullying Divyansh for coming in at
2nd place in the 2 am push up challenge.
6. Alert! Graphics starts here
We put our analytical caps asides and put on an artist’s beret. Tried to figure out the action triggers
on the app. We knew our app would be like the TrueCaller for content, so why not use their actions
and functions!?
The problem was accessing entire messages from the notification is quite complicated or maybe
even impossible in Android. But we found out something interesting, accessing individual
messages for scanning might not be possible but accessing an entire screen is! All we have to do
is take permission to overlay on any app from the user. Thus, we started working in this direction.
Unable to figure out any solution for parsing messages to scan for fake content, Karanjot out of his
usual habit started tapping the screen aggressively. It didn’t take time for his finger to land on the
home button and stick there for three seconds. Voila! Google On Tap opened and hence we knew
exactly what our action was going to be.
We drew a few wireframes
It takes genius designer or maybe just 4 other to decipher the wireframe Divyanshu had in mind.
But we did and came out the beauty out of Akhilesh’s hand.
7. The final step
Here it, all the brain, sweat and caffeine put together. Hope you like it.
Bonus: Video that explains how the application solves the problem
Bonus: Behind the Scenes Sneak Peak
We identified two broad principals of calculating the authenticity of a content piece.
Principal 1: Tally with renowned news sources.
Working: Viral extent of a news can be taken from trending tags on social media and other news
sharing apps like newshound, InShorts, etc. Viral extent maybe taken into account for
determining whether news is fake or not. Tags and key words from the headline and content of
the news can be used in a search algorithm that checks whether an article with same keywords
is present on the renowned news sources or not. Accordingly a news can be declared fake or
genuine. For determining a renowned source basic criteria such as year of establishment, user
reviews, etcetera are taken into account.
Pros: No human interaction.
Cons: Might show real news fake if no occurrences found on renowned sources. Can be overcome
by little human input.
Principle 2: Human resource
Can be implemented in two ways:
1. Dedicated experts
2. Crowdsourcing
1. Dedicated experts: Dedicated experts can review after an algorithm does basic filtration.
Pros: High accuracy
Cons: Slow process, expensive deal
2. Crowdsourcing: Volunteers who act as moderators can flag an article fake or genuine and according to the
reports fake news can be segregated from the genuine one. There could be a reward, rating, and
recognition system that motivates users to flag the fake news.
Pros: Can be accurate to a large extent
Cons: Cannot trust all volunteers but if sample size is huge, average reviews can result in
accurate flagging of the news.
DETERMINATION OF PERCENTAGE OF AUTHENTICITY
Over time the algorithm can learn which keywords are highly likely to be used in a fake article.
On this basis, it will increase the confidence of the percentage determined by the algorithm and
consequently decrease human intervention. The weighted mean of all percentages will give the
final percentage.
Confidence of P1 (by computer) = x
Confidence of P2 (by computer) = y
Confidence of P3 (by computer) = z
Where P(i) is the percentage
Final Percentage P = P1x + P2y + P3z
As the computer keeps learning to identify keywords, genres and trends of highly faked news, it
will ask for human intervention for fewer links.
P1 is a function of number of keywords that match patterns of highly faked news, difference
from original source, reliability of original source, genre and trend.
P2 = no. of moderators who marked authentic/ total no. of moderators asked
P3 = no. of people who marked authentic/ total no. of people asked
One such algorithm can be made that understands the context and accordingly matches latest
news from renowned sources and finally determines the percentage of authenticity. For
example, fake news is that a celebrity died in a car accident. The algorithm understands the
context as tragedy and car accident and look for similar occurrences. Say no such keyword was
found that suggests that the celebrity actually died in that. Therefore, taking all the information
into account a percentage of authenticity can be determined.
 |
| After 20 shots of caffeine each! |
Fun days :')
ReplyDelete