CredCount

AI based fact checking through social media data
Batch: 2017 Summer
Status: Unsuccessful

CredCount

AI based fact checking through social media data
Batch: 2017 Summer
Status: Unsuccessful

Company

AI based fact checking through social media data

An analytics platform, backed by data science and machine learning, used to infer the credibility of events in real-time.

Atlanta, GA, United States / New York, New York, United States

Founders

Tanu and Eric have known each other via the PhD program at Georgia Tech. Eric advised Tanu throughout her PhD, and they worked on a number of projects together. A paper from 2014, “Phrases that Predict Success on Kickstarter” (http://comp.social.gatech.edu/papers/cscw14.crowdfunding.mitra.pdf) resulted in many conversations outside of academia, with entrepreneurs wanting this as a product. Christian and Eric worked closely together throughout the summer of 2016 on a project called HackGT@UPC as one of the first international, collaborative hackathons. With Eric’s faculty oversight, Christian built this event over the course of a few months around the subject of “assistive technologies” (https://hack-gtupc.devpost.com/). How long have the founders known one another and how did you meet? Have any of the founders not met in person? Tanu and Christian met through Professor Eric Gilbert at Georgia Tech about 2 months ago. Eric is currently Tanu's PhD advisor and Christian worked with Eric as his teaching assistant during the GT study abroad program in Barcelona, Spain last summer. When Tanu came to Eric about turning her PhD research project into a full-blown company, Eric immediately thought of Christian. Tanu and Christian met a couple days later for coffee/discussion and jumped into customer discovery and a bit of market research, eventually leading them here to this application!

Progress

Starting in 2014, Tanu has been working on this model, the science and the technologies surrounding it for about 3 years now. Our first major paper (http://credcount.com/whitepaper) was reviewed and published in late 2014 in a top CS conference. Although the majority of the work such as data collection was finished in approximately 8 or so months, there are certain analyses that have taken place up until today. That said, we are in what we consider the prototyping stage: the service/technology in its current form is not suited to any specific market but rather a proof of concept. With a little bit of modification to the machine learning models, we feel as though we can target new markets in as little as a couple weeks. Thus, Christian has spent the last few months trying to identify where we fit and have the biggest effect as a company/product. So far, we’ve considered news, politics, finance and the public sector. With contacts such as news anchor George Howell from CNN, political advisor and campaign manager Jason Boles and Senior Vice President of investments Mr. Jeff Wolk from Raymond James in the financial sector, we feel as though our best fit at this moment is to ride the “fake news” wave. Currently there is no window dressing to our model. In its current form, it is command-line driven (literally “python model.py [input]”), but we feel as though who our potential customers might be should be identified before we begin to work on any sort of graphical interface and/or API. That’s why we’ve come to YC. Our idea currently is a product/application built around our technology as an easy to use display of current news items and their credibility rankings such that someone might be able to read through an article with a presumed level of bias or accuracy up front. It comes through the narrative of “news literacy” and an attempt to have all sides of every story. Think if every fact or bullet from a news article could have a certain score for credibility such that the least items with the least standing fall to the way side and only the true and solid facts of the situation are left. How long have each of you been working on this? Have you been part-time or full-time? Please explain. Tanu has been working on CredCount since early 2014 as part of her PhD in social computing with her advisor Eric Gilbert at Georgia Tech. The majority of data collection and analysis led into early 2015, and she’s continued to circle back to it in a “part-time” role as a project with the most potential as a product/service outside of academia. Christian joined on early this year, 2017, in a product manager capacity to build this paper/model into a business. Working part-time while taking classes since his joining, he’s found himself devoting more time towards this venture than schoolwork, such that Tanu and Christian both are ready to commit to the project for the summer and however long after that.

Idea

Tanu was attracted to the project by the prevalence of fake news spreading online following natural disasters such as Superstorm Sandy in 2012. As she saw people sharing a lot of incorrect or misleading information about the events, Tanu decided to track both big stories and smaller rumors with the goal of creating an app that could help ordinary people sort fact from fiction to make decisions that could be crucial. Tanu has an undergraduate degree in computer engineering and is currently finishing her PhD in computer science with a specialization in social computing. Christian is working on his undergraduate degree, but is focusing into the domain of people and society. He’s currently finishing up a Georgia Tech sponsored paper on the website http://voat.co with qualitative research through interviews with users. Based on Christian’s market research, the financial, political and news markets seem to be the most motivated to move forward with our work. In my discussions with Jason Boles, he would like to use our model as a sort of real-time polling for statements made from congressional candidate Kurt Wilson. While discussing the project with Jeff Wolk from Raymond James, Wolk proposed to use the service for adding credibility to events through social media for commodity trading decisions. Also, we have discussed the possibility of parsing analyst papers and summarizing the ideas and decisions they propose with scores as well. Lastly, in talks with George Howell, CredCount would be a great addition to the new media tool-set used in collaboration with pre-existing service like https://www.dataminer.co/. With this, instead of focusing on volume and noise alone, news outlets could intelligently use the world through social media as credibility to a story or event.

Basic event analysis has existed in its current form, based on volume and noise alone. The ability to take collective reactions through social media to help determine reliability of a specific event is what we bring to the table. Think of it as a haystack of social media posts and events that finding a needle in is close to impossible. We help shrink this haystack to make fact checking far easier.

Based on data science and specific to the social computing field, we fear FactMata (https://medium.com/factmata/introducing-factmata-artificial-intelligence-for-political-fact-checking-db8acdbf4cf1) the most. As far as we know, to this day, they have no model or product and haven’t released anything close to the technology described within our whitepaper. That said, other fact checking websites will be our main competitors such as Snopes (http://www.snopes.com/) and PolitiFact. Otherwise, as Google and Facebook begin to roll out their own solutions these will be internal solutions to problems faced by the all media services on the internet. We do not fear these big guys as much because we stand to help all the rest using a paid for plug-in service if we decide to go that route.

The data science and linguistics portion of these potential markets. What sort of language implies the most credible financial analysis of one specific commodity? Or, how well a certain political statement goes over with a constituency? Or, what sort of language might you look out for when flagging something as potentially fake? Currently with some sort of human touch these sorts of questions can be answered but it might take a bit of time while doing so. We’re attempting to help narrow into the credible pieces.

A specific case we’ve looked into is a contract or subscription service with big media companies. Similarly, DataMiner has made millions helping the news guys observe events as they happen going from labels of “this might be newsworthy” to “lots of people are talking about this; you should check this out”. We hope that our service will be an even bigger asset to news media companies bringing more than volume statistics through analytics. With early research, these contracts have figures in the millions.

For a product of our own we’d need to attract potential users by having a narrative for the social benefit of “news literacy”. Minds are changed through emotion rather than factual concrete evidence so this application would be more of a training to see both sides idea instead of showing someone raw evidence against their position. In this manner we’d be very chicken-and-egg. As a service, we’ll get users by sales and marketing showing how much value we could add to markets like high frequency trading or politics. We’d have to lineup demos and grow through grassroots from the beginning.

Others

Christian’s ideas include:

- A social experiment much like Facebook where every login or session has you assume the role and identity of a fictitious person. All data created by you while in the role of this person persists upon session expiration or sign out. Each time you log back in you must put yourself in the shoes of the previously created identity and post as such breaking the echo chamber that is social media today while also having fun.
- Event-going today has turned into a very spontaneous unscheduled sort of thing. Often times I find myself wishing to get in touch with the people I met within the context of some random evening for pictures, videos and just networking in general. Think of this as a digital collection of your concert/sports tickets where each one expands into your experience prior and after the specific event!
- Having spent 3 summers abroad in Barcelona working closely with nightlife promoters, one thing they always talk about for general improvement of their day to day is a way to more easily get in touch with their network to predict figures for guest lists and to not spam as much. Think of this as a 1 degree removed connection of the promoter and I want to hand over my network list of those specific few that are headed to Barcelona for vacation. Hopefully this would be that he’d have my network only for the time that my friends are travelling to cut through on the spam.

Christian: I over analyze social settings a bit too much and over the years I’ve noticed just how much weight we give first impressions… With that, I always try and give strangers and new contacts the time of day/benefit of the doubt while attempting to be my best self in return.

Curious

When the triple request for startups based around news, jobs and democracy dropped, Tanushree and Christian immediately decided that it was time to apply. After years mulling around with the idea that this model could potentially be a business, we decided to jump all in.

Online and through friends

Comments

Get notified
When there are
0 Comments
Inline Feedbacks
View all comments