PopTheBubble — A Product Idea for Measuring Media Bias

Author(s): Sanghamesh Vastrad

Originally published on Towards AI.

Product Management

A Product Manager’s Perspective on building a Crowdsourced Media Bias Tracker and Anonymous Political News Aggregator

A couple of months ago, I decided to try something new. The MVP Lab by Mozilla is an 8-week incubator for pre-startup teams to explore product concepts and, over the 8 weeks of the program, ship a minimum viable product that people want to use. My team worked hard to come up with a product idea and submit an application for consideration. We didn’t make the cut; however, I learned many product ideations and management concepts along the way. Below is the product idea that I believe is worth sharing!

Description of product idea: What will the Minimum Viable Product be in 8 weeks?

The “personalization” on the web or the algorithmically confounded choices given to us can lead to filter bubbles that often limit us from reading ideologically diverse ideas, opinions, and more recently, news. Media bias is a huge problem all around the world, but according to studies, media houses are often characterized into left leaning or right leaning without proper analysis. A fair and balanced political environment can be created when media houses are tracked for bias daily and people get to see all angles to the story by popping the bubble!

Minimum Viable Product in 8 weeks: A website and an app where users will be able to read anonymous articles and rate content (without having their own bias about media houses) and also see how the media bias changes over time in media outlets for a variety of issues. The intend is to not only become a credible source of media bias quantification like Snopes.com is for accuracy, but also to bring people and communities together for a less polarized, filter bubble-free world.

Media bias is believed to operate via two mechanisms: selective coverage of issues, known as issue filtering, and how issues are presented, known as issue framing. Also, there is no agreed-upon methodology or source for quantifying media bias. The MVP will be based on a research paper “Fair and Balanced?” which uses a combination of machine learning and crowdsourcing techniques to effectively tackle this problem.

To address the problem of quantifying issue framing, an NLP based classifier will be built that will be able to classify articles into “political” and “non-political” categories and further into subsets like news, opinion, and also issues such as healthcare, economy, etc. This classifier will be used to filter out non-political content from our website/app and also tag articles with issues or label them as news or opinion. For issue framing, content-based quantification methodologies are often preferred over audience-based ones. The website/app will track media bias on a daily basis through crowdsourced content analysis.

A user entering PopTheBubble will be presented with anonymized political articles. After reading each anonymized article, the user will be asked to rate it on a scale of left-leaning to right-leaning. All such results will in turn be used to quantify slant in issue framing. This way, not only do slants of media houses get captured at an outlet level but also an issue level.

People want unbiased news and are willing to explore newer platforms that provide it. Many people also prefer aggregation of various issue-based political news but news channels and even social media (through their “personalized” news feed) nowadays are rarely moderate; they write or promote articles either far right or far left which can provide a skewed outlook of reality. Hence, people will use PopTheBubble to check the polarity of media outlets and get ideologically differing news and opinions.

The above-discussed methodologies will allow the creation of a website/app that tracks bias/slant in real-time which no other website or app does right now. In the future, PopTheBubble can be extended for a variety of purposes including giving smaller media houses a platform to publish their articles, becoming a preprint bias checking tool for larger media houses who want results in real-time, and allowing people to express their opinions on the articles. The startup will be a Software as a Service (SaaS) for media houses and a product for the consumers.

Competitors:

Very few third party companies exist that check and track the bias of big media companies. Even a thorough search found no big players in the market. Two sites that had similar functionalities were allsides.com and mediabiasfactcheck.com. While allsides.com classifies and presents facts from different perspectives, they only let voting/rating at a media outlet level (which isn’t fair) and there is no way to interact or vote for these articles. Also, they still show the news outlet which wrote the articles. Mediabiasfactcheck.com is pretty much static and doesn’t let users have a say. Although similar, these sites are way off from our goal. At present many news companies have an internal system for checking the biases of their content but no third party crowdsourced provider. PopTheBubble can be that in the long run.

User Acquisition:

The first 1000 users will be attracted via our professional networks and social networks (LinkedIn, Facebook, Instagram, Twitter) and by writing articles on LinkedIn and Medium to make people aware of our platform. Once all the social media resources are used up, the option of online advertisements will be explored.

First two weeks milestone:

A simple website which aggregates all the news and allows the user to rate it on a scale of left-leaning to right-leaning. The steps involved would be gathering the data through APIs and aggregating them and then automating it using AWS lambda. Meanwhile, two of our developers would start building a react backend for the website and one of the developers will learn react-native for the app. Also, some time will be spent on creating mockups and designing the system. Our NLP classifiers should have labeled data and a final model to train on by the end of two weeks.

Technical Details:

As explained in the product idea, we’ll tackle the problem of quantifying and tracking media bias through content-based crowd-sourcing:

1. Popular US media websites (around 20) will be scraped on a daily or hourly basis. This can be done using general news APIs like Google News API, News API, Bing Search API, etc, or from specific media houses like News York Times API, BBC News API, etc.
2. We pass it through a classifier that classifies it as political or not, news or opinion, and tags it with issues within the political landscape, etc. To build the classifier, we will need to train an NLP based model on labeled data. We intend to do this using already available datasets like News Category Dataset on Kaggle and using the data obtained from APIs from step 1 to create our own dataset. In case of creating our own labeled dataset, the labeling task will be crowdsourced on Amazon MTurk.
3. Anonymize the article and present it to users who’ll read it on our website and rate it to be left-leaning, right-leaning, etc (on a scale). The display strategy for our NewsFeed can be popularity based, issue-based, or just timestamp-based.
4. Use the classification results from step 2 and crowdsourced results from step 3 to quantify, display, and track media bias. This way we precisely quantify and display how bias in issue filtering and issue framing changes overtime for every media outlet overall, at news/opinion level, and an issue level.

Challenges to anticipate with this idea:

One of the biggest issues that this product may face in the future is potential privacy/legal issues from news outlets and media companies. We are displaying articles anonymously coming from many big news companies and they would need some sort of accreditation or reference within the product. To tackle this, we plan to include a “View Publisher” toggle button that allows users to view a reference to the particular news outlet that generated the article that the user is looking at. Since viewing the reference of the article would result in bias, the option for the user to review the article would be disabled once this toggle button is pressed.

Our second challenge will be monetization. A monetization plan will be heavily beneficial to retrieve early-stage investors and bootstrap our finances. Through the development and deployment of our idea, the information and insights that we are collecting could be invaluable to media outlets. Through analytics and insights of what the consumer is inputting into the product, we can leverage information on specific articles and approach media outlets to potentially sell this information or act as the MTurk for them! Another convenient source of funding is through advertisements, either through Google Ads or other vendors. As our customer base grows (1000+ users), we can leverage our marketability and outreach by putting in ads into our product. The growth of our product will then correlate with the revenue that the product is bringing in.

I’d love to hear your thoughts on the product idea. I’m new to the Product Management world and would be grateful for your feedback! You can find me on LinkedIn or comment below. Thank you for reading!

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI