Skip to content

dmtrung14/tiktok-techjam24-factcheck

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

tiktok-techjam24-factcheck

TL;DR

Please like our project on devpost and Youtube. This project is a part of TikTok TechJam 2024

logo

Inspiration

Every day, millions of users spend hours browsing TikTok videos. The attention to short content is unequivocally high. Nonetheless, this also poses a challenge: how do news, science, and information show compete with music covers and comedy sketches to find a place in a world of short-form entertainment? FactCheck solves this problem by gamifying information videos to capture user attention. Through the short pop quiz, we hope to introduce a novel form of "TL;DR" for media files and deliver information in 10-15 seconds.

What it does

  • State-of-the-art: We adapted a state-of-the-art Generative AI model to extract the content of uploaded videos and provide a small pop quiz for users.
  • Extensive: The model is extensively trained to cover a wide range of topics, including but not limited to Science & Technology, Finance, Sports, Healthcare, and Politics.
  • Personalized: Additionally, the quiz will be provided in the user's preferred language to promote equal access to the tool.

How we built it

We used the Qwen2 model with 1.5 billion parameters to generate the quiz for the video. The model can adapt its response to a specified conversation history, enabling tailored responses and ensuring ethical constraints to generated questions.

Our application is developed using Flutter and made available on iOS 14+ operations systems, with a fastapi backend. Our model is preloaded, and generated quizzes are also cached in a Firestore (Firebase) database to enable quick access for other users.

We used TikTok Display API to get user information. This information is later used to cache responses and record answered quizzes on our database.

Challenges we ran into

  • Ethics of AI: How to tune the model to fit with the ethical values of AI: accuracy of information, denial of service for prohibited contents, and respect for users' privacy.
  • Data & Privacy: Collecting users' preferences without encroaching on their privacy.
  • Balancing Human & AI Expertise: How to use AI efficiently but still stay consistent with the educational intentions of the creators.

Accomplishments that we're proud of

  • Devised a new way to access information for TikTok users
  • Enabled fast generation for content creators using state-of-the-art Generative AI and fast access for users using streamlined backend protocols
  • Bridged the gap between short-form content and educational channels.

What we learned

  • How to apply Generative AI to solving a real-world pressing problem
  • TikTok FactCheck requires continuous development to match the trends and needs of the online world.
  • Balancing the use of AI with ethical constraints.

What's next for TikTok FactCheck

Applications are generally more useful when people can interact with each other. Some possible features are:

  • Build leaderboards and scoring systems for users to share.
  • Sharing quizzes/videos they find interesting and "challenge" their friends.

About Us

We are:

  • Trung Dang, Senior at University of Massachusetts Amherst
  • Giap Nguyen, Junior at University of Massachusetts Amherst
  • Hung Nguyen, Senior at University of Massachusetts Amherst

Releases

No releases published

Packages

 
 
 

Contributors