Inspiration

The biggest irony today is despite the advent of the internet, students and adults are more oblivious than ever to world events, and one can easily understand why. Of course, Facebook, YouTube, and League will be more interesting than reading Huffington Post; coupled with the empirical decrease in the attention span of younger generations, humanity is headed towards disaster.

What it does

Our project seeks to address this crisis by informing people in a novel and exciting way. We create a fully automated news extraction, summarization, and presentation pipeline that involves an AI-anime character news anchor. The primary goal of our project is to engage and educate an audience, especially that of younger students, with an original, entertaining venue for encountering reliable news that will not only foster intellectual curiosity but also motivate them to take into deeper consideration of relevant issues today, from political events to global warming.

The animation is basically a news anchor talking about several recent news, where related news is discussed in a short blurb.

Demo Video Explanation

The demo video generally performs well, except for the first few seconds and the Putin/Taliban part. This is because the clusters are too small so many clusters get merged together as our kmeans has fixed number of clusters. A quick fix is to simply calculate the internal coherence of the cluster and filter based on that. more advanced methods can be based on those described in the Scatter Gather paper by Karger et al.

How we built it

News Summarization

For extraction and summarization, our first web scrapes news articles from trusted sources (CNN, New York Times, Huffington Post, Washington Post, etc…) to obtain the texts of recent news articles. Then it generates a compact summary of these texts using an in-house developed two-tier text summarization algorithm based on state-of-the-art natural language processing techniques. The algorithm first does an extractive summarization of individual articles. Next, it computes an overall 'topic feature' embedding. This embedding is used to cluster related news, and the final script is generated using these clusters and DL-based abstractive summarization.

News Anchor Animation

Furthermore, using the google cloud text-to-speech API, we generate speech with our custom pitch and preferences and we then have code that generates a video using an image of any interesting, popular anime character. In order for the video to feel natural to the audience, we accounted for accurate lip and facial movement; there are calculations made using specific speech traits of the .wav file that produces realistic and not only educational but also humorous videos that will entertain the younger audience.

Audience Engagement

Moreover, we wrote code using the Twitter API to automate the process of uploading videos to our Twitter account MinervaNews which is integrated within the project’s server that uploads a video initially when the server starts and automatically generates every 24 hours after a new video using the new articles from the sources.

What's next for Minerva Daily News Reporter

Our project will have a lasting impact on the education of an audience ranging in all age groups. Anime is one great example of a venue that can broadcast news, and we selected anime characters as a humorous and eye-catching means to educate the younger audience. Our project and its customization allow for the possibility of new venues and greater exploration of making education more fun and accessible to a vast audience. We hope to take our project further and add more animations as well as more features.

Challenges

Our compute platform, Satori has a unique architecture called IBM ppe64le that makes package and dependency management a nightmare.

What we learned

8 hours in planning = 24 hours in real time.

Github

https://github.com/gtangg12/liszt

Share this project:

Updates