Rosetta News

• 

  Site Homepage

• 

  Machine in Action

Inspiration

With all of the misinformation and propaganda spread today on all sides, it is important to go straight to the source. This is why we created Rosetta News.

What it does

Rosetta News takes an article input and uses a few Natural Language Processing models to detect the language, translate it, and summarize it. This allows you to get correctly translated news quickly.

How we built it

For the NLP models, we had a summarization model, a detection model, and a translation model, all of which were from Hugging Face. For the summarization model we used facebook/bart-large-cnn, detection we used papluca/xlm-roberta-base-language-detection, and for translation we used Helskinki-NLP.

Challenges we ran into

Challenges we ran into were significant and frequent. When starting on this journey, the goal was to train our own model and apply it to the dataset. Unfortunately, the ram on our computers were not sufficient to train our own model (despite hours) of training, so pre-trained was the way to go. Due to other various errors, fine-tuning this model did not go so well either so we ended up using pipelines. In terms of web development some issues we faced implementing the language detection model on the website, so we ended up making buttons. Some other web development challenges we faced included navigating through pages and files in VS Code, having the correct syntax with Tailwind CSS, and understanding flexbox and divs.

Accomplishments that we're proud of

Two of the people on our team learned web development for the first time ever during this hackathon and it resulted in a great website. Furthermore, we are proud that we were able to exactly implement our idea into the website as we had planned. We are most proud of that for every obstacle we faced, we were able to overcome it with an alternate solution which resulted in a cohesive project.

What we learned

We learned so much about web development and NLP. We learned about React, Tailwind, and specific hugging face libraries and NLP techniques. Some techniques we learned about were abstractive vs extractive summarization, translative models, and language detection models.

What's next for Rosetta News

There is a lot upcoming for Rosetta News. Firstly, we would like to add even more languages to increase accessibility and applications. Next, we would like to train our own models and fine-tune them for optimal results. Another thing we would like to implement is a web scraper that would allow you to place the link in the website and automatically pull the text. We were not able to implement this feature for this hackathon since the majority of news sites have paywalls which prevent you from doing this.

Built With

• huggingface
• natural-language-processing
• railway
• react
• tailwind