Main inspiration for this project came with the recent developments in understanding how human brain and language works. Many of the methods used in NLP and ML gives us a deeper and deeper insight in understanding human mind and how have we evolved as a species.

Regarding the exact application of ML, text summarization- the data and knowledge we as humanity produce grows in unimaginable speeds, if genius of past like Leonardo Da Vinci were positioned with the ability to reach almost all of the available human intelligence, now-days it's impossible. And this gap keeps and keeps widening, with no signs of stopping. We understand that the use of machine intelligence is necessary for well informed, rational and educated citizens of the future.

What it does

We've built a tool to help writers to worry less about manual advertising work and do things more efficiently! It allows writers to post just a link to their website, the app generates the perfect summary and ad for the case.

How we built it

We took a Recurrent neural net, as designed by Abi See link. This model combines the extractive and abstractive types of recurrent neural networks to generate summaries that introduce new words, while also having sementically correct sentences. The model was trained it with 90'000 CNN and Daily Mail articles to summarize text accurately. Wrapped it into Python backend with React webapp on the front.

Challenges we ran into

1.Finding the best NN model 2.Finding/extracting dataset that matches our goals 3.AWS instance crashing 4.Emotional instability of Bram

Accomplishments that we're proud of

Although we spent the first 24 hours in total chaos and panic, nothing seemed to work, But we managed to escape this and actually achieve the goals we set ourselves up to!

What we learned

AWS Finding a dataset that suits your needs can be challenging Insights into the inner workings of the RNN

What's next for Chestnut Delightful Stallion

Implementing new ML models to get better results, train them with dataset of ads that have received positive feedback. More features for application in A/B testing (multiple summary suggestions), different summaries (videos, images). Integrating Facebook, Twitter, Google Ads API.

Share this project: