RESBERT - Real Estate Extractive Summarization with BERT

Inspiration

Commercial Real Estate is a challenging fast paced industry where time matters. Real Estate Agents often need to keep up with the latest real estate news to stay competitive; however, fully reading through every real estate news article is far too of a time consuming process. We decided to build RESBERT to put the power of state of the art extractive summarization modeling into the hands of your average real estate agent for real-time news article summarization.

What it does

RESBERT helps users get straight to the point of real estate news articles through state of the art extractive text summarization models, with specially fine tuned transformer layers for the task. We packaged RESBERT in a modern and responsive web app that allows users to simply copy paste entire long, multi-paragraph real estate news articles and have them instantaneously summarized into only a handful of sentences in real-time.

How we built it

We built it using PyTorch, Python, and Flask!

Challenges we ran into

Remapping the sentence probability from the output of RESBERT to an actual summarization.

Accomplishments that we're proud of

We're proud that we were able to specially fine tune our summarization model using the CoStar Commercial Real Estate dataset for enhanced performance.

What we learned

Transfer learning can be used to learn latent representations of entire documents from a pre-trained BERT model with weights that represent latent sentence embeddings.