Costar:AirBNB analysis- A Novel Tagger and NLP based model

AirBNB commercial listings- A Novel Tagger and NLP the superhero to Ishwar's Secret model's rescue!

Comment

Inspiration

Words have always fascinated me, and I've had a deep connection with them which is why I love languages and NLP. If properly captured and understood, we can truly do great things. The tagger is my favourite brainchild and I've had it for years. I tried multiple evaluation mechanisms including Flesh and Gunning scores and discovered the simplest solution is the best.

What it does

This is a full scale exploration of the Boston AirBNB dataset, attempted in a different way in the hope of discovering something new. It identifies visually and aesthetically pleasing words to possibly help marketing and advertising. Additionally, we have exploited natural language processing on various textual attributes and worked a way to combine then and generate much more useful representations capable of better capturing a relationship to prices.

How I built it

I ran several setups along various stages and ended up with the most optimal one.

Challenges I ran into

Visually and aesthetically pleasing words is a seemingly abstract concept and the time and effort spent in thinking how to map the abstract to something concrete which a computer understands. In the end, the solution was so simple and elegant but powerful.

Word vectors for this task needed to be approximated and aggregated to an usable fashion. This took some experimenting and working around. Finally, I settled on averaging them. Also, individual classifiers were not good enough. Testing various ensemble classifiers and finding their best configuration took some time.

Accomplishments that I'm proud of

With regard to this specifically? It's always going to be the tagging part :D

I'm proud that I got multiple steps further in this hackathon and it was time well spent. I loved how responsive the organizers and you guys at the company were.

I was a national contestant for India's Biggest Networking Championship and I had the pleasure of learning something new at every hackathon I attended and earned prizes each time.

I'm proud of the fact that I wake up each day with new challenges and things to learn. I'm proud that I'm submitting this and everything has been because of people I've met along the way. Thank you for taking the time to read this!

What I learned

What's next for Costar:AirBNB analysis- A Novel Tagger and NLP based model

I've listed that in the Python notebook! I plan to leverage deep learning and generate vectors specifically tuned to this field using more compute power.

Built With

Share this project:

Updates