What it does
Our deepReccomender leverages cutting edge natural language processing techniques to recommend real-estate domain relevant news articles given an initial text. The crux of our deepReccomender is a machine learning model trained during the hackathon to assign multiple real-estate specific document tags to an input text. The tags, predicted from contextualized text representations, are compared in Jaccard similarity to known tags of documents in a fixed collection to produce reccomendations.
How I built it
Our solution leverages a three-tiered architecture powered by orchestrated containerization (docker). Raw data is shared with a Flask python API which assigns worker threads to requests received from a front end React powered application.
Challenges I ran into
- Working on a large project with multiple contributers and maintaining clean code. ## Accomplishments that I'm proud of
- We defined the problem we wanted to solve from the start, narrowed down the scope of solution and delivered.
What's next for deepRecommender
- Increased predictive performance
- Expansion to a larger database of indexed articles to recommend.