Mladen Jovanovic
Data Scientist | Bravo Systems
I love finding question/answers in the data itself.
I’m part of the Data Science team at Bravo Systems, a strong performance driven digital media
company. Previously worked as a Lead Bioinformatics Analyst at Seven Bridges, dealing with
various genomics data driving public and private healthcare research.
In past professional life I worked as a software/web developer. It was bumpy but enjoyable a
decade long journey.
I love freediving (member of the serbian national team). When not submerged underwater, I
also enjoy freeclimbing and brewing my own beer.
Last, but certainly not the least,
I value a good team more than anything else.
PyCon Balkan 2018 Talks
Recommender systems are considered as an inevitable part of any system that is offering some
kind of products/services to the final user. System complexity can range anywhere from simple
webstores displaying few dozen articles to big web applications with complex architecture
offering millions of items to as many users.
As the systems grew bigger, rising need of efficiently handling and presenting that amount of
data in a meaningful manner emerged. Early steps were focused on good categorization of
available items, improving browsing capabilities and providing intelligent search. But the main
aspect of user engagement remained the same. User had to take action in order to have the
items presented to them. This is where the recommender systems came in, radically changing
the way how the end users are experiencing the whole platform. Instead of browsing the
platform in order to discover something new and relevant, the items are directly presented to the
user based on previous experience, not just from their particular history of actions but from the
experience of the whole user community as well.
Over the years, variety of techniques have been developed for building recommender systems,
including content-based, collaborative each having their own pros and cons. We’ll cover both of
them offering relative insight how to treat some of the challenges involved in dealing with this
kind data like matrix sparsity, dimensionality reduction etc.