Content discovery is one of the common browsing use cases since the first days of the internet. Similar in volume and frequency to content search, it stems from curiosity and represents a basic though different need of online users – to explore and uncover new things, as well as learn more about things they already like. Since then, recommender systems in general and content recommendation engines in particular have been in the focus of the machine learning and information retrieval research communities.
In this talk we will call out several challenges in the art of recommendation technology, research challenges as well as big-data implementation challenges. We will touch on related algorithmic and learning problems such as user and item cold start handling; and on the more technical end tradeoffs between online vs. offline scoring, ranking and candidate selection. Analytical and scale difficulties such as creating proxies for reader engagement, dynamic schemas under new interaction scenarios, click-through probability estimation, and recommending item sets and sequences will also be discussed.
Ronny Lempel - Yahoo! Research
Ron Karidi - Outbrain.