Relevance Optimization for Check-In Candidate Lists
Check-in services like Foursquare and Nokia Pulse are attracting millions of users. In check-in apps the user is usually first asked to select a check-in place from a list of nearby candidate places. In this talk I will discuss how to optimize the relevance of such check-in candidate lists. I will explain the ingredients of different data-driven relevance models. These models mix several relevance signals like geo-distance and number of historic check-ins on a candidate place. I am also talking about personalization. The models are conceptually simple and I will try to explain them in a way so that audience without statistics background can understand and reimplement them. I will describe how the model parameters are automatically tuned with machine learning. I will show how to implement the model in production using SOLR. And at the end I will present evaluation results based on millions of real check-ins from Nokia users. Beyond the specific use-case, you can learn many general concepts from this talk on relevance optimization in a location-based setting, machine learning, parameter tuning, and model evaluation.
Watch the video of Steffen Bickel's talk here.