arrow_back Recommendation @ Scale
Using Locations for Online-Behaviour Prediction with Sparse Data
Submitted by Nishant Oli (@nishantoli) on Tuesday, 30 April 2019
Section: Crisp talk Technical level: Intermediate Session type: Lecture
User behaviour varies with their current location, as a consequence their engagement with online media (say Ads) varies with where they are; Knowing the type of location can help us target the user better and recommend better.
After the advent of GPS technology, it is easy and inexpensive to get location information from devices, but the data is often sparse for an ad network as they only get a partial view of user data. We present a novel algorithm to classify locations using the GPS data along with anonymous ad-request data such that it performs optimally with sparse GPS signals with no supervised data at all.
Our results show that the algorithm has a better prediction accuracy compared to existing methods and correlates well with the user behaviours.
- Introduction / Context Setting
- Why the need to classify the locations
- Location as the key attribute for recommendations
- Data exploration and modeling.
- Deploying the model using Spark
- How users interact differently at different locations
- Visualization of results for Bangalore and Bay Area.
- Results and take home message
Nishant Oli is a Data Scientist at InMobi where he works on user inferences and geo analytics. He graduated with a Master’s degree from IIIT Bangalore. Previously he worked at Siemens Corporate Research & Mazumdar Shaw Center of Translational Research in the field of Meta-Learning for computer vision and biomedical computer vision.