Conversion Prediction in User Click Streams
The goal of recommender systems is to provide users with a list of recommended items that match their preferences to help users navigate through large amounts of data (e.g., music to listen to, products to purchase, etc.). Session-aware recommender systems particularly leverage user actions (e.g., clicks, purchases, item views) over time to build user models and discover usage patterns across time, mostly based on (anonymous) user sessions.
In this master thesis, we are partnering with ADDITIVE to predict user actions on their platform. Particularly, we aim to investigate prediction algorithms for user sessions based on interaction logs as well as further offline (outside the platform) logs of user actions. As a first step, an exploratory data analysis aims to infer the characteristics of users and uncover potential groups of users sharing common features. Subsequently, we aim to investigate the following task: Given one or multiple sessions (click-streams) of a user, we aim to predict the conversion probability for a given user and a given item.