What is it about?
We want to choose multiple adverts to display to web users. If the adverts are too similar to each other then some will redundant. Therefore the set needs to be appropriately diverse while still matching the users' interests. The users' preferences vary and cannot be known exactly, but diverse adverts make it more likely that one is of interest. The quality of the adverts is also unknown but we learn about which are best through experimentation and observing user responses.
Featured Image
Why is it important?
The problem of choosing multiple web elements to display to a user is common in many applications beyond advertising, for example in recommender systems where the elements might be news stories, retail items, or movies. Making these element sets appropriately diverse is a current challenge in the area. The model in this work incorporates uncertainty about the users' preferences, and our methods respond with element sets with the correct diversity for balancing short term optimisation and longer term learning.
Perspectives
Contextual bandit problems assume the context (information specific to the current decision) is known and fixed. Our model and solution includes a richer form of context which is a probability distribution, which allows extra uncertainty about the current state of the system to be incorporated into the problem.
James Edwards
Lancaster University
Read the Original
This page is a summary of: Selecting multiple web adverts: A contextual multi-armed bandit with state uncertainty, Journal of the Operational Research Society, February 2019, Taylor & Francis,
DOI: 10.1080/01605682.2018.1546650.
You can read the full text:
Contributors
The following have contributed to this page
