← Back to the feed
AI & softwareResearch findings

Proposal of a Multi-Task Offline Reinforcement Learning Model for Advertising Recommendation

Original published 2025.06.29 14:05 KSTAdded to Blackdeck 2026.10.23 13:04 KST

What has changed?

A Markov Decision Process tailored to advertising scenarios is established, and a causal state encoder with causal attention mechanisms captures dynamic user兴趣,

Key points

  1. Proposes a multi-task offline RL model to address budget constraints and distributional shifts in advertising recommendation.
  2. Introduces a causal state encoder and causal attention mechanism to model dynamic user interests over time.
  3. Simultaneously optimizes channel recommendation and budget allocation through multi-task learning.

Connect it to your work

Simultaneously optimizes channel recommendation and budget allocation through multi-task learning, addressing sparsity and budget constraints.

What remains unproven

This research note is based on an abstract; check the original for test conditions and application scope.

Originals used for this note

Links open the original publisher in a new tab.

  1. arxivMulti-task Offline Reinforcement Learning for Online Advertising in Recommender Systems