Optimizing Long-term Value for Auction-Based Recommender Systems via On-Policy Reinforcement Learning.
Ruiyang XuJalaj BhandariDmytro KorenkevychFan LiuYuchen HeAlex NikulkovZheqing ZhuPublished in: RecSys (2023)
Keyphrases
- recommender systems
- long term
- reinforcement learning
- optimal policy
- policy search
- short term
- action selection
- markov decision process
- exploration exploitation tradeoff
- collaborative filtering
- policy gradient
- markov decision processes
- partially observable
- actor critic
- partially observable environments
- function approximation
- policy iteration
- state space
- reward function
- reinforcement learning problems
- control policies
- state action
- markov decision problems
- partially observable markov decision processes
- reinforcement learning algorithms
- policy evaluation
- matrix factorization
- action space
- function approximators
- temporal difference
- user modeling
- state and action spaces
- control policy
- model free
- information overload
- approximate dynamic programming
- user profiling
- user profiles
- continuous state
- decision problems
- pricing mechanism
- continuous state spaces
- online auctions
- transition model
- transfer learning
- combinatorial auctions
- user preferences
- dynamic programming
- product recommendation
- cold start problem
- average reward
- optimal control
- electronic commerce
- rl algorithms
- reinforcement learning methods
- auction mechanisms
- recommendation algorithms
- infinite horizon
- long run
- evaluation function
- partially observable domains
- user model