Large-Scale Interactive Recommendation with Tree-Structured Policy Gradient

Authors: Haokun Chen, Xinyi Dai, Han Cai, Weinan Zhang, Xuejian Wang, Ruiming Tang, Yuzhou Zhang, Yong Yu3312-3320

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on carefully-designed environments based on two real-world datasets demonstrate that our model provides superior recommendation performance and significant efficiency improvement over state-of-the-art methods.
Researcher Affiliation Collaboration 1Shanghai Jiao Tong University 2Huawei Noah s Ark Lab
Pseudocode Yes Algorithm 1 Learning TPGR
Open Source Code Yes The implementation code3 of the TPGR is available online.3https://github.com/chenhaokun/TPGR
Open Datasets Yes Movie Lens (10M).1 A dataset consists of 10 million ratings from users to movies in Movie Lens website. Netflix.2 A dataset contains 100 million ratings from Netflix s competition to improve their recommender systems. 1http://files.grouplens.org/datasets/movielens/ml-10m.zip 2https://www.kaggle.com/netflix-inc/netflix-prize-data
Dataset Splits No For each dataset, the users are randomly divided into two parts where 80% of the users are used for training while the other 20% are used for test.
Hardware Specification Yes The experiments are conducted on the same machine with 4-core 8-thread CPU (i7-4790k, 4.0GHz) and 32GB RAM.
Software Dependencies No The paper mentions software components like "fully-connected neural network", "REINFORCE algorithm", and "simple recurrent unit (SRU)" but does not provide specific version numbers for any libraries, frameworks, or solvers used.
Experiment Setup Yes In our experiments, the length of an episode is set to 32 and the trade-off factor α in the reward function is set to 0.0, 0.1 and 0.2 respectively for both datasets. For TPGR, we set the clustering tree depth d to 2 and apply the PCA-based clustering algorithm with rating-based item representation when constructing the balanced tree... all the models adopt the neural networks with the same architecture which consists of three fully-connected layers with the numbers of hidden units set to 32 and 16 respectively