Efficient Ordered Combinatorial Semi-Bandits for Whole-Page Recommendation

Authors: Yingfei Wang, Hua Ouyang, Chu Wang, Jianhui Chen, Tsvetan Asamov, Yi Chang

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide experimental results on both synthesis datasets and Yahoo! Front Page Webscope datasets. We demonstrate the ability of our approach to solve real-world problems that can be modeled as ordered contextual combinatorial semi-bandits and that have not been throughly studied before.
Researcher Affiliation Collaboration Yingfei Wang,1 Hua Ouyang,2 Chu Wang,3 Jianhui Chen,4 Tsvetan Asamov,5 Yi Chang6 1Department of Computer Science, Princeton University, yingfei@cs.princeton.edu 2Apple Inc., hua ouyang@apple.com 3Nokia Bell Labs, chu.wang@nokia-bell-labs.com 4Yahoo Research, jianhui@yahoo-inc.com 5Department of Operations Research and Financial Engineering, Princeton University, tasamov@princeton.edu 6Huawei Research America, yichang@acm.org
Pseudocode Yes Algorithm 1: Thompson sampling
Open Source Code No The paper does not contain any explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We provide experimental results on both synthesis datasets and Yahoo! Front Page Webscope datasets. We use Yahoo! Webscope datasets (Yahoo! Webscope 2009) ... http://webscope.sandbox.yahoo.com/.
Dataset Splits No The paper mentions using 'training data' and evaluates performance on datasets, but it does not provide explicit details about training, validation, and test dataset splits (e.g., percentages or sample counts) needed for reproduction. It mentions 'Yahoo! Webscope datasets' but no specific split method.
Hardware Specification Yes The experiments are ran on a Linux server [Intel(R) Xeon(R) CPU X5650 2.67GHz, 8G memory].
Software Dependencies No The paper mentions software components like 'linear programming solver' and 'Bayesian logistic regression' but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes For Gaussian processes, we start with a mean vector of zeros and choose the Squared Exponential Kernel k(x, x ) = α2 exp( β1(k k )2 β2(m m )2) with α = 100, β1 = 0.2, β2 = 0.1 in the experiments. For logistic regression, the 'Regularization parameter λ > 0 mj = 0, qj = λ' is mentioned. Additionally, 'we model this behavior by refreshing the system after every 10 minutes'.