A Domain Generalization Perspective on Listwise Context Modeling

Authors: Lin Zhu, Yihong Chen, Bowen He5965-5972

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our techniques on benchmark datasets, demonstrating that QILCM outperforms previous state-of-the-art approaches by a substantial margin.
Researcher Affiliation Industry Lin Zhu, Yihong Chen, Bowen He Ctrip Travel Network Technology Co., Limited. {zhulb, yihongchen, bwhe}@ctrip.com
Pseudocode No The paper describes the model architecture using equations and descriptive text, but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper provides links to the implementations of baseline methods (DCM and DLCM) but does not provide a link or explicit statement for the open-source code of their proposed QILCM.
Open Datasets Yes We used two largescale LETOR datasets: Istella-S3 (Lucchese et al. 2016) and Microsoft Letor 30K4 (Qin and Liu 2013). ... For this task, we used Airline Itinerary5, which is an anonymized version of the dataset used in (Mottini and Acuna-Agost 2017)...
Dataset Splits Yes Each dataset is split in train, validation and test sets according to a 60%-20%-20% scheme.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions the use of the Adam algorithm and an open-source implementation of Lambda MART, but does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes More specifically, the dimensions of the nonlinear transformations (1) in the Input Encoder were fixed as 100, while MLPs used in (3) and (9) consist of 2 hidden layers with either 256 or 128 ELUs. The models were trained with the Adam algorithm (Kingma and Ba 2014) with a learning rate of 0.001, batch size of 80. Training generally converged after less than 100 passes through the entire training dataset.