Leveraging Distribution Alignment via Stein Path for Cross-Domain Cold-Start Recommendation
Authors: Weiming Liu, Jiajie Su, Chaochao Chen, Xiaolin Zheng
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our empirical study on Douban and Amazon datasets demonstrates that Dis Align significantly outperforms the state-of-the-art models under the CDCSR setting. |
| Researcher Affiliation | Academia | Weiming Liu, Jiajie Su, Chaochao Chen, and Xiaolin Zheng Zhejiang University, Hangzhou, China {21831010,sujiajie,zjuccc,xlzheng}@zju.edu.cn |
| Pseudocode | No | The paper describes optimization steps in paragraph form but does not include structured pseudocode or an algorithm block. |
| Open Source Code | No | The paper does not provide any explicit statement or link indicating the availability of open-source code for the described methodology. |
| Open Datasets | Yes | We conduct extensive experiments on two popuarly used real-world datasets, i.e., Douban and Amazon. First, the Douban dataset [50, 51] has three domains, i.e., Book, Music, and Movie, which contains ratings, reviews, tags, and item details. Second, the Amazon dataset [49, 27] has two domains, i.e., Movies and TV (Movie), and CDs and Vinyl (Music). |
| Dataset Splits | No | The paper does not explicitly state training, validation, and test dataset splits by percentage, count, or specific cross-validation methodology. It mentions that the target domain data (RT) is only used for testing, but no explicit validation set details are given. |
| Hardware Specification | No | The paper does not provide specific details about the hardware specifications (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions 'Adam [16] as optimizer' but does not specify version numbers for any software dependencies, libraries, or programming languages used. |
| Experiment Setup | Yes | Hyper-parameter settings: We set batch size N = 256 for both the source and target domains. The latent embedding dimension is set to D = 128. For the rating prediction module, we set the balance hyper-parameters as η = 0.01 and ζ = 0.01, and number of cluster K = 5 for item unsupervised clustering. For the stein path alignment module, we set the moving step size as ϵ = 0.01 and the kernel bandwidth as σ = 0.5. For the proxy stein path alignment module, we set α = 0.1 and H = 64 according to Section 2.3.3. Finally, for the balance parameters, λSP and λP SP are first selected according to accuracy on Douban Movie Douban Book and then fixed as the best values, i.e., λSP = λP SP = 0.5. |