Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Efficient Knowledge Transfer in Federated Recommendation for Joint Venture Ecosystem
Authors: Yichen Li, Yijing Shan, YI LIU, Haozhao Wang, Cheng Wang, wangshi.ww, Yi Wang, Ruixuan Li
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments A thorough experimental study has been conducted to assess the performance of the FR-JVE in two popular scenarios with four recommendation datasets: (1) Rating Prediction: Movie Lens-100K (ML-100K)[7] and Movie Lens-1M (ML-1M). (2) Top-K Recommendation: Last FM-2K(Last FM) [3] and QB-article [41]. The details of these datasets are outlined in Appendix D. |
| Researcher Affiliation | Collaboration | Yichen Li1, Yijing Shan1, Yi Liu2 Haozhao Wang1, Cheng Wang1, Wei Wang2, Yi Wang2, Ruixuan Li1 1School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China 2Chongqing Ant Consumer Finance Co., Ltd, Ant Group, Chongqing, China EMAIL |
| Pseudocode | Yes | A Algorithm Flow for FR-JVE Algorithm 1: FR-JVE Input :T: communication round; K: client number; Dk: local dataset for the client k; uk,e: local user embedding; vk,e: local item embedding; ϕk( ): local bridge function; fk: local inversion model. Output :ve: Item Embedding. |
| Open Source Code | No | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We could provide our code if required. |
| Open Datasets | Yes | A thorough experimental study has been conducted to assess the performance of the FR-JVE in two popular scenarios with four recommendation datasets: (1) Rating Prediction: Movie Lens-100K (ML-100K)[7] and Movie Lens-1M (ML-1M). (2) Top-K Recommendation: Last FM-2K(Last FM) [3] and QB-article [41]. The details of these datasets are outlined in Appendix D. |
| Dataset Splits | Yes | Unless otherwise mentioned, to build the system with partially overlapped users and items, we first assign each client with random private users and items and then use the Dirichlet distribution Dir(α = 10) [32] to distribute shared users to yield data heterogeneity for all datasets where a smaller α indicates higher data heterogeneity. Here, we report the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as the metrics for the rating prediction [28] and Hit Rate (HR@K) and Normalized Discounted Cumulative Gain (NDCG@K) for the top-k recommendation [27, 21]. In this work, we set K = 10. We illustrate all the settings with all the benchmark parameters in Appendix D. ... Table 11: Experimental Details. Analysis of various considered settings of different datasets in the experiments section. Attributes Rating Prediction Top-K Recommendation ML-100K ML-1M Last FM QB-article Ratings 100,000 1,000,209 92,834 348,736 Users 943 6,040 1,892 24,516 Items 1,682 3,952 17,632 7,355 Sparsity s = 93.70% s = 95.81% s = 99.72% s = 99.81% Batch Size b = 32 b = 64 b = 256 b = 256 Learning Rate l = 0.001 l = 0.001 l = 0.04 l = 0.01 Local Users 100 600 200 2,400 Local Items 200 400 1,700 700 Shared Users 443 3,040 892 12,516 Client numbers C = 5 C=5 C=5 C=5 Top-K Search K = 200 K=3000 K=1600 K=7000 Local training epoch E = 5 E = 5 E = 5 E = 5 Client selection ratio k = 0.6 k = 0.6 k = 0.6 k = 0.6 Communication Round T = 200 T = 200 T = 200 T = 200 |
| Hardware Specification | No | The computation is completed in the HPC Platform of Huazhong University of Science and Technology. |
| Software Dependencies | No | We use Adam as an optimizer with a linear learning rate schedule. We set the remaining parameters according to the values in the original open-source code. |
| Experiment Setup | Yes | Table 11: Experimental Details. Analysis of various considered settings of different datasets in the experiments section. Attributes Rating Prediction Top-K Recommendation ML-100K ML-1M Last FM QB-article Ratings 100,000 1,000,209 92,834 348,736 Users 943 6,040 1,892 24,516 Items 1,682 3,952 17,632 7,355 Sparsity s = 93.70% s = 95.81% s = 99.72% s = 99.81% Batch Size b = 32 b = 64 b = 256 b = 256 Learning Rate l = 0.001 l = 0.001 l = 0.04 l = 0.01 Local Users 100 600 200 2,400 Local Items 200 400 1,700 700 Shared Users 443 3,040 892 12,516 Client numbers C = 5 C=5 C=5 C=5 Top-K Search K = 200 K=3000 K=1600 K=7000 Local training epoch E = 5 E = 5 E = 5 E = 5 Client selection ratio k = 0.6 k = 0.6 k = 0.6 k = 0.6 Communication Round T = 200 T = 200 T = 200 T = 200 |