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 [1].

Multifaceted User Modeling in Recommendation: A Federated Foundation Models Approach

Authors: Chunxu Zhang, Guodong Long, Hongkuan Guo, Zhaojie Liu, Guorui Zhou, Zijian Zhang, Yang Liu, Bo Yang

AAAI 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental validations on benchmark datasets demonstrate the superior performance of our proposed method. We evaluate the performance of our method on real-world industrial datasets. Experimental results show that our proposed MRFF can empower various foundation model architectures and achieve superior performance compared to state-of-the-art baselines on widely-used click-through rate (CTR) prediction task. Additionally, we conduct comprehensive experiments to analyze our model s efficacy in terms of user-grouping capability. Furthermore, we demonstrate the practical feasibility of MRFF by assessing its efficiency and privacy-preserving capabilities.
Researcher Affiliation Collaboration 1 College of Computer Science and Technology, Jilin University, China 2 Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, China 3 Australian Artificial Intelligence Institute, FEIT, University of Technology Sydney 4 Kuaishou Technology 5 Institute for AI Industry Research, Tsinghua University
Pseudocode No The paper describes the method using mathematical formulations and descriptive text, but does not include a clearly labeled pseudocode or algorithm block.
Open Source Code Yes Code https://github.com/Zhangcx19/AAAI-25-MRFF
Open Datasets Yes We conduct experiments on three practical datasets: Kuai Rand-Pure (Gao et al. 2022), Kuai SAR-R and Kuai SAR-S (Sun et al. 2023).
Dataset Splits Yes This paper focuses on the click-through rate (CTR) prediction task and we adopt the prevalent leave-one-out dataset split, following the setting in (Kang and Mc Auley 2018).
Hardware Specification No The paper discusses the efficiency of the model in terms of parameter size and testing time on client devices, but does not specify the hardware used for conducting the experiments.
Software Dependencies No All experiments are implemented using the Py Torch framework and repeated 5 times, with average results reported to ensure statistical reliability.
Experiment Setup Yes For two non-sequential baselines Fed NCF and Fed PA, we fix batch size as 1,024. For three transformer-based baselines and our enhanced versions, we set transformer blocks as 2 for a fair comparison. Besides, we set the total communication rounds as 500 for all models to ensure convergence.