Bridging the User-side Knowledge Gap in Knowledge-aware Recommendations with Large Language Models
Authors: Zheng Hu, Zhe Li, Ziyun Jiao, Satoshi Nakagawa, Jiawen Deng, Shimin Cai, Tao Zhou, Fuji Ren
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on three real-world datasets to validate the effectiveness of our method. Our approach achieves state-of-the-art performance compared to competitive baselines, particularly for users with sparse interactions. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, University of Electronic Science and Technology of China 2Graduate School of Information Science and Technology, University of Tokyo 3Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China |
| Pseudocode | No | The paper describes the method using mathematical formulas and conceptual steps in the 'Method' section, but does not include a clearly labeled pseudocode or algorithm block. |
| Open Source Code | Yes | Code https://github.com/laowangzi/CIKGRec |
| Open Datasets | Yes | We conduct extensive experiments on three real-world datasets: DBbook2014 (Cao et al. 2019), Book Crossing (Dong et al. 2017), and Movie Lens-1M (Noia et al. 2016) |
| Dataset Splits | Yes | We construct the training, validation, and test sets by randomly splitting the historical behavior of each user in a ratio of 7 : 1 : 2. |
| Hardware Specification | No | The paper mentions 'we adopt the gpt-3.5-turbo-0125 model as the LLM', which is a software model, but does not specify any hardware details like GPU/CPU models, memory, or specific computing resources used for experiments. |
| Software Dependencies | Yes | In this paper, we adopt the gpt-3.5-turbo-0125 model as the LLM. |
| Experiment Setup | Yes | For our model, we search the learning rate from 0.0001 to 0.005. We test the user interest reconstruction initial mask rate α from 0.05 to 0.15 and the maximum mask rate ω from 0.35 to 0.95. In this paper, we adopt the gpt-3.5-turbo-0125 model as the LLM. The required convergence epoch Λ is chosen in {80, 160, 320}. ... an embedding dimension of 64 ... An early stopping strategy with a maximum of 2000 epochs is applied to both the baselines and our model. |