LLM-ESR: Large Language Models Enhancement for Long-tailed Sequential Recommendation

Authors: Qidong Liu, Xian Wu, Yejing Wang, Zijian Zhang, Feng Tian, Yefeng Zheng, Xiangyu Zhao

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To verify the effectiveness and versatility of our proposed enhancement framework, we conduct extensive experiments on three real-world datasets using three popular SRS models.
Researcher Affiliation Collaboration 1 School of Auto. Science & Engineering, MOEKLINNS Lab, Xi an Jiaotong University 2 City University of Hong Kong 3 Jarvis Research Center, Tencent You Tu Lab, 4 Jilin University 5 School of Comp. Science & Technology, MOEKLINNS Lab, Xi an Jiaotong University 6 Medical Artificial Intelligence Lab, Westlake University
Pseudocode Yes Due to the limited space, the algorithm lies in Appendix A.2 for more clarity. Algorithm 1 Train and inference process of LLM-ESR
Open Source Code Yes The implementation code is available at https://github.com/Applied-Machine-Learning-Lab/LLM-ESR.
Open Datasets Yes There are three real-world datasets applied for evaluation, i.e., Yelp, Amazon Fashion and Amazon Beauty. ... Yelp3 is the dataset that records the check-in histories and corresponding reviews of users. ... Amazon4 [42] is a large e-commerce dataset
Dataset Splits Yes As for the data split, the last item vnu and the penultimate item vnu 1 of each interaction sequence are taken out as the test and validation, respectively.
Hardware Specification Yes The hardware used in all experiments is an Intel Xeon Gold 6133 platform with Tesla V100 32G GPUs
Software Dependencies Yes the basic software requirements are Python 3.9.5 and Py Torch 1.12.0.
Experiment Setup Yes The hyper-parameters N and α are searched from {2, 6, 10, 14, 18} and {1, 0.5, 0.1, 0.05, 0.01}. ... the batch size and learning rate are set as 128 and 0.001 for all datasets. The embedding size is 128 for all baselines, while 64 for LLM-ESR. ... Then, we choose the Adam as the optimizer.