Improving LLMs for Recommendation with Out-Of-Vocabulary Tokens
Authors: Ting-Ji Huang, Jia-Qi Yang, Chunxu Shen, Kai-Qi Liu, De-Chuan Zhan, Han-Jia Ye
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate META ID on five downstream recommendation tasks: sequential recommendation, direct recommendation, rating prediction, explanation generation, and review related tasks. We analyze the influence of critical components in META ID and assess the ID representations through visualization and our proposed metrics. Table 1 presents our findings for sequential recommendation task. |
| Researcher Affiliation | Collaboration | 1School of Artificial Intelligence, Nanjing University 2National Key Laboratory for Novel Software Technology, Nanjing University 3We Chat Technical Architecture Department, Tencent Inc. 4Software Institute, Nanjing University. |
| Pseudocode | No | The paper describes the methodology, including meta-path-based embedding and OOV token generation, through descriptive text and diagrams, but does not include a formally structured pseudocode block or algorithm. |
| Open Source Code | Yes | Code is available at https: //github.com/Tingji2419/META-ID. |
| Open Datasets | Yes | We evaluate our META ID framework on three public real-world datasets from the Amazon Product Reviews dataset (Ni et al., 2019), focusing specifically on Sports, Beauty, and Toys. The datasets are processed following the methodology in P5 (Geng et al., 2022). 2https://nijianmo.github.io/amazon |
| Dataset Splits | Yes | For rating, explanation, and review task families, we randomly split each dataset into training (80%), validation (10%) and testing (10%) sets, and ensure that there is at least one instance included in the training set for each user and item. |
| Hardware Specification | Yes | For LLM fine-tuning, we pre-train T5 for 10 epochs using Adam W optimizer on two NVIDIA RTX 3090 GPUs with a batch size of 64... We use the lora (Hu et al., 2022) technique to fine-tune the token embedding layer and linear head layer of LLa MA2-7b for 1 epochs using Adam W optimizer on two NVIDIA RTX A6000 GPUs... |
| Software Dependencies | No | The paper mentions specific LLM models like T5 (Raffel et al., 2020b) and LLa MA2-7b (Touvron et al., 2023), and techniques like LoRA (Hu et al., 2022), but does not specify version numbers for underlying programming languages or libraries (e.g., Python, PyTorch/TensorFlow versions). |
| Experiment Setup | Yes | For LLM fine-tuning, we pre-train T5 for 10 epochs using Adam W optimizer on two NVIDIA RTX 3090 GPUs with a batch size of 64, a peak learning rate of 1e 3. We apply warm-up for the first 5% of all training steps to adjust the learning rate, a maximum input token length of 1024. We use the lora (Hu et al., 2022) technique to fine-tune the token embedding layer and linear head layer of LLa MA2-7b for 1 epochs using Adam W optimizer on two NVIDIA RTX A6000 GPUs with a batch size of 28, a peak learning rate of 1e 5, the lora attention dimension of 16 and the alpha parameter of 32. |