Recommender Forest for Efficient Retrieval
Authors: Chao Feng, Wuchao Li, Defu Lian, Zheng Liu, Enhong Chen
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental studies are performed on six popular recommendation datasets: with a significantly simplified training cost, Rec Forest outperforms competitive baseline approaches in terms of both recommendation accuracy and efficiency. |
| Researcher Affiliation | Collaboration | 1School of Computer Science and Technology University of Science and Technology of China, Hefei, China 2 Microsoft Research Asia, Beijing, China |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/wuchao-li/Rec Forest. |
| Open Datasets | Yes | We evaluate the Rec Forest with six real-world recommendation datasets, which can be downloaded from the url*. The datasets are Movie Lens 10M (abbreviated as Movie), Amazon Books (abbreviated as Amazon), Tmall Click (abbreviated as Tmall), Gowalla Check-in Dataset (abbreviated as Gowalla), Microsoft News Dataset (abbreviated as MIND). The url* is https://drive.google.com/drive/folders/1ahi Lmz U7c GRPXf5q GMqt AChte2e Yp9g I. |
| Dataset Splits | Yes | In each dataset, we randomly choose 10% users as validation users, 10% users as test users, and all the left users as training users. |
| Hardware Specification | Yes | These experiments are done on a Linux server with Tesla V100 GPUs. |
| Software Dependencies | No | The paper mentions implementing SCANN and IPNSW with PyTorch but does not specify a version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | The latent dimensionality is set to 96 in all methods. The beam size for any beam search is set to 100 unless specified. To learn models, the learning rate is all set to 1e-3 with exponential decay. The items representations for constructing trees are obtained from item embedding of the pre-trained DIN on each dataset. Rec Forest uses 2 trees on the Movie and MIND, and 4 trees on other datasets. |