RepeatNet: A Repeat Aware Neural Recommendation Machine for Session-Based Recommendation
Authors: Pengjie Ren, Zhumin Chen, Jing Li, Zhaochun Ren, Jun Ma, Maarten de Rijke4806-4813
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We report on extensive experiments on three benchmark datasets. |
| Researcher Affiliation | Collaboration | 1Shandong University, Jinan, China 2University of Amsterdam, Amsterdam, The Netherlands 3Data Science Lab, JD.com, Beijing, China |
| Pseudocode | No | The paper describes the model architecture and mathematical formulations, but it does not contain explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | To facilitate reproducibility of the results in this paper, we are sharing the code used to run the experiments in this paper at https://github.com/Pengjie Ren/Repeat Net. |
| Open Datasets | Yes | We carry out experiments on three standard datasets, i.e., YOOCHOOSE, DIGINETICA, and LASTFM. YOOCHOOSE is a public dataset released by the Rec Sys Challenge 2015. DIGINETICA is released by the CIKM Cup 2016. LASTFM is released by (Celma 2010) and widely used in recommendation tasks (Cheng et al. 2017). |
| Dataset Splits | Yes | The splitting of the datasets are the same as (Li et al. 2017b). Table 1: Repeat ratio (%) on three benchmark datasets. Datasets Train Validation Test. Table 2: Statistics of three datasets (number of sessions and items). Dataset Training Validation Test Items. |
| Hardware Specification | Yes | The model is written in Chainer (Tokui et al. 2015) and trained on a Ge Force GTX Titan X GPU. |
| Software Dependencies | No | The paper states 'The model is written in Chainer (Tokui et al. 2015)' but does not provide a specific version number for Chainer or any other software dependencies. |
| Experiment Setup | Yes | We set the item embedding size and GRU hidden state sizes to 100. We use dropout (Srivastava et al. 2014) with drop ratio p = 0.5. We initialize model parameters randomly using the Xavier method (Glorot and Bengio 2010). We use Adam as our optimizing algorithm. For the hyper-parameters of the Adam optimizer, we set the learning rate α = 0.001, two momentum parameters β1 = 0.9 and β2 = 0.999, respectively, and ϵ = 10 8. We halve the learning rate α every 3 rounds. We also apply gradient clipping (Pascanu, Mikolov, and Bengio 2013) with range [ 5, 5] during training. To speed up the training and converge quickly, we use mini-batch size 1024 by grid search. |