Dual Sparse Attention Network For Session-based Recommendation

Authors: Jiahao Yuan, Zihan Song, Mingyou Sun, Xiaoling Wang, Wayne Xin Zhao4635-4643

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on two real public datasets show that the proposed method is superior to the state-of-the-art sessionbased recommendation algorithm in all tests and also demonstrate that not all actions within the session are useful.
Researcher Affiliation Academia 1 Shanghai Key Laboratory of Trustworthy Computing, East China Normal University, Shanghai, China 2 Shanghai Institute of Intelligent Science and Technology, Tongji University, Shanghai, China 3 Gaoling School of Artificial Intelligence, Renmin University of China
Pseudocode No The paper does not contain explicit pseudocode or algorithm blocks.
Open Source Code Yes To make our results reproducible, we have published our code on https://github.com/SamHaoYuan/DSANForAAAI2021.
Open Datasets Yes To evaluate the effectiveness of the proposed model, we use two real-world representative datasets, i.e., Diginetica1 and Retailrocket2. For simplicity, we name them DN and RR. The DN dataset comes from CIKM Cup 2016, and we only used the released transaction data. The RR is a dataset on a Kaggle contest published by an e-commerce company, which contains the user s browsing activity within six months. Both two datasets are publicly available. 1http://cikm2016.cs.iupui.edu/cikm-cup 2https://www.kaggle.com/retailrocket/ecommerce-dataset
Dataset Splits No The paper mentions that hyperparameters are optimized via grid search on each dataset, implying a validation set was used, but it does not explicitly state the specific validation split percentages or sample counts.
Hardware Specification No No specific hardware (e.g., GPU/CPU models, memory) used for running the experiments is mentioned.
Software Dependencies No The paper states, 'We implement the model by Pytorch,' but it does not specify a version number for Pytorch or any other software dependencies.
Experiment Setup Yes In this paper, all hyper-parameters are optimized via grid search on each dataset, respectively. According to the experimental results, the optimal hyperparameters are {η : 0.001, ε : 0.5, wk : 20} on two datasets, where η is the learning rate, ε is the dropout rate and wk is the normalized weight. We use Adam as the model optimizer, and explore the case that embedding dimension d = 100 for a fair comparison, which is the hyper-parameter in the previous work. We implement the model by Pytorch, and the mini-batch settings are {batch size: 512, epoch: 50}.