Graph-Enhanced Multi-Task Learning of Multi-Level Transition Dynamics for Session-based Recommendation
Authors: Chao Huang, Jiahui Chen, Lianghao Xia, Yong Xu, Peng Dai, Yanqing Chen, Liefeng Bo, Jiashu Zhao, Jimmy Xiangji Huang4123-4130
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three real-world datasets demonstrate the superiority of MTD as compared to state-of-the-art baselines. We highlight key contributions of this paper as follows: Our extensive experiments on three real-world datasets demonstrate that MTD outperforms different types of baselines in yielding better recommendation results. Also, we show the efficiency of our developed model as compared to representative competitors and perform case studies with qualitative examples to investigate the interpretation capability of our MTD model. |
| Researcher Affiliation | Collaboration | Chao Huang1, Jiahui Chen2, Lianghao Xia2, Yong Xu2,3,4*, Peng Dai1, Yanqing Chen1, Liefeng Bo1, Jiashu Zhao5, Jimmy Xiangji Huang6 1JD Finance America Corporation, USA 2South China University of Technology, China, 3Peng Cheng Laboratory, China 4Communication and Computer Network Laboratory of Guangdong, China 5Wilfrid Laurier University, Canada, 6York University, Canada |
| Pseudocode | No | The paper does not contain any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Source code is released at the link https://github.com/sessionRec/MTD. |
| Open Datasets | Yes | Yoochoose Data1. This data comes from an online retailing site to log half year of user clicks (released by Recsys 15 Challenge). Following the pre-processing strategies in (Li et al. 2017; Liu et al. 2018), the sessions with the length of 2 and items with the appearing frequency of 5 are kept in the training and test set. Diginetica Data2. This data is collected from the CIKM Cup platform 2016 which records the user clicks from the time period of six months. To be consistent with the settings in (Wu et al. 2019; Liu et al. 2018), we do not include the sessions that contains single clicked item. Sessions in the test set are generated from the last week. Retailrocket Data3. It contains the user browse data from another e-commerce company. Following the same settings in (Xu et al. 2019), we filter out the items with the browsed frequency less than 5 and sessions with the length of less than 2. We set the data from the last week for test and the remaining part for training. 1http://cikm2016.cs.iupui.edu/cikm-cup 2http://2015.recsyschallenge.com/challenge.html 3https://www.kaggle.com/retailrocket/ecommerce-dataset |
| Dataset Splits | Yes | The data statistics with training/test detailed split settings are shown in Table 1. Following the pre-processing strategies in (Li et al. 2017; Liu et al. 2018), the sessions with the length of 2 and items with the appearing frequency of 5 are kept in the training and test set. Sessions in the test set are generated from the last week. We set the data from the last week for test and the remaining part for training. |
| Hardware Specification | No | The paper states, "Our implement is based on Tensorflow." but does not provide any specific hardware details such as CPU/GPU models, memory, or cloud instances used for the experiments. |
| Software Dependencies | No | The paper mentions "Our implement is based on Tensorflow." but does not provide specific version numbers for TensorFlow or any other software dependencies, which are necessary for reproducibility. |
| Experiment Setup | Yes | The embedding dimensionality d is set as 100. We assign the regularization penalty λ2 = 10 6. All models are optimized using the Adam optimizer with the batch size and learning rate as 512 and 1e 3, respectively. The training frequency f in each epoch is set as 1, 4, 6 corresponding to the Yoochoose, Diginetica, Retailrocket, respectively. Furthermore, the dropout technique is applied in the training phase to alleviate the overfitting issue, with the ratio of 0.2. |