Exploiting Behavioral Consistence for Universal User Representation
Authors: Jie Gu, Feng Wang, Qinghui Sun, Zhiquan Ye, Xiaoxiao Xu, Jingmin Chen, Jun Zhang4063-4071
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on benchmark datasets show that our approach can outperform state-of-the-art unsupervised representation methods, and even compete with supervised ones. |
| Researcher Affiliation | Industry | Jie Gu*, Feng Wang , Qinghui Sun, Zhiquan Ye, Xiaoxiao Xu, Jingmin Chen, Jun Zhang Alibaba Group, Hangzhou, China {yemu.gj,wf135777,yuyang.sqh,beichen.yzq,xiaoxiao.xuxx,jingmin.cjm,zj157077}@alibaba-inc.com |
| Pseudocode | No | The paper does not include a clearly labeled pseudocode block or algorithm block. |
| Open Source Code | Yes | Source codes of SUMN will be released at https://github.com/m2408gj/SUMN. |
| Open Datasets | Yes | Amazon Dataset1 This dataset includes product reviews... 1https://nijianmo.github.io/amazon/index.html; Twitter Dataset We download the twitter archives... 2https://archive.org/ |
| Dataset Splits | No | The paper states "The training is stopped when the loss converges on the validation set." for SUMN training and "For all evaluation datasets, we randomly select 80% of the samples for training downstream models and the rest for performance test." for downstream tasks. While validation sets are used, explicit split percentages or specific sample counts for the validation portion are not provided. |
| Hardware Specification | Yes | The models are trained on one V100 GPU and can achieve reasonable performances quickly. |
| Software Dependencies | No | The paper mentions optimizers (Adam) and models (BERT, Text CNN, HAN) by name but does not specify software dependencies (e.g., libraries, frameworks) with version numbers. |
| Experiment Setup | Yes | For all datasets, the dimension of all embeddings in SUMN, namely d, is set to be 256, and the number of hops is set to be 5. The loss function (Equation 5) is optimized by the Adam optimizer (Kingma and Ba 2014) with a learning rate of 0.001 and a batch size of 256. |