Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
You can not select more than 25 topics
Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
tal
2d544e4866
|
1 year ago | |
|---|---|---|
| data | 1 year ago | |
| src | 1 year ago | |
| LICENSE | 1 year ago | |
| README.md | 1 year ago | |
README.md
NeuCrowd
This repository is the implementation of 'NeuCrowd: Neural Sampling Network for Representation Learning with Crowdsourced Labels'.
Dependencies
- Python == 3.6
- tensorflow == 1.8
- scikit-learn (for matrics evaluation)
To develop locally, please follow the instruction below:
git clone git@github.com:tal-ai/NeuCrowd_KAIS2021.git
cd NeuCrowd_KAIS2021
Data Preparation
Please convert your data into three csv data files (train.csv, valid.csv, test.csv) of correct format, each of which contains a label column which stand for the category(0/1).
The rest columns are the feature columns.
Two example data sets used in the paper are stored in data folder.
Train & Test
You can train the model as follow:
cd src/
python trainRLL.py your_data_set_path
You can test the model as follow:
cd src/
python inference_neucrowd.py your_data_set_path
Contact
If you have any problem to the project, please feel free to report them as issues.
该算法提出一个基于众包标签的监督表示学习 (SRL) 统一框架 NeuCrowd。可以缓解由于数据隐私、预算限制、特定领域标注人员短缺等导致的众包标签的数量有限的问题。该框架 (1) 通过safety-aware 抽样和稳健的锚点生成，创建了大量高质量 n 元组训练样本；(2)学习一个抽样神经网络，自适应地为 SRL 网络选择有效样本。在酒店评论数据集上，accuracy 达到 87.1%；在 Pre-K Children Speech 数据集上accuracy 达到 86.7%。
CSV Python
