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.
czp
da53ad0154
|
2 years ago | |
|---|---|---|
| src | 2 years ago | |
| README.md | 2 years ago | |
| create_dataset.py | 3 years ago | |
README.md
AttentionLSTM
目录
AttentionLSTM概述
给定一个句子和句子中出现的某个aspect，aspect-level 情感分析的目标是分析出这个句子在给定aspect上的情感倾向。
例如：great food but the service was dreadful! 在aspect “food”上，情感倾向为正，在aspect “service”上情感倾向为负。Aspect level的情感分析相对于document level来说粒度更细。
论文： Attention-based LSTM for Aspect-level Sentiment Classification [https://www.aclweb.org/anthology/D16-1058.pdf]
模型架构
AttentionLSTM模型的输入由词向量和Aspect向量组成，LSTM是单向单层的结构，输出的状态向量再和Aspect向量concatenate，并输入到Attention结构中计算Attention权值。最后拿Attention权值和LSTM输出的状态向量一起计算情感极性。
数据集
- semEval 2014 Task 4
- 预训练词向量：Glove
环境要求
- 硬件：Ascend处理器
- 框架：mindspore
快速入门
在Ascend处理器上运行
cd src
python train_atae_lstm.py
脚本说明
.
├── AttentionLSTM
├── README.md # AttnetionLSTM相关说明
├── script
│ ├── run_eval.sh #
│ └── run_train.sh #
├── src
│ ├── config.py # 参数配置
│ ├── eval_atae_lstm.py # 推理脚本
│ ├── load_dataset.py # 加载数据集
│ ├── model.py # 模型结构
│ ├── model_for_test.py # 模型推理
│ ├── model_for_train.py # 模型训练
│ ├── my_utils.py # LSTM配置
│ ├── rnn_cells.py # LSTMCell
│ ├── train_atae_lstm.py # 训练脚本
│ └── rnns.py # LSTM
├── eval.py # GPU、CPU和Ascend的评估脚本
└── train.py # GPU、CPU和Ascend的训练脚本
No Description
Python