## Results
### 词向量+LSTM
| 词向量+LSTM | Acc | Pr | Re | F1 | Trainable | epoch |
| -------------- | -------- | -------- | -------- | -------- | -------- | -------- |
| 自训练word2vec(100d) | 0.625731 | 0.522778 | 0.226636 | 0.292154 | False | 50 |
| 自训练word2vec(100d) | 0.606561 | 0.465614 | 0.492967 | 0.460935 | True | 50 |
| 预训练glove(100d) | 0.624939 | 0.501283 | 0.431366 | 0.440754 | False | 50 |
| 预训练glove(100d) | 0.605527 | 0.471022 | 0.476533 | 0.456550 | True | 50 |
| 预训练wiki word2vec(100d) | 0.630228 | 0.506221 | 0.436347 | 0.449099 | False | 50 |
| 预训练wiki word2vec(100d) | 0.620865 | 0.479311 | 0.505142 | 0.475388 | True | 50 |
| fasttext(100d) | 0.655658 | 0.578113 | 0.264163 | 0.350707 | False | 50 |
| fasttext(100d) | 0.607963 | 0.469200 | 0.487640 | 0.457745 | True | 50 |
| Bert(768d) | 0.638890 | 0.526669 | 0.435295 | 0.456044 | False | 50 |
### 词向量+BiLSTM+Attention
| BiLSTM+Attention | Acc | Pr | Re | F1 | Trainable | epoch |
| -------------- | -------- | -------- | -------- | -------- | -------- | -------- |
| 自训练word2vec(100d) | 0.594484 | 0.411021 | 0.194717 | 0.252234 | False | 50 |
| 自训练word2vec(100d) | 0.630865 | 0.511518 | 0.287036 | 0.355171 | True | 50 |
| 预训练glove(100d) | 0.617624 | 0.508118 | 0.216726 | 0.288202 | False | 50 |
| 预训练glove(100d) | 0.625258 | 0.511204 | 0.261211 | 0.331538 | True | 50 |
| 预训练wiki word2vec(100d) | 0.622811 | 0.479011 | 0.241808 | 0.306475 | False | 50 |
| 预训练wiki word2vec(100d) | 0.609992 | 0.479359 | 0.476048 | 0.457980 | True | 50 |
| fasttext(100d) | 0.637248 | 0.566077| 0.182606 | 0.265081 | False | 50 |
| fasttext(100d) | 0.618331 | 0.489203 | 0.281450 | 0.340978 | True | 50 |
### 词向量+TextCNN(待更新实验结果)
| Dataset | 词向量+TextCNN | Acc | Pr | Re | F1 | Trainable | epoch |
| ------ | -------------- | -------- | -------- | -------- | -------- | -------- | -------- |
| SDS | 自训练word2vec(100d) | 0.599138 | 0.460517 | 0.274916 | 0.329898 | False | 50 |
| SDS | 自训练word2vec(100d) | 0.629213 | 0.514082 | 0.291974 | 0.358257 | True | 50 |
| SDS | 自训练fasttext(100d) | 0.626891 | 0.534870 | 0.168099 | 0.245896 | False | 50 |
| SDS | 自训练fasttext(100d) | 0.621121 | 0.500141 | 0.272007| 0.336070 | True | 50 |
| SDS | 预训练glove(100d)(100d) | 0.624519 | 0.513552 | 0.261986 | 0.330023 | False | 50 |
| SDS | 预训练glove(100d)(100d) | 0.637153 | 0.533213 | 0.290446 | 0.362039 | True | 50 |
### Bert
| Dataset | 方式 | Acc | Pr | Re | F1 | Trainable | epoch |
| ------ | -------------- | -------- | -------- | -------- | -------- | -------- | -------- |
| SDS | 提取特征+Dense | 0.684022 | 0.630039 | 0.309357 | 0.397258 | False | 50 |
| SDS | 提取特征+TextCnn | 0.675988 | 0.600909 | 0.296789 | 0.378142 | False | 50 |
| SDS | 提取特征+BiLSTM | 0.671320 | 0.590214 | 0.300830 | 0.378835 | False | 50 |
| SDS | 提取特征+BiLSTM+Attention | 0.673399 | 0.595025 | 0.300186 | 0.380878 | False | 50 |
## Some Records
### 1. 关于词向量
本实验共使用了
* 基于SDS数据集训练的word2vec
* 预训练的glove_100d (这个我忘记从哪下载的了 hxd对8起)
* 预训练的wiki_word2vec_100d [下载地址](https://wikipedia2vec.github.io/wikipedia2vec/pretrained/)
* 基于SDS数据集训练的fasttext_100d
* 通过`Bert-as-service`获取的Bert_768d词嵌入 使用同上一级页面的[Bert句向量获取](../Bert#调用bert-as-service获取句向量)
通过`main.py`中的options进行选择。
### 2. 关于模型
本实验共使用了
* LSTM (代码里直接换成了BiLSTM，但实验结果还是用的LSTM的，懒的改了)
* BiLSTM+Attention
* TextCNN (只跑了一个Bert_768d的，其他的词向量的结果还没来得及跑)
基于语义理解的软件摘要生成技术.zip
版权申诉
97 浏览量
2023-05-18
23:32:06
上传
评论
收藏 2.37MB ZIP 举报
基于语义理解的软件摘要生成技术.zip （118个子文件） 
data.csv 394KB test.csv 394KB data_sds.csv 357KB result.csv 69KB settings.json 78B README.md 4KB README.md 1KB README.md 877B README.md 540B README.md 335B main.py 10KB main.py 8KB textrank.py 5KB eval.py 4KB utils.py 4KB utils.py 4KB utils.py 4KB utils.py 4KB Attention_define.py 3KB Attention_define.py 3KB eval.py 2KB textrank.py 1KB save_data.py 1KB word2vec_train.py 1KB preprocess.py 1KB main.py 968B preprocess.py 960B text_cnn.py 950B fasttext_train.py 383B word2vec_train.py 371B utils.cpython-37.pyc 3KB utils.cpython-37.pyc 3KB Attention_define.cpython-37.pyc 3KB eval.cpython-37.pyc 2KB textrank.cpython-37.pyc 1KB preprocess.cpython-37.pyc 1KB text_cnn.cpython-37.pyc 1018B
26.txt 8KB 22.txt 8KB 5.txt 8KB 3.txt 7KB goldset_ads.txt 7KB 33.txt 7KB 22.txt 7KB goldset_sds.txt 7KB goldset.txt 7KB goldset.txt 7KB 34.txt 6KB 12.txt 6KB 8.txt 6KB 24.txt 6KB 26.txt 6KB 17.txt 6KB 5.txt 6KB 3.txt 5KB 15.txt 5KB 25.txt 5KB 2.txt 5KB 33.txt 5KB 27.txt 5KB 32.txt 5KB 30.txt 5KB 12.txt 4KB 34.txt 4KB 29.txt 4KB 4.txt 4KB 8.txt 4KB 19.txt 4KB 24.txt 4KB 15.txt 4KB 7.txt 4KB 23.txt 4KB 16.txt 4KB 17.txt 4KB 11.txt 4KB 14.txt 4KB 10.txt 4KB 2.txt 4KB 25.txt 4KB 1.txt 3KB 32.txt 3KB 35.txt 3KB 31.txt 3KB 13.txt 3KB 30.txt 3KB 4.txt 3KB 23.txt 3KB 27.txt 3KB 19.txt 3KB 36.txt 3KB 7.txt 3KB 29.txt 3KB 11.txt 3KB 20.txt 3KB 10.txt 3KB 1.txt 3KB 6.txt 3KB 14.txt 3KB 16.txt 3KB 18.txt 3KB共 118 条
- 1
- 2
资源评论
