利用Tensorflow实现基于CNN的中文短文本分类

Overview

Text Classification with CNN

使用卷积神经网络进行中文文本分类

CNN做句子分类的论文可以参看: Convolutional Neural Networks for Sentence Classification

还可以去读dennybritz大牛的博客:Implementing a CNN for Text Classification in TensorFlow

以及字符级CNN的论文:Character-level Convolutional Networks for Text Classification

本文是基于TensorFlow在中文数据集上的简化实现,使用了字符级CNN对中文文本进行分类,达到了较好的效果。

文中所使用的Conv1D与论文中有些不同,详细参考官方文档:tf.nn.conv1d

环境

  • Python 2/3
  • TensorFlow 1.3以上(我的是2.x)
  • numpy
  • scikit-learn
  • scipy

数据集

使用THUCNews数据集的一个子集进行训练与测试,数据集可在THUCTC:一个高效的中文文本分类工具包下载,请遵循数据提供方的开源协议。

本次训练使用了其中的10个分类,每个分类6500条数据。

类别如下:

体育, 财经, 房产, 家居, 教育, 科技, 时尚, 时政, 游戏, 娱乐

这个子集可以在此下载:链接: https://pan.baidu.com/s/1hugrfRu 密码: qfud

数据集划分如下:

  • 训练集: 5000 x 10
  • 验证集: 500 x 10
  • 测试集: 1000 x 10

从原数据集生成子集的过程请参看helper下的两个脚本。其中,copy_data.sh用于从每个分类拷贝6500个文件,cnews_group.py用于将多个文件整合到一个文件中。执行该文件后,得到三个数据文件:

  • cnews.train.txt: 训练集(50000条)
  • cnews.val.txt: 验证集(5000条)
  • cnews.test.txt: 测试集(10000条)

预处理

data/cnews_loader.py为数据的预处理文件。

  • read_file(): 读取文件数据。
  • build_vocab(): 构建词汇表,使用字符级的表示,这一函数会将词汇表存储下来,避免每一次重复处理。
  • read_vocab(): 读取上一步存储的词汇表,转换为{词:id}表示。
  • read_category(): 将分类目录固定,转换为{类别: id}表示。
  • to_words(): 将一条由id表示的数据重新转换为文字。
  • process_file(): 将数据集从文字转换为固定长度的id序列表示。
  • batch_iter(): 为神经网络的训练准备经过shuffle的批次的数据。

经过数据预处理,数据的格式如下:

Data Shape Data Shape
x_train [50000, 600] y_train [50000, 10]
x_val [5000, 600] y_val [5000, 10]
x_test [10000, 600] y_test [10000, 10]

CNN卷积神经网络

配置项

CNN可配置的参数如下所示,在cnn_model.py中。

class TCNNConfig(object):
    """CNN配置参数"""

    embedding_dim = 64      # 词向量维度
    seq_length = 600        # 序列长度
    num_classes = 10        # 类别数
    num_filters = 128       # 卷积核数目
    kernel_size = 5         # 卷积核尺寸
    vocab_size = 5000       # 词汇表达小

    hidden_dim = 128        # 全连接层神经元数目

    dropout_keep_prob = 0.5 # dropout正则化保留比例
    learning_rate = 1e-3    # 学习率

    batch_size = 64         # 每批训练大小
    num_epochs = 10         # 总迭代轮次

    print_per_batch = 100   # 每多少轮输出一次结果
    save_per_batch = 10     # 每多少轮存入tensorboard

CNN模型

具体参看cnn_model.py的实现。

大致结构如下:

image-20211110151539493

训练与验证

用cmd命令在代码文件所在目录运行 python run_cnn.py train,可以开始训练。

若之前进行过训练,请把tensorboard/textcnn删除,避免TensorBoard多次训练结果重叠。

Configuring CNN model...
Configuring TensorBoard and Saver...
Loading training and validation data...
Time usage: 0:00:14
Training and evaluating...
Epoch: 1
Iter:      0, Train Loss:    2.3, Train Acc:  10.94%, Val Loss:    2.3, Val Acc:   8.92%, Time: 0:00:01 *
Iter:    100, Train Loss:   0.88, Train Acc:  73.44%, Val Loss:    1.2, Val Acc:  68.46%, Time: 0:00:04 *
Iter:    200, Train Loss:   0.38, Train Acc:  92.19%, Val Loss:   0.75, Val Acc:  77.32%, Time: 0:00:07 *
Iter:    300, Train Loss:   0.22, Train Acc:  92.19%, Val Loss:   0.46, Val Acc:  87.08%, Time: 0:00:09 *
Iter:    400, Train Loss:   0.24, Train Acc:  90.62%, Val Loss:    0.4, Val Acc:  88.62%, Time: 0:00:12 *
Iter:    500, Train Loss:   0.16, Train Acc:  96.88%, Val Loss:   0.36, Val Acc:  90.38%, Time: 0:00:15 *
Iter:    600, Train Loss:  0.084, Train Acc:  96.88%, Val Loss:   0.35, Val Acc:  91.36%, Time: 0:00:17 *
Iter:    700, Train Loss:   0.21, Train Acc:  93.75%, Val Loss:   0.26, Val Acc:  92.58%, Time: 0:00:20 *
Epoch: 2
Iter:    800, Train Loss:   0.07, Train Acc:  98.44%, Val Loss:   0.24, Val Acc:  94.12%, Time: 0:00:23 *
Iter:    900, Train Loss:  0.092, Train Acc:  96.88%, Val Loss:   0.27, Val Acc:  92.86%, Time: 0:00:25
Iter:   1000, Train Loss:   0.17, Train Acc:  95.31%, Val Loss:   0.28, Val Acc:  92.82%, Time: 0:00:28
Iter:   1100, Train Loss:    0.2, Train Acc:  93.75%, Val Loss:   0.23, Val Acc:  93.26%, Time: 0:00:31
Iter:   1200, Train Loss:  0.081, Train Acc:  98.44%, Val Loss:   0.25, Val Acc:  92.96%, Time: 0:00:33
Iter:   1300, Train Loss:  0.052, Train Acc: 100.00%, Val Loss:   0.24, Val Acc:  93.58%, Time: 0:00:36
Iter:   1400, Train Loss:    0.1, Train Acc:  95.31%, Val Loss:   0.22, Val Acc:  94.12%, Time: 0:00:39
Iter:   1500, Train Loss:   0.12, Train Acc:  98.44%, Val Loss:   0.23, Val Acc:  93.58%, Time: 0:00:41
Epoch: 3
Iter:   1600, Train Loss:    0.1, Train Acc:  96.88%, Val Loss:   0.26, Val Acc:  92.34%, Time: 0:00:44
Iter:   1700, Train Loss:  0.018, Train Acc: 100.00%, Val Loss:   0.22, Val Acc:  93.46%, Time: 0:00:47
Iter:   1800, Train Loss:  0.036, Train Acc: 100.00%, Val Loss:   0.28, Val Acc:  92.72%, Time: 0:00:50
No optimization for a long time, auto-stopping...

在验证集上的最佳效果为94.12%,且只经过了3轮迭代就已经停止。

准确率和误差如图所示:

accuracy_1

测试

用cmd命令在代码文件所在目录下运行 python run_cnn.py test 在测试集上进行测试。

Configuring CNN model...
Loading test data...
Testing...
Test Loss:   0.14, Test Acc:  96.04%
Precision, Recall and F1-Score...
             precision    recall  f1-score   support

         体育       0.99      0.99      0.99      1000
         财经       0.96      0.99      0.97      1000
         房产       1.00      1.00      1.00      1000
         家居       0.95      0.91      0.93      1000
         教育       0.95      0.89      0.92      1000
         科技       0.94      0.97      0.95      1000
         时尚       0.95      0.97      0.96      1000
         时政       0.94      0.94      0.94      1000
         游戏       0.97      0.96      0.97      1000
         娱乐       0.95      0.98      0.97      1000

avg / total       0.96      0.96      0.96     10000

Confusion Matrix...
[[991   0   0   0   2   1   0   4   1   1]
 [  0 992   0   0   2   1   0   5   0   0]
 [  0   1 996   0   1   1   0   0   0   1]
 [  0  14   0 912   7  15   9  29   3  11]
 [  2   9   0  12 892  22  18  21  10  14]
 [  0   0   0  10   1 968   4   3  12   2]
 [  1   0   0   9   4   4 971   0   2   9]
 [  1  16   0   4  18  12   1 941   1   6]
 [  2   4   1   5   4   5  10   1 962   6]
 [  1   0   1   6   4   3   5   0   1 979]]
Time usage: 0:00:05

在测试集上的准确率达到了96.04%,且各类的precision, recall和f1-score都超过了0.9。

损失函数变化如图所示:

loss

从混淆矩阵也可以看出分类效果非常优秀。

预测

为方便预测,predict.py 展示了一个简单demo的预测。

Owner
Jeremiah
如今的现在早已不是当初的未来、
Jeremiah
Detecting and Tracking Small and Dense Moving Objects in Satellite Videos: A Benchmark

This dataset is a large-scale dataset for moving object detection and tracking in satellite videos, which consists of 40 satellite videos captured by Jilin-1 satellite platforms.

Qingyong 87 Dec 22, 2022
Official repository with code and data accompanying the NAACL 2021 paper "Hurdles to Progress in Long-form Question Answering" (https://arxiv.org/abs/2103.06332).

Hurdles to Progress in Long-form Question Answering This repository contains the official scripts and datasets accompanying our NAACL 2021 paper, "Hur

Kalpesh Krishna 41 Nov 08, 2022
RMNA: A Neighbor Aggregation-Based Knowledge Graph Representation Learning Model Using Rule Mining

RMNA: A Neighbor Aggregation-Based Knowledge Graph Representation Learning Model Using Rule Mining Our code is based on Learning Attention-based Embed

宋朝都 4 Aug 07, 2022
FaceVerse: a Fine-grained and Detail-controllable 3D Face Morphable Model from a Hybrid Dataset (CVPR2022)

FaceVerse FaceVerse: a Fine-grained and Detail-controllable 3D Face Morphable Model from a Hybrid Dataset Lizhen Wang, Zhiyuan Chen, Tao Yu, Chenguang

Lizhen Wang 219 Dec 28, 2022
Code for "Neural Body: Implicit Neural Representations with Structured Latent Codes for Novel View Synthesis of Dynamic Humans" CVPR 2021 best paper candidate

News 05/17/2021 To make the comparison on ZJU-MoCap easier, we save quantitative and qualitative results of other methods at here, including Neural Vo

ZJU3DV 748 Jan 07, 2023
An Approach to Explore Logistic Regression Models

User-centered Regression An Approach to Explore Logistic Regression Models This tool applies the potential of Attribute-RadViz in identifying correlat

0 Nov 12, 2021
Ivy is a templated deep learning framework which maximizes the portability of deep learning codebases.

Ivy is a templated deep learning framework which maximizes the portability of deep learning codebases. Ivy wraps the functional APIs of existing frameworks. Framework-agnostic functions, libraries an

Ivy 8.2k Jan 02, 2023
A PyTorch library for Vision Transformers

VFormer A PyTorch library for Vision Transformers Getting Started Read the contributing guidelines in CONTRIBUTING.rst to learn how to start contribut

Society for Artificial Intelligence and Deep Learning 142 Nov 28, 2022
PointCNN: Convolution On X-Transformed Points (NeurIPS 2018)

PointCNN: Convolution On X-Transformed Points Created by Yangyan Li, Rui Bu, Mingchao Sun, Wei Wu, Xinhan Di, and Baoquan Chen. Introduction PointCNN

Yangyan Li 1.3k Dec 21, 2022
ML-Decoder: Scalable and Versatile Classification Head

ML-Decoder: Scalable and Versatile Classification Head Paper Official PyTorch Implementation Tal Ridnik, Gilad Sharir, Avi Ben-Cohen, Emanuel Ben-Baru

189 Jan 04, 2023
PyTorch implementation of Tacotron speech synthesis model.

tacotron_pytorch PyTorch implementation of Tacotron speech synthesis model. Inspired from keithito/tacotron. Currently not as much good speech quality

Ryuichi Yamamoto 279 Dec 09, 2022
Code for Talking Face Generation by Adversarially Disentangled Audio-Visual Representation (AAAI 2019)

Talking Face Generation by Adversarially Disentangled Audio-Visual Representation (AAAI 2019) We propose Disentangled Audio-Visual System (DAVS) to ad

Hang_Zhou 750 Dec 23, 2022
NU-Wave: A Diffusion Probabilistic Model for Neural Audio Upsampling

NU-Wave: A Diffusion Probabilistic Model for Neural Audio Upsampling For Official repo of NU-Wave: A Diffusion Probabilistic Model for Neural Audio Up

Rishikesh (ऋषिकेश) 38 Oct 11, 2022
Code for ACM MM 2020 paper "NOH-NMS: Improving Pedestrian Detection by Nearby Objects Hallucination"

NOH-NMS: Improving Pedestrian Detection by Nearby Objects Hallucination The offical implementation for the "NOH-NMS: Improving Pedestrian Detection by

Tencent YouTu Research 64 Nov 11, 2022
Canonical Capsules: Unsupervised Capsules in Canonical Pose (NeurIPS 2021)

Canonical Capsules: Unsupervised Capsules in Canonical Pose (NeurIPS 2021) Introduction This is the official repository for the PyTorch implementation

165 Dec 07, 2022
Package to compute Mauve, a similarity score between neural text and human text. Install with `pip install mauve-text`.

MAUVE MAUVE is a library built on PyTorch and HuggingFace Transformers to measure the gap between neural text and human text with the eponymous MAUVE

Krishna Pillutla 182 Jan 02, 2023
Pytorch implementation code for [Neural Architecture Search for Spiking Neural Networks]

Neural Architecture Search for Spiking Neural Networks Pytorch implementation code for [Neural Architecture Search for Spiking Neural Networks] (https

Intelligent Computing Lab at Yale University 28 Nov 18, 2022
Python Implementation of the CoronaWarnApp (CWA) Event Registration

Python implementation of the Corona-Warn-App (CWA) Event Registration This is an implementation of the Protocol used to generate event and location QR

MaZderMind 17 Oct 05, 2022
ArcaneGAN by Alex Spirin

ArcaneGAN by Alex Spirin

Alex 617 Dec 28, 2022
A PyTorch implementation of Sharpness-Aware Minimization for Efficiently Improving Generalization

sam.pytorch A PyTorch implementation of Sharpness-Aware Minimization for Efficiently Improving Generalization ( Foret+2020) Paper, Official implementa

Ryuichiro Hataya 102 Dec 28, 2022