Chinese | English

FastTextClassification

Fast text classification for you, Start your NLP journey

开源实现 / 简单 / 全面 / 实践

**The function is free, the code is open source, everyone can use it with confidence, welcome to contribute! **

[💥Latest News](#Latest News)
[💫Development plan](#Development plan)
[⛏️Steps to use](#Steps to use)
[📄Run Example](#Run Example)
[📄Result Display](#Result Display)
[🛠️Common error report](#Common error report)
[💐References & Acknowledgments] (#References & Acknowledgments)
[🌟Sponsor us](#Sponsor us)
[🌈Starchart] (#Starchart)
[🏆Contributors] (#Contributors)

Latest News

2023/03/23: FastTextClassification V0.0.1 is officially open source, version features:
- Support bilingual text classification in Chinese and English
- Support multiple text classification models: traditional machine learning shallow model, deep learning model and transformers class model
- Support multi-label text classification
- Support multiple embedding methods: inner/outer/random

Steps for usage

Clone this project

git clone https://github.com/fast-llm/FastTextClassification.git

Dataset download and preprocessing

I will add another automation script when I have time, please handle it by yourself or refer to preprocessing.py for the time being.

It is best to put the data in the data directory, such as data/dbpedia, and then divide it into 3 subdirectories. Input stores the original data set (the data set you downloaded), and data stores the preprocessed formatted data set (text-label format), saved_dict stores the training results (model and log, etc.).

Run the example

The tested development environment is as follows, for reference only, almost all environments should be able to run.

python: 3.6/3.7
torch: 1.6.0
transformers: 4.18.0
torchtext: 0.7.0
scikit-learn: 0.24.2 -tensorboardX:2.6 -nltk:3.6.7
numpy: 1.18.5
pandas: 1.1.5

Select the module to run according to your own needs, see the next section for details.

python run.py

Run the example

Run the DNN/transformers class model for text classification

python run.py

Run the traditional shallow machine learning model for text classification

python run_shallow.py

Run the DNN/transformers class model to do multi-label text classification

python run_multi_label.py

The following table is the reference result of running the demo directly:

Operating environment: python3.6 + T4

demo	Dataset	Example Model	Acc	Time Consumption	Remarks
run.py	THUCNews/cn	TextCNN	89.94%	~2mins
run_multi_label.py	rcv1/en	bert	61.04%	~40mins	See running results for other indicators
run_shallow.py	THUCNews/cn	NB	89.44%	105.34 ms

Result display: to be continued

The author provides detailed experimental results from shallow to deep to multi-label for your reference. However, due to limited time and computing power, many experiments may not be optimal, please know! Therefore, everyone is very welcome to actively contribute, supplement related experiments, codes, new models, etc., and build FastTextClassification together.

For the time being, only part of the summary results are provided. I will add detailed experimental results and parameters when I have time. There are many, and it will take some time to sort them out.

Data	Model	tokenizer	Min_word_len	Min_df	ngram	binary	Use_idf	Test acc	备注
THUCNews	LR	lcut	1	2	(1,1)	False	True	90.61%	C=1.0, max_iter=1000 词表61549； train score: 94.22% valid score: 89.84% test score: 90.61% training time: 175070.97 ms
	MultinomialNB(alpha=0.3)	lcut	1	2	(1,1)	False	True	89.86%	词表61549； training time: 94.18ms
	ComplementNB(alpha=0.8)	lcut	1	2	(1,1)	False	True	89.88%	词表61549； training time: 98.31ms
	SVC(C=1.0)	lcut	1	2	(1,1)	False	True	81.49%	词表61549；维度200 training time: 7351155.59 ms train score: 85.95% valid score: 80.07% test score: 81.49%
	DT	lcut	1	2	(1,1)	False	True	71.19%	max_depth=None training time: 149216.53 ms train score: 99.97% valid score: 70.57% test score: 71.19%
	xgboost	lcut	1	2	(1,1)	False	True	90.08%	XGBClassifier(n_estimators=2000,eta=0.3,gamma=0.1,max_depth=6,subsample=1,colsample_bytree=0.8, nthread=10) training time: 1551260.28 ms train score: 99.00% valid score: 89.34% test score: 90.08%
	KNN	lcut	1	2	(1,1)	False	True	83.34%	k=5 training time: 22.14 ms train score: 89.57% valid score: 82.69% test score: 83.34%