Skip to content

Use several classical deep learning models to solve multi-label NLP classification problem

Notifications You must be signed in to change notification settings

frankhjh/Abnormality-Recognition-in-Medical-Image-Report

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Abnormality Recognition in Medical Image Report

Introduction

It is a NLP task in the medical field. Based on doctors' descriptions of CT scans, we need to build models to determine whether anomalies exist in specific areas of the patients' bodies.

Data

For the reason of privacy protection, the text description has already replaced by the number,that is, each unique word is translated into the unique number. the label is just the ID of anomalies areas,if no any anomalies areas, then the label is empty.

Modeling

For this task, I tried following classical deep learning models and the combination of them.

1.Text-CNN

2.Bi-LSTM

3.Attention

4.Attention+Text-CNN

5.Bi-LSTM+Text-CNN

6.Attention+Bi-LSTM+Text-CNN

Besides, I also compare the performance of each model on both the training set and validation set.Below shows the result.

Training Loss

Validation Loss

Try yourself

Feel free to use my model in ./model for your own project.

If you want to run my code to train and make prediction by yourself,you can simply run the following command:

python main.py --model_name ['cnn'/'rnn'/'attn'/'attn_cnn'/'lstm_cnn'/'attn_lstm_cnn'] --seed 1 --epochs 20 --device 'cpu'.

Of course, you can change the parameters as you want,for the --model_name argument,you should choose one you want from the above list. For example

--model_name 'cnn'

The final prediction result of test set will be stored in ./pred_out/submission.csv

Some References

Convolutional Neural Networks for Sentence Classification

LONG SHORT-TERM MEMORY BASED RECURRENT NEURAL NETWORK ARCHITECTURES FOR LARGE VOCABULARY SPEECH RECOGNITION

Attention is all you need

About

Use several classical deep learning models to solve multi-label NLP classification problem

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages