BrambleXu / Nlp Beginner Guide Keras

In this repo, I will implement some NLP models for the nlp beginner learner. In each project folder contains a notebook floder to show the output in detail. I think this would help the beginner to understand what happens in the model.

I will list what resource used for each model implementation. All project based on Python3.6 and Keras2.1.6 with TensorFlow1.8 backend.

Deep Models for NLP beginners

You can find detail instruction in each project. Here I will list what you can learn in each project.

Cnn-text-classification

• A keras implementation of CNN-non-static for text classification
• train a word2vec model
• Load pre-trained word2vec weights
• Save and load a keras model
• Use tensorboard to visualize your neural networks

word_embedding

• Load pre-trained GloVe weights
• How Keras deal with OOV token

char-level-cnn

What you can learn in this implementation:

• Using Keras function to preprocess char level text, article, notebook
• Constructing the char-cnn-zhang model, article, notebook

sentiment-comparison

In this project, I use three embedding levels, word/character/subword, to represent the text. And test them with two model, CNN and LSTM.

According to the result, subword-level embedding is useful for the dataset with many unknown words. The CNN not only achieve the better performance, but also take less training time. So if you want to implement a simple and powerful sentiment classification model, I highly recommend to use the CNN model.

Read the complete report on my blog: The Best Embedding Method for Sentiment Classification

bilstm-crf

I create this project to learn the NER task and implement the BiLSTM-CRF model by using keras-contrib.

• notebook

Requirements

• Python 3.6
• Keras 2.1.6
• Tensorflow 1.8
• tensorboard 1.8
• gensim 3.4.0
• scikit-learn 0.19.1
• numpy 1.14.3
• h5py 2.8

I use conda to construct the environment, and I highly recommend you do it too. After clone this project, you can run the following command to construct the whole environment. Make sure you already install the conda tool.

conda env create -f py36.yml

After install the whole environment, you can use following command to switch to the py36 environment.

source activate py36

Bibtex

Please use the following bibtex, when you refer my implementations in your papers.

@misc{liang2018kerasnlpmodel,
    title  = {Keras Implementations for NLP Models},
    author = {Xu, Liang},
    url    = {https://github.com/BrambleXu/nlp-beginner-guide-keras},
    year   = {2018}
}