text-classification-by-cnn

A convolutional neural network model for sentence classification by using tensorflow.

Requirement

• python3

• tensorflow 1.6.0

• numpy

• python-gflags (for processing the input arguments)

• gensim (for loading the pre-train word vectors)

Model structure

Start training

python train.py

Visualizing results in TensorBoard

tensorboard --logdir /"PATH_OF_CODE"/log/"TIMESTAMP"/summaries/

Descreption of files

• inputs (directory):

• rt.pos & rt.neg: The dataset contains 10,662 examples of movie review sentences. download from -> "Movie Review data from Rotten Tomatoes". One positive sentences file, one negative sentences file.

• vectors.bin: Pre-trained word vectors by google's word2vec, you can download the source-archive from this github project. You can also see the gensim (topic modelling for human) web page for getting more help of how to use this.

• data_helpers.py: It contains functions for the data loading (include pre-trained word vectors), data clean and generating batch data for training.

• text_cnn.py: The core function for generating a cnn for text classification. Model structure: embedding layer -> convolutional layer -> max-pooling layer -> softmax layer.

• train.py: It implements the reading parameters, data preperation and training procedure.

Refer to the papers

[1]. Kim, Y. (2014). Convolutional neural networks for sentence classification. Eprint Arxiv.

[2]. Zhang, Y., & Wallace, B. (2015). A sensitivity analysis of (and practitioners' guide to) convolutional neural networks for sentence classification. Computer Science.