The goal of this project is to analys your sentiment through the review typed on the page. If the background turns into green, that means you have the positive review. Red is the opposite. This project acheived 91% accuracy.

Demo server is based on Flask. To run the server, execute demo_server.py

python demo_server.py

Next, once you go to the root page of your server, you will get the simple page. You can just type the review of what you watched recently for the test.

This open source is based on Python 3.5

pip install -r requirement.txt

IMDb Dataset: http://ai.stanford.edu/~amaas/data/sentiment/

GloVe pre-trained vector: https://nlp.stanford.edu/projects/glove/

if __name__ == "__main__":
	manager = Manager(ngram_range=2)
    manager.load_dataset()
    model = CNNModel().build(embedding_matrix=manager.embedding_matrix,
                             max_features=manager.max_features).model
    manager.train(model)
    manager.store_model("cnn_and_bi-gram")

To train a model, go to manager.py, and then we need to set up some parameters. An n-gram is a contiguous sequence of n items from a given sequence of text. You can set the range through ngram_range From my experience, I recommend you to use bi-gram, ngram_range=2. embedding_matrix means pre-trained matrix. Some model doesn't use this parameter. In that case, you need to remove that parameter. You don't need to consider max_features Lastly, your trained-model will be stored by the parameter of store.model method. In this example, cnn_and_bi-gram will be your filename.

if __name__ == '__main__':
    movie_review_classifier = Classifier(filename="cnn_and_bi-gram_90.484acc_model")
    movie_review_classifier.build()

    app.run(
        host="0.0.0.0",
        port=int("8888")
    )

Once you trained, you would get your own model. In this example, I will use the cnn model that has 90.484 accuracy. Go to demo_server.py and set up your model. Make sure you should take off the file extension. That's all. Run it!!:)

Coming soon.

Welcome!

• James Pak

This project is licensed under Gridspace.

• Word embedding
• DNNs