Enhance Sentiment Analysis Using Transfer Learning
This example demonstrates how word embeddings trained on a word-prediction task can be used to train a better sentiment analysis network.
Load a pretrained word-embedding layer from the neural net repository.
Construct a classification net starting with the embedding layer.
Load sentiment-labeled movie reviews.
Train the net using the LearningRateMultipliers option to freeze the embedding layer.
Try the network on a few examples.
Compute the accuracy of the network. Thanks to the pre-trained embedding layer, a decent sentiment-analysis net has been trained in about one minute and only using a small amount of data.