Hands-On Transfer Learning with Python

Creating document summaries with CNN model

A review has many sentences. Some of these sentences are neutral and some are redundant to determine the polarity of the overall document. Summarizing the review or highlighting the sentences in the review where the user actually expresses the opinion is very useful. In fact, it also gives an explanation for the prediction we make and thus makes the model interpretable.

As explained in the paper, the first step for text summarization is to create a saliency map for the document by assigning an importance score to each sentence. To generate the saliency map for a given document we can apply the following technique:

1. We first perform a forward pass through the network to generate a class prediction ...