Muli-class text classification using neural network (Reuters)

Project description

Input: 8982 news that belong to 46 mutually exclusive categories. The first 1000 news are kept as the development set while the rest is used for training set. There is an additional test set of 2246 news.

Goal: Classifying a given news as one of the 46 mutually exclusive categories.

Network model: It is shown in the following figure:

The resulting loss and accuracy are shown in the following figures.

Modification:

To combat the overfitting problem, early stoppage is used and the training is done only for 9 epochs. The resulting accuracy is 0.8.

The source code of this project is available in my Github page.