Sentiment Analysis

Sentiment analysis on small datasets using fine-tuning technique on a neural network.

Since the number of data records in an small dataset is not enough for a relative accurate result, I used Keras IMDB large dataset to train a neural network, then save the weights of the layers in neural network. On the next step I used that weights to train another neural network this time for small dataset with one additional layer to fit better to the small dataset. We can obtain better results using this technique.

The procedure is specially useful for languages which not have enough data for a proper modeling. In this project I used Farsi (Persian) comments from "digikala.com" translated to English using "Google Translate". You can access datasets here: datasets