This demo estimates a sentiment (positive, negative, or neutral) for each sentence in the input text, and also for the input text as a whole.
The sentiment analysis relies on three resources:
Word embeddings calculated from a corpus of Finnish text.
Product reviews harvested from the Internet.
A word-based convolutional neural network with 100 kernels each of sizes 2, 3, 4 and 5 words. The neural network is trained to predict the rating associated with product reviews, and the prediction it gives to the input text is converted to a sentiment.