Hierarchical Neural Network Implementation: Emotion Recognition for Food Security Comments on Twitter

Modern Hierarchical Neural Network (HNN) implementation combines several deep learning algorithms working together, connected in a hierarchy layer. For this HNN architecture to work well, the problem and the data must be in a hierarchical format. Emotion recognition is the best example of a layered problem where each emotion is attached to a sentiment. This research proposes an HNN model to solve the emotion recognition problem with three deep learning, one for the sentiment in the first layer and two models for the emotion prediction in the second layer. There are two combinations to be compared, full-LSTM and full-CNN. Surprisingly, the overall HNN performance for both combinations is similar, and both are below a control model without HNN architecture. However, solving the emotion recognition problems in the food security domain was still possible despite poor performance. The application result creates a rough estimation of what people feel about the current food security trend.