Bacterial Evolutionary Algorithm-based Feature Selection for Word Sentiment Interpolation in Hungarian Language

With the advancement of social artificial agents the need for correct understanding of sentiment is growing. In this paper we propose a method for building a context-less word-level emotional model of words in the Hungarian language based on Russell's Circumpex model of affect. By utilizing Bacterial Evo-lutionary Algorithm for feature selection, a method for efficient web-based annotation is proposed. Using the latent information of word embeddings multi-layer perceptron networks are trained to realize an interpolative function of two-dimensional emotion vectors over the embedding space. Dimensionality reduction via correlation analysis is also discussed.