A web-based system for emotion vector extraction

The ability of assessing the affective information content is of increasing interest in applications of computer science, e.g. in human machine interfaces, recommender systems, social robots. In this project, the architecture of a semantic system of emotions is designed and implemented, to quantify the emotional content of short sentences by evaluating and aggregating the semantic proximity of each term in the sentence from the basic emotions defined in a psychological model of emotions (e.g. Ekman, Plutchick, Lovheim). Our model is parametric with respect to the semantic proximity measures, focusing on web-based proximity measures, where data needed to evaluate the proximity can be retrieved from search engines on the Web. To test the performances of the model, a software system has been developed to both collect the statistical data and perform the emotion analysis. The system automatizes the phases of sentence preprocessing, search engine query, results parsing, semantic proximity calculation and the final phase of ranking of emotions.