Content discovery using perceptual automation
Alexander I. Iliev · 2018
In this work, an innovative media content discovery and selection system is proposed based on human behavior. There were two parts to this project: first, in the perception phase, a speech signal is used for analysis with the intention to detect and extract emotional cues; and second, in the automation phase, textual information was used to determine the sentiment using natural language processing methodology. The speech cues are first captured from the signal and then classified in three emotional categories: upbeat, positive or happy; emotionless, flat or neutral; and downbeat, negative or sad. Then in order to match the collected emotions from speech to specific contextual information captured from a real source, using natural language processing techniques five books from The Game of Thrones were processed. The results were summarized and mapping between the two systems was drawn so that books can be chosen based on the degree of each emotion portrayed from speakers. This can be done in a non-intrusive fission without direct and explicit human-machine intervention.