Abstract:

The task of emotion recognition is to automatically identify human emotions using temporal data of various types. Audio, text and visual information are the three main sources of data that are used in emotion recognition. Compared to sentiment analysis, emotion recognition is considerably more complex and difficult to achieve high accuracy due to less available labelled data for training and subjectivity of human emotion. There is evidence that there could be a disparity of emotion recognition when applied to people of different gender or race. There are also concerns about where emotion recognition could be used and the effectiveness of the method involved. For example, if emotion recognition had been involved as part of the assessment for a job interview, it could potentially biased views about the candidate if the method involved is not accurate. However, applications in different domains, such as personalized education, can provide new dimensions for efficient delivery of the knowledge.

The aim and scope of this study is to propose a prototype audio recommendation application on a mobile platform that uses emotion recognition methods to detect user’s emotions and recommend audio files that will improve mood of the listener. Applications of the prototype can range from development of individual learning environment to use in therapies where audio is involved. Technologies used for development of the prototype include streaming audio server, neural network model to detect emotions and front-end interface for interaction with the user.

Discussion and analysis about maturity of automated sentiment analysis and emotion recognition include key considerations about possible applications of the technology and the prototype for learning environments where diversity of race, culture backgrounds, gender and age of learners are included.

Keywords:

Sentiment analysis, emotion recognition, learning interfaces.