Automatic Annotation of Corpora For Emotion Recognition Through Facial Expressions Analysis

The massive adoption of social networks has madeavailable an unprecedented amount of user-generated content,which may be analyzed in order to determine people’s opinionsand emotions on a large variety of topics. Research has mademany efforts in defining accurate algorithms for the analysis ofemotions conveyed by texts, however their performance oftenrelies on the existence of large annotated datasets, whose currentscarcity represents a major issue. The manual creation of suchdatasets represents a costly and time-consuming activity andhence there is an increasing demand for techniques for theautomatic annotation of corpora. In this work we present amethodology for the automatic annotation of video subtitleson the basis of the analysis of facial expressions of peoplein videos, with the goal of creating annotated corpora thatmay be used to train emotion recognition algorithms. Facialexpressions are analyzed through machine learning algorithms,on the basis of a set of manually-engineered facial featuresthat are extracted from video frames. The soundness of theproposed methodology has been evaluated through an extensiveexperimentation aimed at determining the performance on realdatasets of each methodological step.