Parody Detection: An Annotation, Feature Construction, and Classification Approach to the Web of Parody

Abstract

In this chapter, we discuss the problem of how to discover when works in a social media site are related to one another by artistic appropriation, particularly parodies. The goal of this work is to discover concrete link information from texts expressing how this may entail derivative relationships between works, authors, and topics. In the domain of music video parodies, this has general applicability to titles, lyrics, musical style, and content features, but the emphasis in this work is on descriptive text, comments, and quantitative features of songs. We first derive a classification task for discovering the ``Web of Parody.'' Furthermore, we describe the problems of how to generate song/parody candidates, collect user annotations, and apply machine learning approaches comprising of feature analysis, construction, and selection for this classification task. Finally, we report results from applying this framework to data collected from YouTube and explore how the basic classification task relates to the general problem of reconstructing the web of parody and other networks of influence. This points toward further empirical study of how social media collections can statistically reflect derivative relationships and what can be understood about the propagation of concepts across texts that are deemed interrelated.