Information extraction from user-generated content in the classical music domain

Abstract

The applications of Information Extraction (IE) on User-generated Content (UGC) have widely benefited from the emergence of microblogging services in the last decade. In particular, Twitter has been at the center of attention of many studies because of its widespread use and easy accessibility. Among the several fields which have benefited from this source, in particular Named Entity Recognition (NER) has demonstrated how challenging can be obtaining useful information from the noisy space of tweets . From another perspective, recently in the field of Music Information Retrieval (MIR) researches have shown how NLP techniques such as IE and NER can be an important resource to improve accuracy and precision in tasks like Music Recommendation, Artist Similarity or Genre Classification . The objective of this thesis is to investigate methods to extract information from user-generated content in a specific channel related to Classical Music, BBC Radio 3 , through the use of NER techniques. We investigate how state-of-the-art methods in NER can be applied to detect entities in the music domain, and how contextual information can contribute with NER in this particular case.