The importance of quickly finding relevant information in an ever-growing ocean of online content, raises the requirements on search engine ranking quality. Tags can help sort online content and facilitate ranking, but manually tagging massive amounts of content is futile. Deep learning methods have become increasingly good at performing complex classification tasks, and could potentially relieve the problem by automating the tagging process. The aim of this project was to evaluate the effect that automatically generated music tags have on the ranking quality of musical content returned by a search engine. The state-of-the-art music tagging library, musicnn, was used to automatically tag the music collection at the company Epidemic Sound. The ranking quality for a large set of user requests was measured using the Mean Average Precision metric before and after the tags were added. Whether the tags had a positive or negative effect on the ranking quality was deemed inconclusive due to a sparsity in the ground truth data.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:liu-170252 |
| Date | January 2020 |
| Creators | Weberyd, Emma |
| Publisher | Linköpings universitet, Medie- och Informationsteknik, Linköpings universitet, Tekniska högskolan |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0017 seconds