master's thesis
Abstract
Music streaming services have become a key part of the music industry in recent years. Among other things, they allow users to create their own playlists and continue them with similar tracks after the last track ends, but these are not always relevant. We aimed to improve the relevance of recommended tracks in playlist continuation using tune-based recommendation.
Our solution is based on the similarity of note sequences of different tracks. We present the entire workflow of our solution, from candidate selection and music transcription to the measurement of the similarity between the tracks. Unfortunately, our solution did not achieve satisfactory results and also has a high time complexity, therefore we do not consider it suitable for playlist continuation in a real-world scenario. However, we believe that our method has potential in other areas of use, such as detecting plagiarism in music.
Keywords
playlist continuation;music information retrieval;recommendation system;computer science;master's thesis;
Data
Language: |
English |
Year of publishing: |
2023 |
Typology: |
2.09 - Master's Thesis |
Organization: |
UL FRI - Faculty of Computer and Information Science |
Publisher: |
[M. Kosi] |
UDC: |
004:78(043.2) |
COBISS: |
170117891
|
Views: |
76 |
Downloads: |
18 |
Average score: |
0 (0 votes) |
Metadata: |
|
Other data
Secondary language: |
Slovenian |
Secondary title: |
Nadaljevanje glasbenih seznamov predvajanja na podlagi melodij pesmi |
Secondary abstract: |
Storitve za pretočno predvajanje glasbe so v zadnjih letih postale ključen člen glasbene industrije. Te uporabnikom med drugim omogočajo, da si sestavijo svoje glasbene sezname, in jih, ko se konča zadnja skladba, samodejno nadaljujejo s podobnimi skladbami, ki pa včasih niso najustreznejše. Naš cilj je bil izboljšati ustreznost teh skladb s priporočanjem na podlagi melodij skladb.
Rešitev, ki smo jo zasnovali, temelji na iskanju podobnosti med zaporedji not različnih skladb. V magistrskem delu predstavimo celoten potek delovanja naše rešitve, od izbire nabora najbolj podobnih skladb in glasbene transkripcije do merjenja podobnosti med posameznimi skladbami. Naša rešitev žal ni dosegla zadovoljivih rezulatov in ima obenem še zelo visoko časovno zahtevnost, zato smo sklenili, da za nadaljevanje glasbenih seznamov v praksi ni primerna. Menimo pa, da ima potencial na drugih področjih, na primer iskanju plagiatov v glasbi. |
Secondary keywords: |
nadaljevanje glasbenih seznamov;pridobivanje informacij iz glasbe;priporočilni sistem;magisteriji;Glasba;Računalništvo;Univerzitetna in visokošolska dela; |
Type (COBISS): |
Master's thesis/paper |
Study programme: |
1000471 |
Embargo end date (OpenAIRE): |
1970-01-01 |
Thesis comment: |
Univ. v Ljubljani, Fak. za računalništvo in informatiko |
Pages: |
VIII, 51 str. |
ID: |
20010350 |