Vis enkel innførsel

dc.contributor.authorJohansen, Einar Søreide
dc.date.accessioned2018-10-22T16:00:11Z
dc.date.available2018-10-22T16:00:11Z
dc.date.issued2018-06-15
dc.date.submitted2018-06-14T22:00:11Z
dc.identifier.urihttp://hdl.handle.net/1956/18659
dc.description.abstractThis thesis explores whether recommender systems can be used to create personalized content. To this end a prototype was created that generates music based on a user’s preferences. The prototype, named RecOrder, is therefore a song composer. The song creation process of RecOrder combines small audio clips into a song. 37 audio clips were created specifically for the prototype to use. These clips are short recordings of a singular instrument and are designed to be modular and fit together in any order. The selecting of what audio clips to combine, is based on the preference of the users interacting with the prototype. This is done by the implemented recommendation system, which is a weighted hybrid system. Hybrid recommendation systems consist of more than one recommendation algorithm. For the prototype, a collaborative item-based recommendation algorithm called Slope One was selected. Additionally, a custom knowledge-based algorithm was created as the secondary recommendation algorithm. Once the song has been created, the prototype moves on to a feedback phase. With the feedback phase, the user rates different sections of the generated song. This feedback, which is given on a scale from one to five, will help guide the recommendation system to more accurate recommendations. An empirical evaluation was conducted, which aimed to establish whether the generated songs were successfully tailored to the tests subjects’ personal preferences. 15 test subjects partook in the experiment, selected without any preference to musical background. In the experiment, the test subjects used the prototype to create multiple songs. They also answered questions in a survey, regarding the prototype and the song creation process. The results from the empirical evaluation show a positive trend in terms of user satisfaction. As the prototype creates more songs, the user felt that the songs were becoming gradually more personalized. Using recommendation systems for personalized content creation is also possible. However, there are some limitations that recommendation systems pose towards what domains are suitable. The limitations are related to the cold-start problem. If the recommendation system cannot learn the preferences of the user, it will not be able to yield recommendations. Similarly, the number of available items is also a critical factor for a suitable domain. Too few available items, and the domain is not fitting for a recommendation system approach.en_US
dc.language.isoengeng
dc.publisherThe University of Bergeneng
dc.titlePersonalized Content Creation using Recommendation Systemseng
dc.typeMaster thesisen_US
dc.date.updated2018-06-14T22:00:11Z
dc.rights.holderCopyright the author. All rights reserved.en_US
dc.description.degreeMasteroppgave i informasjonsvitenskap
dc.description.localcodeINFO390
dc.subject.nus735115eng
fs.subjectcodeINFO390
fs.unitcode15-17-0


Tilhørende fil(er)

Thumbnail

Denne innførselen finnes i følgende samling(er)

Vis enkel innførsel