지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학술대회자료
- 저자정보
- 발행연도
- 2014.5
- 수록면
- 677 - 680 (4page)
이용수
초록· 키워드
오류제보하기
Recently the interests with culture technologies have been heightened and accordingly many studies for recommendation systems considering individuals’ preferences have been done. As a part of this trend, music recommendation systems have been developed in content-based and collaborative filtering methods. Previous studies have emphasized on technical aspects such as the extraction of music items’ features and the development of comparison algorithms for them. However, there are few studies to develop a music recommendation system based on individual users’ cognitive characteristics. To contribute to this issue, the present study proposes a music recommendation method considering serendipity based on individual users’ play records. Serendipity occurs when users meet familiar music items unexpectedly. The serendipity is determined by the comparison between users’ listening records in the past and the recent days. This is to find music items which users have often listen to in the past but do not anymore, and to recommend new music items similar to them. To apply the serendipity in the recommendation process, we utilize artificial neural network models. The artificial neural network models consist of long-term and short-term models, each of which is used to analyze user’s listening behaviors. Music items in a user’s playlist are used to train the models based on the last played times. The models use music items’ features as input values and play counts (the numbers of play times) as output values. Music items’ features are expressed by MFCC(Mel-frequency cepstral coefficients). After the artificial neural network models are trained, each model allocates the predicted play counts for music items in database. The music items’ serendipity degrees are calculated by difference between the predicted play counts in long-term model and short-term model. A high serendipity degree of item means users would satisfy the item with high probability. Finally, the music items in database are appeared in the recommendation list according to the serendipity degrees. The music recommendation method suggested in this study is a new approach based on individual users’ listening habits and preferences. It is expected that this method can improve users’ satisfaction with recommended items.
목차
ABSTRACT
1. Introduction
2. Backgrounds
3. Model development
4. Conclusion and discussion
References
참고문헌 (0)
참고문헌 신청이 논문의 저자 정보
최근 본 자료
댓글(0)
0
UCI(KEPA) : I410-ECN-0101-2016-530-002490691