The emergence of music recommendation systems calls for the development of new data management technologies able to query vast music collections. In this paper, we define fuzzy song sets and an algebra to manipulate them. We present a music warehouse prototype able to perform efficient nearest neighbor searches in an arbitrary song similarity space. Using fuzzy song sets, the music warehouse offers a practical solution to the all musical data management scenarios provided: song comparisons, user musical preferences and user feedback. We investigate three practical approaches to tackle the storage issues of fuzzy song sets: tables, arrays and bitmaps. Finally, we confront theoretical estimates to concrete implementation results and prove that, from a storage perspective, arrays and bitmaps are both effective data structure solutions.
Music recommendation systems have recently gained a tremendous popularity. Music lovers discover new ways of searching and sharing their favorite music. However, at such growing speed, the database element of any recommendation systems will soon become a bottleneck. Hence, appropriate musical data management tools are needed. Music Warehouses (MWs) are dedicated data warehouses optimized for the storage and analysis of music content. They are currently developed to respond to this is call. The contributions of this paper are threefold. First, motivated by a case study , we propose three generic usage scenarios illustrating the current demands in musical data management. To answer these demands, we define fuzzy song sets and develop an algebra. Second, to demonstrate the usefulness of fuzzy song sets, a prototypical MW composed of two multidimensional cubes is presented. For each cube, concrete examples of queries inspired by the usage scenarios are provided. Fuzzy song sets prove to be an adequate data structure to manipulate musical information. Third, we discuss three solutions for storing fuzzy song sets and we construct theoretical estimates. A practical implementation shows that the structure overhead represents a major part of the storage consumption and that two solutions are viable for very large music collections. A lot of attention was drawn to enable music lovers to explore individual music collections . Within this context, several research projects have been conducted in order to pursue a suitable similarity measure for music . A music data model, an algebra and a query language are introduced by Wang et al. However, the model lacks an adequate framework to perform similarity searches. Jensen et al. address this issue and offer a model
that supports dimension hierarchies . This paper tackles the storage issues when the scalability does not remain limited to a few hundred thousands songs.
Publisher : OCG
By : Franc¸ois Deliege and Torben Bach Pedersen
File Information: English Language/ 6 Page / size: 198 KB
سال : 2007
ناشر : OCG
کاری از : Franc¸ois Deliege and Torben Bach Pedersen
اطلاعات فایل : زبان انگلیسی / 6 صفحه / حجم : KB 198
لینک دانلود : روی همین لینک کلیک کنید