Information Technology Reference
In-Depth Information
Encyclopedic Data
Genre
Musical Career
Artist and Band
Genre Relation
liked/rate
Album
Album Release
Album Tracks
Track
User
e.g. Facebook or LastFM
Fig. 7.14
The semantic dataset with music information and user profiles linked to it
Table 7.1
Music information contained in the Freebase dataset
Entities
Number of entities
Number of edges
Musicians
Genre
Albums
Tracks
Musicians
417,217
-
79,543
374,445
-
Genre
3,082
79,543
-
90,444
-
Albums
438,180
37,445
90,444
-
1,048,565
Tracks
1,048,576
-
-
1,048,565
-
a graph containing the user profiles and the Freebase data interlinked. The linkage
is needed as our enrichment algorithm is a graph-based method. Without connected
data, the profile enrichment cannot be computed. Thus, it is necessary to know that an
entity such as 'Facebook#The_Beatles' in a user profile is similar to the entity 'Free-
base#Beatles' in the Freebase dataset and to create a link between them. Figure
7.15
shows the situation before and after the linkage. Linkage is done using a set of rules
that connect the profiles. First, we check if we have a MusicBrainz ID (which is the
case if we got the user data from LastFM). If we have the MusicBrainz ID the linkage
is easy as this information is also part of the meta-information that Freebase provides
about the artists. If no MusicBrainz ID is available we try to link entities based on the
artist name in different spellings and languages offered by Freebase. If more than one
Freebase node matches the rules and we cannot disambiguate the correct node this
entity is disregarded. While we assume that this method minimizes the number of
false positive linked entities, there still may be incorrectly linked entities that might
lead to a reduced recommendation quality.
Having connected the user profile with the Freebase dataset, the derived semantic
network can be used for enriching user profiles.
Search WWH ::
Custom Search