Information Technology Reference
In-Depth Information
(1a)
(1b)
(2a)
(2b)
(3a)
(3b)
Fig. 7.20
Results for Evaluation on the Facebook (
first column
) and LastFM (
second column
)
datasets.
a
Shows the results for users with an unusual music taste,
b
shows the results for users
with popular music taste,
c
shows results over complete dataset
other users based on the enriched profile and hence items to recommend. This effect
becomes even more visible in scenarios where recommendations for users with an
uncommon taste are computed. In these scenarios the strategy
CF
enriched profiles
outperforms all other approaches. As CF depends on a sufficient amount of neighbors
to compute recommendations and finding similar users for users with an uncommon
taste is more difficult, enrichment helps to overcome this problem. The results for
the LastFM user profiles (Fig.
7.19
) confirm the findings on the Facebook dataset.
On the LastFM dataset, we used a more restrict threshold to distinguish between
users with common and uncommon taste. The evaluation results show that for users
with a uncommon taste
CF
+
+
Most Popular
recommendations perform bad while
the
CF
enriched profiles
recommender really improves recommendation quality.
For common taste users and all users, the
CF
+
Most Popular
recommender performs
best. Both combined strategies outperform the
standard CF
recommender.
+
Search WWH ::
Custom Search