Page 238 - Big Data Analytics for Intelligent Healthcare Management

9.2 BACKGROUND 231

ways: building a recommendation system that combines both of the above two techniques, applying
some collaborative filtering in a content-based approach, and using some content-based filtering in
the collaborative approach. This technique uses different hybrid methods such as cascade hybrid,
weighted, mixed, and switching hybrid according to their operations.

9.2.3.2 Collaborative-based filtering recommendation system
Collaborative filtering predicts unknown outcomes by creating a user-item matrix consisting of users’
product preferences or interests. Similarities between users’ profile are measured by matching the user-
item matrix with users’ preferences and interests. The neighborhood is made of groups of users. The
user who has not rated specific items before, receives recommendations for a particular product by
considering positive ratings given by users in his neighborhood. The CF (Collaborative Filtering) rec-
ommendation system is used for two purposes. They are recommendation and prediction. Prediction
refers to a rating value R i,j of item j for user i. There are two types of techniques: memory-based and
model-based collaborative filtering. The following figure explains the process of collaborative filtering
technique [16, 17] (Fig. 9.3).

A. Memory-based collaborative filtering
Item and user are two key factors in this filtering technique. There are two types of memory-based
collaborative filtering: item-based and user-based collaborative filtering. Prediction is calculated by
measuring similarity among the items. This technique constructs a model on similarities of the item
by considering the active user-rated products from the user-item matrix, by which we can measure
the similarity among the target item and all retrieved items. Then we select k most similar items and
find the prediction by calculating a weighted average of the active user rating on similar items k.
We use different mathematical approaches to calculate user-user and item-item similarity. These
are correlation-based similarity measure, cosine-based similarity measures, and Pearson’s correlation
coefficient [16]. Pearson’s coefficient is described in Eq. (9.1).
n
X
ð
ð r a,i r a Þ r u,i r u Þ
i¼1
ð (9.1)
sa, uÞ ¼ s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n n
X X
2 2
ð r a,i r a Þ ð r u,i r u Þ
i¼1 i¼1
i1 i2 i3 i4 i5
u1 4 4 1
Prediction
u2 4 3 Output
u3 5 2 1 interface
Recommendation
u4 4 5
u5 5 4
u6 5 3
(User-item rating matrix) CF-algorithm
FIG. 9.3
Collaborative filtering technique.