News Recommendation in Real-Time - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 166

items. The algorithm can be altered to consider different timeframes by restricting
the interactions which it receives (see Sect. 6.4.2 ).

Algorithm 1 Most Popular Recommender
Input matrix of interactions R , set of items I , number of items to recommend k

Output list of k items sorted by popularity

1: for all i ∈ I do

) ₐ u ∈ U I ( 2: popularity ( i i , u)

3: end for
4: recommendations ₐ top (k, popularity , I )

6.5.1.2 Most Recent

The most recent recommender builds upon the notion of recency. As Algorithm 2
illustrates, most recent recommendation ranks items according to their appearance in
the collections. The algorithm takes the set of items, their creation time, the current
time, and a specification of how many items to recommend as input. Subsequently,
we obtain an items age subtracting the date of creation from the current time. The
method determines which items to recommend by cutting the list of items ordered
by their ages at position

. As new items enter the collection, they move on top of
the list replacing the former top-ranked ones. Thus, the method keeps the items to
recommend up to date (see Sect. 6.4.3 ).

k

Algorithm 2 Most Recent Recommender
Input set of items I , timestamps of item creation ˄( i ) , current time T , number of items to recom-
mend

k
Output

k

items sorted by date of creation
1: for all i ∈ I do
2:

list of

t ( i ) ₐ T − ˄( i )

3: end for
4: recs ₐ top (k, − t , I)

6.5.1.3 Random

Recommending random items represents another simple method. Randomly picking
items yields the risk of suggesting irrelevant items. On the other hand, it could
provide access to items which are neither popular nor recent and thus would not have
been found by users. Algorithm 3 depicts the random recommendation procedure.
It randomly adds items to the list of recommendations until the list has the desired
capacity. Items may not be redundant.