![](bg3.png)
is shown as TV Guide + Classification in Figure 1.
These programs with predicted ratings are nothing but
recommendation list for each individual user. Finally,
social choice theory strategies are used to get a com-
bined (group) recommendation list.
Learning algorithm plays major role in content
based recommendation approach. It is used to learn
user profiles. Our learning algorithm (DLRL) as
mentioned in Figure 1 is a decision list rule learner
based on RIPPER (Cohen, 1995) and FOIL (Quin-
lan, 1996) rule learners. It is a multi-class rule learner
wherein there are five classes : bad, average, above
average, good, excellent. Initially, all training ex-
amples are divided into two sets: training data and
prune data. Training data is used to learn the set
of rules. Prune data is used to prune the rules to
avoid over-fitting. FOIL Information gain is given
as FOIL Gain(L, R) = t(log
2
(
p
1
p
1
+n
1
) − log
2
(
p
0
p
0
+n
0
))
where L is the candidate literal to add to rule R, p
0
is
the number of positive bindings of R, n
0
is the number
of negative bindings of R, p
1
is the number of positive
bindings of R + L, n
1
is the number of negative bind-
ings of R + L, t is the number of positive bindings of
R also covered by R + L. The formula used to prune
the rule is defined as v =
(p−n)
(p+n)
where p is the number
of positive examples covered by the rule in prune data
set and n is the number of negative examples covered
by the rule in the prune data set. Pruning criteria is
deleting the final sequence of conditions that maxi-
mizes v. The different steps involved in our learning
algorithm is shown in Algorithm 1.
Algorithm 1: Learning Algorithm used in VSW.
Input: Train Data,Prune Data
Output: set of rules
Step 1: foreach class do
Find the number of training examples for that class;
Take the class with maximum number of examples, make that as
Default class;
Step 2: Take an empty RuleSet;
while No class has left do
take the next smallest class;
Consider training examples for that class as positive,
remaining as negative;
while All positive examples covered do
Take empty Rule;
Add conjuncts to rule as soon as it improves FOIL
Information gain;
prune the rule by deleting any final sequence of
conditions;
Mark covered positive examples by this rule;
Add this rule to RuleSet;
Step 3:Add Default Rule to RuleSet;
Return RuleSet;
Table 1: Example to demonstrate Social choice Strategies.
Tv-Programs A B C D E F G H I J
John 10 4 3 6 10 9 6 8 10 8
Adam 1 9 8 9 7 9 6 9 3 8
Mary 10 5 2 7 9 8 5 6 7 6
In VSW, mention is made about four social
choice theory strategies as out- lined in (Masthoff,
2003). The main idea behind social choice the-
ory is (also called as group decision making) decid-
ing what is best for a group given the opinions of
individuals. The strategies used in the implemen-
tation are (1) Utilitarian Strategy (Hogg and
Jennings, 1999): In this strategy, instead of using
ranking information, utility values are used. This can
be done in multiple ways, additive or multiplicative.
For example, the utility values for the 10 programs in
Table 1. will be 21, 18, 13, 22, 26, 26, 17, 23, 20,
22 respectively (column-wise addition). The TV pro-
gram E and F are having highest utility values. So
either E or F is the recommended program. (2) Least
misery strategy (O’Connor et al., 2001): In
this strategy, the item with large minimum individ-
ual rating will be recommended. The idea behind this
strategy is that a group is as happy as its least happy
member. For Example, the group rating for the 10 TV
programs based on Least Misery Strategy will be 1, 4,
2, 6, 7, 8, 5, 6, 3, 6 respectively. From the above group
ratings, F has the highest rating. So TV program F
is recommended by Least misery strategy. (3) Most
pleasure strategy (Masthoff, 2004): Making
new list with the maximum of individual ratings. For
Example, from Table 1 the group rating for 10 TV
programs based on this strategy will be 10, 9, 8, 9,
10, 9, 6, 9, 10, 8 respectively. From the above group
ratings, A, E, I are having the highest rating values.
So either A, E or I will be the recommended TV
program. (4) Average without Misery strategy
(McCarthy and Anagnost, 1998): In this strat-
egy, a new list of ratings is made with the average of
the individual ratings, but without items that score be-
low a certain threshold for individuals. The item with
maximum value will be recommended. For example,
from Table 1. considering a threshold of 4 the average
values for 10 TV programs will be -, 18, -, 22, 26, 26,
17, 23, -, 22 respectively. The TV program E and F
are having highest utility values. So either E or F is
the recommended program.
(Padmanabhan et al., 2011) suggested that a single
strategy alone would not be sufficient to get the most
accurate result as far as group recommendation is con-
cerned. To address this problem a combined strategy
was put forward that considers three factors: (1) Least
group member happy (like least misery strategy) (2)
Most group member happy (like most pleasure strat-
ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence
372