Social Media Mining: An Introduction

P1: Trim: 6.125in×9.25in Top: 0.5in Gutter: 0.75in
CUUS2079-09 CUUS2079-Zafarani 978 1 107 01885 3 January 13, 2014 17:28

256 Recommendation in Social Media

Our goal in this section is to formalize how existing methods for recom-

mending to a single individual can be extended to a group of individuals.

Consider a group of individualsG={u 1 ,u 2 ,...,un}and a set of prod-

uctsI={i 1 ,i 2 ,...,im}. From the products inI, we aim to recommend

products to our group of individualsGsuch the recommendation satisfies

the group being recommended to as much as possible. One approach is to

first consider the ratings predicted for each individual in the group and then

devise methods that can aggregate ratings for the individuals in the group.

Products that have the highest aggregated ratings are selected for recom-

mendation. Next, we discuss these aggregation strategies for individuals in

the group.

Aggregation Strategies for a Group of Individuals

We discuss three major aggregation strategies for individuals in the group.

Each aggregation strategy considers an assumption based on which ratings

are aggregated. Letru,idenote the rating of useru∈Gfor itemi∈I.

DenoteRias the group-aggregated rating for itemi.

Maximizing Average Satisfaction. We assume that products that satisfy

each member of the group on average are the best to be recommended to the

group. Then,Rigroup rating based on the maximizing average satisfaction

strategy is given as

Ri=

1

n

∑

u∈G

ru,i. (9.33)

After we computeRifor all itemsi∈I, we recommend the items that

have the highestRi’s to members of the group.

Least Misery.This strategy combines ratings by taking the minimum of

them. In other words, we want to guarantee that no individuals is being

recommended an item that he or she strongly dislikes. In least misery, the

aggregated ratingRiof an item is given as

Ri=minu∈Gru,i. (9.34)

Similar to the previous strategy, we computeRifor all itemsi∈Iand

recommend the items with the highestRivalues. In other words, we prefer

recommending items to the group such that no member of the group strongly

dislikes them.