User interface design. An individual’s satisfaction with a group recommendation may be increased by good user interface design. For example, when showing an item, users could be shown what the next item will be (e.g. in a TV programme through a subtitle). This may inform users who do not like the current item that they will like the next one better.
• Group aggregation strategies for cold-start problems. In Section 21.7.2, we have sketched how group aggregation can be used to help solve the cold-start problem. However, our study in this area was very small, and a lot more work is required to validate and optimise this approach.
• Dealing with uncertainty. In this chapter, we have assumed that we have accurate profiles of individuals’ preferences. For example, in Table 21.1, the recommender knows that Peter’s rating of item B is 4. However, in reality we will
often have probabilistic data. For example, we may know with 80% certainty
that Peter’s rating is 4. Adaptations of the aggregation strategies may be needed
to deal with this. DeCampos et al try to deal with uncertainty by using Baysian
networks [4]. However, they have so far focussed on the Average and Plurality
Voting strategies, not yet tackling the avoidance of misery and fairness issues.
• Empirical Studies. More empirical evaluations are vital to bring this field for-
wards. It is a challenge to design well-controlled, large scale empirical studies
in a real-world setting, particularly when dealing with group recommendations
and affective state. All research so far (including my own) has either been on a
small scale, in a contrived setting or lacks control.