-
This keynote discusses why (and what kind of) personalization is needed to
produce persuasive arguments and the challenges of investigating this topic.
Personalization is when a system automatically adapts its actions to a user;
in this case, we are interested in all actions related to the use of persuasive
arguments. We will use the term `hearer' to indicate the user, who in this case
is the target of the persuasive argument.
The need for personalization in persuasive systems in general has been argued
before [
13
]. Examples of where personalization is needed speci c to the case of
persuasive arguments include the following.
Deciding on the type of argument. There is a growing body of research
showing that adapting the type of argument to the hearer matters [
11, 3, 21, 20, 4, 10
]:
which argumentation scheme [
27
] or Cialdini's principle [
2
] is used, and which
framing, with di erent types of hearers more susceptible to di erent kinds of
arguments.
Deciding on the content of the argument. Even when using the same type
of argument, the content of the argument needs adapting. First, the position of
the argument matters. When an argument's position is too far from the current
position of the hearer, it falls outside the hearer's so-called latitude of acceptance
[
25
]. In such a case, the argument will not move the hearer's position towards
the speaker's, but on the contrary, may make the hearer even more ingrained in
their original position. Nguyen et al. [
19
] investigated the hearer's position after
hearing arguments, based on argument strength, argument position and receiver
involvement.
Position is not just a point on a linear scale (as it was in the research by
Nguyen et al). Often, there are multiple ways in which a behaviour can be
improved. For example, there are many ways in which a person's diet can become
healthier, and it may be easier to convince some people to eat more vegetables
(and even a particular kind of vegetable, c.f. [
8
]) and other people to eat more
fruit. Such cases could be regarded as having multiple position scales, e.g., one
for eating vegetables and one for eating fruit.
Second, the hearer's involvement with the topic of the argument matters [
9
].
For example, when selling a house, one could highlight that the house has a
garage that can hold two cars, but if the prospective buyer does not drive, this
argument will not help to sell the house. Instead, one might want to highlight
that that space could be used as a kids' playroom (if the buyer has kids) or a
gym (if the buyer is into tness). Involvement is related to the hearer's values,
preferences, interests, and goals [
1, 8, 14
]. Similarly, research in the Superhub and
MyWay European projects on behaviour change for sustainable travel shows that
hearer attitudes matter [
23, 7
], with hearers from di erent attitudinal segments
being in uenced by di erent argument contents.
Deciding the source of the argument. The credibility of the (perceived) source
of the argument in uences the extent to which people feel the argument is
believable, trustworthy, non-biased, and detailed [
17
]. The credibility of a source
depends on the hearer (as exempli es also in politics). Additionally, in certain
cases it may be better for the system not to provide arguments, but instead
encourage the user to make their own arguments, in line with motivational
interviewing [
18
]. This opens a new line of research on how a system can support
users to produce self-persuasive arguments.
Deciding on the timing of the argument. The timing of arguments also
matters. For example, research has shown that people are more persuaded by an
argument when they were in a positive a ective state than a negative one [
12
].
The hearer's cognitive load is also likely to matter.
Deciding whether to use arguments at all. Many di erent persuasion
techniques exist; Michie et al. [
16
] listed 137 techniques based on di erent
theories of behaviour change. Using arguments (as in persuasive messages) is just
one such technique. Whether this technique is appropriate depends on the
behavioural determinants [
15, 5, 6
] a system is trying to tackle, which in its turn
depends on the person using the persuasive technology. Additionally, research
shows that the e ectiveness of techniques also depends on user characteristics
such as personality [
22
], and that techniques that may be e ective for some
people are counterproductive for others [
22
]. This is likely to also hold for the use
of argumentation.
Challenges. Challenges include (1) how to measure actual rather than
perceived persuasiveness of arguments, and sequences of arguments, under
controlled conditions, (2) how to investigate the impact of adaptation in persuasive
argumentation (keeping in mind also the principles of layered evaluation [
24
]),
and (3) how to enable a system to generate the variation of arguments that are
needed for personalization (building on the work in [
26
]).