1. INTRODUCTION
Intelligent technology is ever more pervasive in modern
societies, requiring people to increasingly collaborate with
intelligent agents, such as avatars, robots, smart home systems,
and other intelligent user interfaces that tap into users' daily
habits and routines to provide suitable support, feedback,
and recommendations [
7
]. This development - agents
becoming a part of society and interacting with humans - has
a vital consequence: agents will not only need to be
intelligent, they also need to be social [
9
]. Agents must be capable
of establishing long-lasting personal, well-embedded, con
dential, and respectful relationships with people. In order
Corresponding author.
to establish such relationships, agents must meet two
crucial requirements: personalisation and context-sensitivity,
i.e. modeling the user and interpreting the user's situation
based on real world personal data.
2.
EXPERIENCE SHARING
This paper introduces a personalisation framework
facilitating the establishment of human-agent relationships through
context-sensitive experience sharing. The framework applies
two mechanisms that both aim to develop human-agent
relationships through experience sharing [
1, 4, 5
]:
Sharing past experiences. Triggered by a stimulus (e.g.
a song, a picture, a video, or a social network post), the
agent engages in conversation with the user about the
memories of past experiences evoked by the stimulus. Exactly
which stimulus is used, depends on the context of the
application (also see Section 3). The framework uses an
expressive domain-speci c knowledge representation, i.e. an
ontology, to engage in conversations about the stimuli and to
gradually construct a user model containing personal
information about the user. The agent uses its knowledge about
the user - as available in the ontology - to determine the
intimacy of the relationship. The agent reciprocates by
discussing things deemed appropriate for that level of intimacy.
Sharing new experiences. Based on the user's (social)
context and preferences, the agent invites the user to
engage in joint activities so as to share and create new
experiences together. The agent identi es the people present in
the room using face and speech recognition. It uses their
identities to access the associated user pro les, which are
created manually beforehand and are appended
automatically with the information acquired during interaction (e.g.
when sharing past experiences or when participating in joint
activities). The agent uses information about the number of
people present (e.g. solitary, dyad, or group activities) and
their preferences (e.g. likes to play games or not) to
personalise the proposed activities and the content of the
activities. The agent continues to personalise its behaviour using
a feedback loop (e.g. sensing the users' emotional responses).
This feedback loop enables the agent to personalise its
interaction to the needs and desires of the users, and their social
context.
3.
APPLICATION DOMAIN
The framework is currently used in the ReJAM project1,
which aims to develop Robots engaging elderly in Joint
This song brings back so
many memories!
Could you tell me
about them?
(a)
Are you up
for
something
fun?
Depends…
what do
you
propose?
Well, your friend
Sarah is coming over.
So I would suggest to
either play a game or
make music together.
You both enjoy those
things.
involving
has
evokes
at
Time/age
triggers
Person
has
Emotion
triggers
likes
evokes
Music
from
by belongs to
Activities with Music. The objective of ReJAM is the
promotion of physical, cognitive, emotional, and social
wellbeing [
6
]. The main target group of ReJAM are elderly
with dementia (also see Figure 1(a) and (b)).
Sharing past experiences in ReJAM: Research shows
that talking about past experiences is particularly bene cial
for elderly people with dementia. Patients are still able to
recall past experiences until advanced stages of the disease,
and reminiscing about the past helps improve their mood
and well-being [
2
]. As music has been shown to activate
memory in people with dementia [
6, 8
], ReJAM uses
patients' favourite music to initiate past experience sharing:
we use music as a stimulus because the context of the
application is always music-related. The ontology describes
that music can trigger memories (also see Figure 2), and so
when a song is played - especially in the context of the
reminiscence activity - the agent can infer that it is reasonable
to ask whether the music evokes any special memories. If
the user con rms, the agent can use the information about
memories available in the ontology to ask follow up
questions, e.g. about the associated location or people involved.
The user's answers to these questions are stored in the
ontology for future reference, resulting in the construction of
a model of the user's life story.
Sharing new experiences in ReJAM: A common
challenge in dementia care is apathy of the patient. At advanced
stages of the disease patients become passive, causing family,
friends, and caregivers to experience a deterioration of their
relationship with the patient [
3
]. ReJAM aims to motivate
people with dementia to engage in activities with their
social environment, by proposing music-related activities, e.g.
playing a game with other patients, reminiscing with their
spouse, or singing a song with their grandchildren. All
activities are designed to scale from individual settings to group
settings, making ReJAM well-suited for a broad range of
social contexts.
In this paper we presented our framework for human-agent
experience sharing. The framework enables agents to share
both past and new experiences with their users. The
framework is currently implemented in a proof-of-concept for the
dementia care practice. Future research aims to investigate
its e ects on the human-agent relationship.
5.