3. affective computing:
“relates to, arises from, or influences emotions” …
user modeling:
“systems that adapt […] based on an internal
representation of the user”
4. affective computing:
“relates to, arises from, or influences emotions” …
user modeling:
“systems that adapt […] based on an internal
representation of the user”
5. 1. what role do emotions play in personalised, user-
model based systems?
2. can we measure emotion accurately?
3. what can we compute using (representations of)
emotion and personality?
6. 1. what role do emotions play in personalised, user-
model based systems?
8. research: (mostly) driven by off-line studies
practice: (definitely) driven for recurring interaction
9. research: simulate recurring interaction with one of
three surveys. note: one survey did not change its
recommendations.
10. result (expected): people prefer those
recommendations that are diverse and change over
time.
research: let's build this into recsys & see how it
affects quantitative metrics.
N. Lathia, S. Hailes, L. Capra, X. Amatriain. “Temporal Diversity in
Recommender Systems.” in ACM SIGIR 2010, Geneva, Switzerland.
11. result (unexpected):
“there is a bug in your system.. it sux […]”
“i used to work for [famous recsys company];
building a recommender system is not hard...”
– angry, frustrated; low ratings may encode
“punishing” a system (beyond preference)
13. 2. can we measure emotion accurately?
what are we measuring? how do we measure it?
14. “people might be said to have an implicit theory of
emotions […] laymen's cognitive representation of
emotion is presumably implicit in the sense that few
if any could explicitly state their conceptual
framework...”
- Russel (1980)
J. A. Russel. A Circumplex Model of Affect. Journal of Personality and Social
Psychology. Vol 39, No. 6. 1980.
15. “...investigators who have factor analyzed self-
reported affective states have typically concluded
that there are between six and twelve independent
monopolar factors of affect...”
- Russel (1980)
J. A. Russel. A Circumplex Model of Affect. Journal of Personality and Social
Psychology. Vol 39, No. 6. 1980.
16. users are being asked to perform two separate
tasks: (1) estimate, based on self-knowledge, and
(2) translate onto a rating scale... and this
sometimes causes problems*
* X. Amatriain, J. Pujol, N. Oliver. “I Like It.. I Like It Not... Measuring Users Ratings
Noise in Recommender Systems.” In UMAP 2009, Trento, Italy
17. 2. can we measure emotion accurately solicit quick,
meaningful representations of emotions?
23. early results: consistent, highly correlated usage of
the affect grid.
but.. not everyone is using it correctly;
predictions of r (adjective) ~ (x, y, ….) ongoing
and, more importantly...
24. an earlier study (22 participants, 4 weeks)
experimented with the question “when do we ask
users how they feel?”.
25. N. Lathia, K. Rachuri, C. Mascolo, P. Rentfrow. Contextual Dissonance: Design Bias in
Sensor-Based Experience Sampling Methods. To appear, ACM Ubicomp 2013, Zurich,
Switzerland.
[…] negative affect ratings [...] were significantly
different from one another with at least 90%
confidence […] we observe that our design
parameters influence the outcome...
31. “...the specifics of the context surrounding people’s
day-to-day living are much more subtle, fluid and
idiosyncratic than theories of context have led us to
believe...”
- Y. Rogers
Y. Rogers. Moving on from Weiser's Vision of Calm Computing: Engaging Ubicomp
Experiences. In ACM Ubicomp 2006. Orange County, USA
35. early results: augmenting music recommendation
algorithms with personality data improves average
ranking by more than 10%
but.. this still doesn't seem to outperform just
“working harder” with the rating data
36. 1. what role do emotions play in personalised, user-
model based systems?
2. can we measure emotion accurately?
3. what can we compute using (representations of)
emotion and personality?
37. References
1. R. W. Picard. Affective Computing. M.I.T Media Laboratory Perceptual Computing
Section Technical Report No. 321
2. N. Lathia, S. Hailes, L. Capra, X. Amatriain. Temporal Diversity in Recommender
Systems. In ACM SIGIR 2010, Geneva, Switzerland.
3. J. A. Russel. A Circumplex Model of Affect. Journal of Personality and Social
Psychology. Vol 39, No. 6. 1980.
4. X. Amatriain, J. Pujol, N. Oliver. “I Like It.. I Like It Not... Measuring Users Ratings
Noise in Recommender Systems.” In UMAP 2009, Trento, Italy
5. N. Lathia, K. Rachuri, C. Mascolo, P. Rentfrow. Contextual Dissonance: Design
Bias in Sensor-Enhanced Experience Sampling Methods. To appear, ACM Ubicomp
2013. Zurich, Switzerland.
6. Y. Rogers. Moving on from Weiser's Vision of Calm Computing: Engaging
Ubicomp Experiences. In ACM Ubicomp 2006. Orange County, USA