Sentiment and emotion analysis, on the other hand,
has been applied to both text and multimodal
datasets, but this research has been limited to quantifying well-defined human reactions. Affect analysis
(that is, techniques and applications to understand
the experience of an emotion) in the context of language and text is an up-and-coming research space.
Work on affect analysis in language and text spans
many research communities: computational linguistics, consumer psychology, HCI, marketing science,
and cognitive science. Computational linguists study
how language evokes, as well as expresses, emotion.
Consumer psychology examines human affect by
drawing upon grounded psychological theories of
human behavior. The HCI community studies human
responses as a part of user experience evaluation.
Computational models for consumer psychology theories present a huge opportunity to guide the construction of intelligent systems that understand
human reactions, and tools from linguistics and
machine learning can provide attractive methods to
fulfill those opportunities. Models of affect have
recently been adapted for social media platforms,
enabling new approaches to understanding user’s
opinions, intentions, and expressions.
The workshop focused on the analysis of emotions, sentiments, and attitudes in textual, visual,
and multimodal content for applications in psychology, consumer behavior, language understanding,
and computer vision. Besides original research presentations and posters, the workshop also hosted a
range of keynote speakers who highlighted the state
of the art in affective computing in a range of fields.
James Pennebaker from the University of Texas
Austin provided evidence from a series of studies
about how affect and emotion can be mined from
the words used by people in everyday life. Dipankar
Chakravarti from Virginia Tech discussed some of the
challenges involved in affective analysis of text for
consumer behavior, especially noting the differences
between the experience and the expression of affect.
Bjoern Schuller from the University of Augsberg, Germany, provided insight into a range of applications
of affect analysis from speech, music, and audio.
Rajesh Bagchi from Virginia Tech shared work in the
space of consumer psychology and marketing science, focusing on the affective processing of information and its relationship to consumer behavior.
Cristian Danescu-Niculescu-Mizil talked about the
application of affective computing in conversational
dynamics, in group discussions, and with respect to
the outcomes of decision-making discussions. Jennifer Healey from Intel discussed her work in multimodal affect analysis as a part of the cutting-edge
research on building emotionally aware robots that
can intuit and respond to human emotions.
The workshop ended with a panel discussion
among the keynote speakers, moderated by the
organizers, on the potential grounds for interdisci-
plinary collaborations, as well as venues for such
events in future.
Niyati Chhaya, Kokil Jaidka, Lyle Unger, and P.
Aanandan cochaired the workshop. This report was
prepared by Niyati Chhaya and Kokil Jaidka. The
workshop papers were published in the AAAI digital
and Marketing Science
The growth in online data across marketing, cam-
paign, display, programmatic advertisements, and
social platforms has focused the attention of AI and
machine learning researchers on developing new and
more efficient computational models and tech-
niques. Among marketing science (MS) researchers,
the emphasis has been on exploiting ML methods to
address business problems in marketing resource
optimization, managerial decision-making, competi-
tive behavior modeling, deconstruction of consumer
behavior, and campaign automation and optimiza-
tion. These two vibrant research communities both
investigate problems central to marketing, but pub-
lish in separate journals and conferences. The moti-
vation for the workshop was to start the dialogue for
bridging that divide. The desired, longer term out-
come is the two communities benefiting from each
other’s research, problems, and insights, leading to
the higher effectiveness of models and methods in
both theory and application.
The workshop showcased four invited keynote
speakers to explore a number of key themes: Craig
Boutilier (Google Inc.), Vince Conitzer (Duke University), Harikesh Nair (Stanford University and
JD.com), and Dave Weinstein (Adobe Systems Inc.).
The 11 paper presentations included four posters.
Boutilier presented research focusing on sequential
decision-making, a major area straddling both AI and
MS. His presentation highlighted advances in
Markov decision processes (MDPs) to tackle problems
in online advertising related to understanding the
long-term impact of advertisements. His recent work
includes logistic MDP and stochastic action set MDP.
The latter is useful when new, exogenous actions are
to be considered (for example, a new advertising
campaign). In the context of marketing decision-making, algorithms must address the double issue of
partial observability and multiple goals, which is
where MDPs and reinforcement learning become
particularly relevant. Voice of the customer data,
including information on user goals, can be mined
to help with the goal-identification tasks important
Conitzer presented research that addresses the
problem of pacing in selecting bids on a demand-side
platform. Consider that a marketer asks an agent (the
platform can also be the agent) to bid on its behalf
subject to a daily budget. An important problem is