chatbots. We envision that the chatbot will become
the front-end interface of almost everything digital.
Developing better chatbot technology requires
interdisciplinary attention. The backend technologies of chatbots rely upon domain expertise in infor-mation/cognitive summary, sentiment analysis, topic modeling, natural language processing, text
mining, and machine learning. The front-end technologies are built from a deep understanding of user
behavior and engagement, cognitive thinking, UI
design, and multimodal interaction design. In addition, designing chatbots entails the complicated
social interaction of individuals and groups, which
includes peer communication, conformity, social
and society pressures, trends, personality, social
bonding, and more.
Ying Ding, Bing Liu, James Shanahan, and Jie Tang
organized the workshop. No report was submitted to
Data-Driven Personas and
Automating Customer Insights in
the Era of Social Media
The Data-Driven Personas and Human-Driven Ana-
lytics workshop dealt with the use of online analyt-
ics data for automating customer insights, specifical-
ly via the use of personas. Automation can bring
many benefits, including increased speed of produc-
ing reports, removing judgment errors, and reducing
cost of customer analysis. However, efforts toward
automation also involve many challenges, such as
data retrieval, presentation, and validation of the
usefulness of such systems. We illustrated these prob-
lems by presenting a methodology called automatic
persona generation that summarizes real data on
online behaviors and demographics into easily inter-
pretable personas. This system and its key challenges
were introduced to the participants, after which we
divided into teams, so that each team picked a sub-
problem to solve.
There were four teams working on the problems.
Overall, the participants came up with a rich set of
ideas. One team was interested in the usefulness of
personas. They came up with an idea of an ad creation experiment to test the effect of personas on real
and tangible work output. The transparency of persona profiles was also noted as an issue: from prior
research, we know that not including explanations
of how information is produced can hinder trust in
users and negatively affect their experience of systems that rely on algorithmic filtering. The team concluded that determining how to provide explanations in the context of automatic analytics requires
empirical user studies.
Another team worked on the consistency problem
of automatic persona profiles, meaning that the
selected information in the persona profile is not
always internally coherent (for example, quotes are
from a male while the persona is female). Their
approach involved classifying comments by age, gen-
der, and location and then matching the appropriate
ones to demographic attributes to ensure persona
Another team was interested in the applicability of
social network analysis, that is, graph features, to personas. They postulated that in the absence of direct
information, we could learn about the preferences
and traits of the persona by analyzing the connections, such as followers, of individuals we have information about.
One team was interested in image generation, or
transformation, using generative adversarial networks. The theoretical promise of deep learning
could be to transform a set of seed images to closely
matching facial pictures with varying age, gender,
and even ethnicity. This solution is targeted toward
the challenge of finding and downloading thousands
of individual images to portray automatically generated personas.
In conclusion, even though the automation of customer analytics has many benefits, there are considerable technical and social challenges in the process.
Technical challenges relate to the availability of data
and especially the accuracy of inferring various user
attributes from user-generated social media content.
Moreover, automation has been questioned, as algorithms may pose bias and result in stereotypical or
nonmeaningful data portrayals. In our workshop, we
explored this duality of data- and human-driven analytics. While data-driven analytics is focusing on the
accuracy and availability of data to support decision-making, human-driven analytics is defined as the
presentation and analysis of insights about users or
customers that highlights qualitative attributes over
numbers. Human-driven analytics is based on users’
information needs by automatically adapting or giving users interactive choices for information presentation.
The workshop chairs were Jim Jansen, Joni Salminen, Lene Nielsen, and Matti Mäntymäki. This report
was written by Joni Salminen and Bernard J. Jansen.
Designed Data for Bridging
the Lab and the Field:
Tools, Methods, and Challenges
in Social Media Experiments
The Designed Data for Bridging the Lab and the Field
workshop explored the methodological middle
ground between the two environments of lab and
field. Research using repurposed observational data
from online platforms has transformed the study of
online behavior, and while such data has high exter-
nal validity, it also presents challenges for establish-