New project: Using Natural Language Processing and Generation to develop and test health-related chatbots

Chatbots have great potential for digital communication: they are always accessible, can do many interactions in parallel and engage in personalised interactions. However, the current generation of chatbots tends to have only a very limited set of responses, produce interactions that can be perceived as meandering, and cannot adopt adequately to the user’s knowledge and needs, as derived for the user’s responses. As a result, people tend to dissociate from chatbots quite rapidly, meaning their potential is presently unfulfilled.

In the present project, we integrate expertise from health communication, health psychology, and digital humanities to create and test chatbots that aim to foster long-term relationships between users and chatbots by having chatbots engage in human-like conversations. We aim to do this by iteratively developing chatbots that have a ’theory-of-mind’ and an interaction memory, for which the chatbot can subsequently engage in varied conversations. We use recent insights from computational linguistics, motivational interviewing, and persuasion and test our chatbots for both sexual health promotion and smoking cessation.

The project is a collaboration between University of Amsterdam, Radboud University, and Tilburg University. From the Digital Communication Methods Lab, dr. Gert-Jan de Bruijn is involved in the project and will supervise two PhD students.

Prof. Claes de Vreese appointed Faculty Professor on Artificial Intelligence

Prof. Claes de Vreese has been appointed Faculty Professor on Artificial Intelligence at the Faculty of Social and Behavioral Sciences at the University of Amsterdam as of September 1. Claes is the director of the RPA Communication and its Digital Communication Methods Lab, and Co-PI at the Research Priority Areas Personalised Communication and Human(e) AI.

As a Faculty Professor, Claes will accelerate his work investigating the societal consequences of Artificial Intelligence, including its effects on decision-making, democratic processes and politics. An important aspect of the role is to continue building upon and expanding collaborations within the faculty, as well as across the University of Amsterdam and with external partners, with a strong interdisciplinarity focus.

Selected proposals for Thesis Funding Grants announced!

The Digital Communication Methods Lab is happy to announce the selected proposals for the first edition of the Thesis Funding Grants. These grants provide provide (financial) support for theoretically-relevant and digitally innovative (research) master theses for semester 2 of the 2018-2019 academic year.

Cultivating perceptions of credibility in the context of online conversational agents: the role of similarity in personality and expertise claims

by Eirine Ntaligkari

The adoption of conversational agents (CAs) as social actors that can substitute other humans (Zhao, 2003) is a new form of corporate communication between stakeholders and organizations, as technological advances continue to drive CAs’ capabilities significantly. An interesting scope capacity of CAs is personalization of services, that allows organizations to raise their business intelligence and provide customized products or services, and users to enjoy tailored services that have been found to have multiple benefits (Wang & Li, 2012; Godey et al., 2016; Perna, Runfola, Temperini, & Gregoni, 2018). To achieve this, CAs need to acquire personal information, which users are rather sceptical to disclose due to uncertainty related to security and privacy risks (Aguirre, Roggeveen, Grewal & Wetzels, 2016).

In order for users to feel more comfortable to disclose such information to CAs, trust and credibility must be present. Nonetheless, credibility and its antecedents (namely, trustworthiness and expertise) are values that form gradually and are difficult to acquire in an online environment, especially through micro interactions such as the ones mentioned above. The assessment of credibility and the effects of its antecedents in such environments remains unresearched. By utilizing an CA created with the aid of the Conversational Agent Research Toolkit (CART) of UvA’s Digital Communication Methods Lab, this thesis aims to answer the following question: How can trustworthiness and expertise be embedded in CAs in a meaningful way, in order for an organization to influence not only the source, but also their corporate credibility, and therefore privacy and security risk perceptions?

How can we characterize disinformation in online news? Developing machine learning classifiers for examining structural differences in U.S. and Russian state-backed news in Serbia

by Ognjan Denkovski

Democratic nations globally are experiencing increasing levels of false and misleading information circulating through social media and political websites, often propagating alternative socio-political realities. One of the main actors in this process has been the Russian state, whose organized disinformation campaigns have influenced elections throughout the Western world. A key element of these campaigns has been the dissemination and spread of content produced by outlets like RT and Sputnik – content thereafter spread by underfunded local media and organized online networks which attempt to shape mainstream political narratives. In response to a lack of comprehensive research examining the characteristics of such content, this paper examines whether, and if so how, content produced by Russian websites like Sputnik is structurally distinct from that of mainstream Western outlets. Through text-as-data methods we examine: a) the stylistic and thematic differences in content produced by U.S. and Russian backed outlets in Serbia, a key geopolitical interest for both states and b) which features, if any, of Russian news can be used to characterize content as part of disinformation campaigns. These findings are used for the development and evaluation of supervised machine learning classifiers. The paper contributes towards an understanding of the structural characteristics of disinformation and online political polarization in a novel context – Balkan online news – while also forwarding the application of text-as-data methods in Serbian. Ideally, the project will allow for the development of an online suspicious news identifier for the Balkan languages.

Is it a big problem or not? An analysis of fake news diffusion on social media during the Brazilian 2018 presidential elections

by Pieter Attema Zalis

The goal is to provide new evidences on how ‘fake news’ spreads in social media during electoral campaigns. This study will analyse the subject in the context of the 2018’s Brazilian presidential election. As it occurred in the US with Trump, Brazil elected an unlikely candidate, Jair Bolsonaro. Due to Bolsonaro’s anti-establishment populist style and social media presence, international media outlets nicknamed him as the “Tropical Trump”.

This study has three main goals. First, following previous studies in the US, I’ll analyse if ideology (left vs right) might moderate the differences on fake news diffusion in social media. In the second and third steps, I’ll compare fake news to traditional news. I’ll first look if fake news stories present a more emotional content than traditional news stories and latter investigate if this leads to higher levels of fake news sharing compared to traditional ones. Second, i’ll test, in the Brazilian context, evidences found in Europe and in US that the general audience of fake news is significantly smaller than traditional news. To sum up, I have two research questions about the subject: (1) to what extent are interactions of fake news bigger or smaller than of traditional news in social media? (2) to what extent is the total amount of fake news bigger or smaller than traditional news in social media? I’ll answer these questions with a content analysis of 5,120,892 tweets streamed during the second turn of the presidential election.

Mapping CSR-crisis in issue arenas: applying the Network Agenda-Setting Model in big-data research

by Louelle Jasmin Pesurnaij

In the past decades, more and more plastic is floating in the oceans and seas as a result of disposed plastic products, such as food packaging, fishnets, synthetic clothing, toothbrushes and plastic furniture. In 2011, the Plastic Soup Foundation was founded, whose mission is to tackle the plastic pollution of oceans. Campaigning activities of NGOs such as the Plastic Soup Foundation have proven to increase public awareness and concern regarding social, ethical and environmental issues and have proven to be powerful at setting the public agenda by urging the public towards a pro-environmental stance. However, it remains unclear what the most effective way is to design campaign and/or news messages and how to find the right issues and issue arenas.

The aim of this study is to gain a more complete picture of the agenda-setting role of the media and NGOs in virtual environments. This study seeks to apply a relatively new concept in the field of communications: the Network Agenda-Setting (NAS) model, to research the capability of both news media and NGOs to influence how the public links different messages regarding #PlasticSoup. The NAS model extends traditional agenda-setting research, by asserting that issues and attributes are not just transferred as individual elements but issues and/or attributes are interconnected and transferred in bundles to the public agenda. The research question of this thesis study is as follows: “To what extent are the issue attribute networks regarding #PlasticSoup transferred from the agenda of media and NGOs onto the public agenda?”.

The present study uses innovative, digital research methods, such as automated content analyses of big datasets (including semantic network analysis), to identify issues and issue attributes in tweets about #PlasticSoup and to examine the agenda-setting power of both news media and NGOs.

Paid partnership, #ad or ambiguous hints to brands on Instagram? The impact of sponsorship disclosures, alternative cues and different influencer types on users’ persuasion knowledge

by Céline Marie Müller

Instagram, today’s most relevant platform for influencer marketing has received little scholarly consideration regarding this form of native advertising . Visual attention is a crucial indicator for whether and when people recognize Instagram ads. To date, it is unclear what elements help consumers to identify sponsored content. The few studies that explored users’ responses to sponsorship disclosures focused on different hashtags and Instagram’s platform-based disclosure (Evans, Phua, Lim, & Jun, 2017; Coursaris, Van Osch, & Kourganoff, 2018). But are these disclosures actual key elements that help users recognize sponsored content? Previously, the potential effectiveness of other cues (i.e. brand tags in picture or caption) has been neglected. With an eye-tracking experiment, this study clarifies the cues or combinations that successfully help consumers identify sponsored posts. Furthermore, it detects possible boundary conditions of the different disclosures’ value by distinguishing between posts of macro-, micro- and nano-influencers. A recent study found that users are less likely to identify sponsored posts by micro- compared to macro-influencers (Coursaris et al., 2018). Heretofore, consumers’ ad recognition for sponsored posts by nano-influencers remains underesearched. Based on the eye-tracking findings, a second online experiment elucidates the impact of influencer types and particular disclosures or elements on users’ persuasion knowledge and brand responses. In order to fill the above-mentioned gaps in academia, the thesis seeks to answer the following question:
To what extent do sponsorship disclosures, alternative cues and different influencer types affect users’ visual attention to Instagram posts, their persuasion knowledge and resulting brand responses?

Margot van der Goot wins Strategic Communication Challenge for research on chatbots

Margot van der Goot has won the Strategic Communication Challenge 2019–organized by the “Centrum Strategische Communicatie” of Logeion (the Dutch association of communication professionals). This means that Margot receives €25,000 to conduct a qualitative study that aims to provide in-depth insights in how consumers experience chatbot communication in a customer service setting. More information can be found here.

Virtual patients’ effectiveness in training health providers presented at the Workshop on AI and Conversational Agents @ TU Delft

Can a virtual patient help health care providers to improve their communication skills? Julia van Weert will discuss this during the Workshop Artificial Intelligence and Conversational Agents at the TU Delft on January 24 and 25.

She will present the results of one of the projects from the Digital Communication Methods Lab in which virtual patients (VPs) were integrated in an interactive VP-based learning module, aiming to enhance shared decision making (SDM) skills in oncology.

Based on computational models of data obtained from previous studies on SDM in oncology, the VP provided the trainees (in this study medical students) the illusion of being engaged in a free speech natural dialogue as part of a SDM consultation. Results indicate that using a VP has potential as an effective learning tool for communication skills training.

Lab group on Conversational Agent Research

The Digital Communication Methods Lab is launching a group to discuss conversational agent and AI related research. The idea is to create a space in which we can discuss ongoing research, identify opportunities for collaboration, and build a network of researchers interested in the topic. We’ll organise monthly discussion sessions.

Members of ASCoR, RPA Personalised Communication, and RPA Communication have already joined, and some of the ongoing projects include surveys of usage of chatbots in health-related information and news distribution, analysis of persuasive outcomes of conversational agents and brand communication. For more information, please contact Theo Araujo and Judith Möller.

New projects on Media Exposure

Ewa Maslowska and Lukas Otto
Ewa Maslowska and Lukas Otto will start new projects in the Digital Communication Methods Lab, part of the RPA Communication, as of 2019!

Ewa’s project aims to investigate how we can reliably measure which elements of websites consumers are exposed to, in order to advance our knowledge about the effects of these elements on consumer information processing and decision making. The project will develop eye-tracking measurements for exposure to online content that are predictive of consumer behavior and examine to what extent those measurements correlate with other (digital) methods (e.g., mouse-move tracking). By doing this the project aims to create a stepping stone to understand the broader theme of consumer engagement. Furthermore, it aims to develop methodologies that will be applicable to other context of online communication than the online consumption context.

Memory failures and biased self-assessment are seen as severe challenges for the measurement of media exposure. Facing this challenge, Lukas’ project will add to the methods repertoire by introducing the mobile experience sampling method to media exposure research. Participants will indicate their political media usage via Smartphones directly after media exposure to minimize memory effects and multiple times per day to model media selection dynamics as they unfold over time.

Digicomlab researcher receives Diabetes Breakthrough Grant to develop and test virtual reality training for healthy grocery shopping

Gert-Jan de Bruijn has received a breakthrough grant from Diabetes foundation and ZonMw to develop and test the effectiveness of working memory training in virtual reality settings to promote healthy grocery shopping in virtual and real-life supermarkets.

It is well-known that, when people move to an area where there is a high density of health-harming food outlets, their odds for developing diabetes type 2 can increase by as much as four times. The idea behind this effect is that when people are exposed to unhealthy food cues in those outlets, they are more likely to purchase and consume those unhealthy foods. This is particularly the case when people are vulnerable to those cues, such as when people are in a state of hunger, have high blood glucose levels, or are diabetic.

Because intervening in real-life supermarkets is problematic due to a variety of reasons, virtual reality approaches are gaining traction in studies on health decision making. In this project, Gert-Jan (together with colleagues from Tilburg University, Free University Medical Center, and the Department of Psychology of the University of Amsterdam) will transfer offline working memory training strategies to a virtual reality mobile app setting. In this virtual reality setting, several external (store crowdedness, hedonic temptations) and internal (glucose levels, hunger) will be manipulated. After training, participants will do grocery shopping in an immersive virtual reality supermarket. During the month following the virtual reality training, the real-life shopping behaviours of the participants will also be assessed. The virtual reality training app is scheduled to be released via (among others) Digicomlab.

More details (in Dutch)

Digicomlab project with virtual supermarket at the Innovation Expo in Rotterdam

The virtual supermarket that is currently used in the project Testing Package Complexity in an Innovative 3D Virtual Supermarket Environment by prof. Edith Smit, dr. Nynke van der Laan, and dr. Corine Meppelink will be present at the Innovation Expo in Rotterdam!

The virtual supermarket has been developed at UMC Utrecht and the University of Amsterdam (UvA). This is a fully immersive shopping environment in which people can walk around and pick up products, and is used to understand how people make choices in a supermarket. More information here.

Digicomlab-members receive funding to start a Mobile Lab

Bert Bakker and Claes de Vreese together with Gijs Schumacher and Frans Oort (both UvA) have received an NWO Investment Grant to open a Mobile Lab (more information here). As Bert explains:

Excellent social science research requires studies that are both internally and externally valid. Currently, there are many obstacles to achieve this. Lab studies often produce high internal validity, but primarily rely on student samples (external validity issue) and are conducted in clinical settings (ecological validity issue). To address this researchers move their lab to the field (e.g., schools, sporting events, museums or festivals). But researchers face high transaction costs with these so-called lab-in-the-field studies: there are many practical issues to consider and some of these even threaten the internal validity of the study.

We offer a simple and innovative solution by creating a mobile social science lab – The Mobile Lab. This facility enables researchers to easily set-up a lab across the country on specific locations. We will invest in mobile electroencephalography (EEG), eye-tracking and psychophysiology equipment as well as mobile cameras for in-depth interviews and behavioural observations. The Mobile Lab will be a national infrastructure and enable Dutch social scientists to test their theories on high-quality samples. In the coming months, the team will start developing the lab. More information will follow here soon.