HCI International 2020 – Late Breaking Posters

Inhaltsverzeichnis

1. Interaction and Intelligence

   1. Frontmatter

   2. Experiencing AI in VR: A Qualitative Study on Designing a Human-Machine Collaboration Scenario

      Alexander Arntz, Sabrina C. Eimler

      Abstract

      This paper describes the setup and results of a qualitative interview study, in which participants were given the opportunity to interact with an AI-based representation of a robotic-arm in a virtual reality environment. Nine participants were asked to jointly assemble a product with their robotic partner. The different aspects of their experiences, expectations and preferences towards the interaction with the AI-based industrial collaboration partner were assessed. Results of this study help to inform the design of future studies exploring working arrangements and communication between individuals and robots in collaborating together.

   3. Interacting with a Salesman Chatbot

      Charlotte Esteban, Thomas Beauvisage

      Abstract

      In recent years, chatbots have been spreading on social networks and brand websites, and interactions between users and commercial chatbots have become an ordinary experience in the range of human-computer interactions. Yet, whereas automated conversation has been analyzed in various experimental contexts, only a few studies describe real-world interactions with voicebots [1‐3] or chatbots [4‐6]. How do interactions with chatbots actually take place? What is an AI-driven commercial conversation in practice? To address these questions, we conducted a sociological study of interactions with chatbots, based on the quantitative and qualitative analysis of interaction logs with a vending chatbot, deployed on a French online telecom company. The study relies on a dataset of 9 months of ComBot usage logs in 2019, representing roughly 47,000 interaction sessions. Our analysis shows that interactions with the commercial chatbot are a highly hybrid format between click-based interfaces and conversational interactions. A majority of users mobilize the conventions of commercial conversation to express their need in plain text. However, the rest of the dialogue mainly combines response-buttons, short input text, and hyperlinks. The use of politeness shows that users are keen on following the conversational interaction format offered to them, even if they don’t use it entirely.

   4. An Empirical Study on Feature Extraction in DNN-Based Speech Emotion Recognition

      Panikos Heracleous, Kohichi Takai, Yanan Wang, Keiji Yasuda, Akio Yoneyama, Yasser Mohammad

      Abstract

      The current empirical study focuses on speech emotion recognition using speech data extracted from video clips. Although many studies reported speech emotion recognition, the majority of the studies presented were based on using acted and clean speech. A more challenging and realistic task would be using spontaneous noisy speech from video clips. In the current study, the modern and state-of-the-art i-vector features are applied and experimentally evaluated. Comparisons with the widely used low-level descriptors (LLDs) and functionals are also presented. To improve the classification accuracy, a method based on late fusion is investigated. Using the proposed method, higher accuracies were achieved compared to the sole use of individual features. For classification, a fully connected deep neural network (DNN) with several hidden layers was used.

   5. Develop an Interactive Model of Impact of Basketball Players and Team Performance

      Yun-Chi Huang

      Abstract

      This study aims to develop an analytical model of basketball team performance based on substitution method (Rotational, Fatigue, Best Fit), player’s usage percentage, and individual’s point produce (Calculation extended with field goal percentage/attempt of 2P, 3P and FT). The model is expected to predict game results and analyze team management strategies. This study develops four strategies based on the interactive model: (Rotational versus Best fit) Substitution and (Star player versus Average) Usage Rate with scoring value X (The individual player’s value of shooting percentage) and value S (The summation index of all scoring value). It is found that scoring performance of Best fit Substitution and star player Usage Rate strategy is highly predictable. Also scoring performance of Best fit Substitution is usually more precise than Rotational Substitution.

   6. Human-Centered Artificial Intelligence: Antecedents of Trust for the Usage of Voice Biometrics for Driving Contactless Interactions

      Rohan Kathuria, Ananay Wadehra, Vinish Kathuria

      Abstract

      Covid-19 driven pandemic situation has brought greater visibility to contactless interactions with consumer IoT devices. As industries explore transitioning away from shared touch devices, the role of voice biometrics for authentication becomes critical. Voice biometrics is utilized for voice recognition through analysis of an individual’s pitch, speech, voice, and tone and has been used in back-office operations for customer verification, fraud avoidance, and password reset. However, not much research has been done in the consumer sector and the critical role of trust in driving usage and the adoption of such services.

      Using the existing research on trust in e-commerce and automation, we bring together models from psychology (Theory of Planned Behaviour) and technology (Human-Centered Artificial Intelligence), to explore the various antecedents of consumer trust for voice authentication (Ease of use, self-efficacy, perceived usefulness, reliability, the perceived reputation of the service provider, perceived security, perceived privacy, fraud, and social influence). Special attention is given to the use of vernacular voice, two-step authentication, and their impact on trust. Speaker recognition is a pattern recognition problem and incudes various technologies like frequency estimation, hidden Markov models, Gaussian mixture models, pattern matching algorithms, neural networks, decision trees, and linear predictive coding.

      Through a combination of custom build prototypes, usage scenarios, and qualitative and quantitative analysis, we intend to highlight the components that drive trust for voice authentication so that it can help in the societal transition to contactless interactions. Early results show that people value Security, Privacy, and Reliability as top factors impacting trust in Voice Biometrics. Multi-level authentication, vernacular voice, and initial usage for transactional and low-value financial transactions can help drive trust in the voice biometrics ecosystem.

   7. An HCI Approach to Extractive Text Summarization: Selecting Key Sentences Based on User Copy Operations

      Ilan Kirsh, Mike Joy

      Abstract

      Automatic text summarization is a very complex problem. Despite being intensively researched, automatic summaries are still considered to be of lower quality than manual summaries. This paper introduces a novel HCI approach to web page summarization. The proposed Crowd-Copy Summarizer follows the extractive text summarization approach of summarizing by selecting sentences within the text. The selection is performed by examining how frequently users copy certain sentences to their clipboards, for their own purposes. The most frequently copied sentences are included in the summary. Results from an early experiment are promising, as key sentences, such as introductory sentences, definitions, and important highlights, are copied frequently. Consequently, the generated summaries can provide good coverage of the main topics. This novel text summarization approach combines the best of both worlds: summarization based on collective human wisdom, without the expensive burden of manual summarization work.

   8. Infrequent Use of AI-Enabled Personal Assistants Through the Lens of Cognitive Dissonance Theory

      Nicole O’Brien, Maarif Sohail

      Abstract

      The current availability of several versatile and powerful Artificial Intelligence-Enabled Personal Assistants (AIEPA) along with the unique phenomena of far less usage of these devices serve as the primary reason for motivation for the development and contribution of this research. This research explores the infrequent usage of commonly available commercial Artificial Intelligence Enabled Personal Assistants (AIEPA) like Alexa and Google Assistant with the help of Cognitive Dissonance (CD) theory. We propose a model that helps to investigate the phenomenon of infrequent usage. We also share our view that the theory of cognitive dissonance can be a way forward to study the usage behavior of the end-users, which can improve the performance of these devices as well as reduce concerns that act as barriers in adoption, acceptance, and usage.

   9. Concept for Human and Computer to Determine Reason Based Scene Location

      Adrienne Raglin, Andre Harrison

      Abstract

      The motivation of human computer interaction (HCI) is to improve how humans use computers. One approach can be developing technology that allows the human and computer to collaborate. In some cases creating this type of collaboration or joint approach to for example solve a problem or identify an important piece of information. While some researchers investigate ways that do not include the human, HCI focuses on keeping the technology linked to the human. At this stage in our research, the human is the collaborator utilizing the computer as a teammate to accomplish the task. The task for this work falls under the area of ongoing research in scene understanding. We present a concept that allows the location of an image to be determined based on object detection performed by the computer and the human using reasoning to generate possible candidates for the location that an image can represent. The human may use his or her own knowledge to reason about the options or again working with the computer to glean from reasoning engines that include knowledge. The paper will present this idea and the work that has started.

   10. A Neural Affective Approach to an Intelligent Weather Sensor System

       John Richard, James Braman, Michael Colclough, Sudeep Bishwakarma

       Abstract

       The ability to capture data from our surrounding environment while learning user preferences has the potential to make our everyday decisions more straightforward and informed. Based on the idea of capturing weather data, we present our current design of a weather sensor network using several Raspberry Pi’s, combined with external resources, that presents recommendations based on personalized affect data. The goal of the system is to learn user preferences in combination with providing emotive output utilizing a neural network. The system has visualization capabilities that can interface with the web, along with other features based on preferences set by the user. This paper presents an overview of the system prototype.

   11. Role-Based Design of Conversational Agents: Approach and Tools

       Ilaria Scarpellini, Yihyun Lim

       Abstract

       The wide adoption of conversational agents in delivering services to users asks for user-centric approach to design of the experience. We propose a Role-Based design approach, introducing 9 archetypal roles/purposes of conversational agents, resulting from case study research of conversational interfaces available in the market. These roles cover a range of behaviors (reactive, proactive) and features (relational, operational) that will enable the design of human-like, bi-directional conversational experience. We exemplify this approach through workshop sessions that involved stakeholders from financial industry in using the developed toolkits to envision financial service experiences enabled by conversational-agent.

   12. How Users Reciprocate to Alexa

       The Effects of Interdependence Florian Schneider

       Abstract

       With the launch of Siri, a conversational assistant presented by Apple in 2011, voice-enabled personal assistants have since been accessible to the masses (Hoy 2018). As speech is the main channel for communication between humans (Flanagan 1972; Schafer 1995) and is considered to be an innate human behavior (Pinker 1994), interacting with a voice interface is intuitive (Cohen, Giangola and Balogh 2004). Studies conducted under the Computers Are Social Actors (CASA) paradigm indicate that speech-output and interactivity are two main factors to elicit social reactions in users (Nass et al. 1993). Users adopt human principles like reciprocity and team affiliation when interacting with computers (Nass and Moon 2000). As voice assistants are able to send social cues we assumed that subjects will show social reactions towards an Amazon Echo. Focussing on the social norm of reciprocity, we measured if people provide more help to the assistant after being told that they are interdependent of each other when compared to being independent.

       A laboratory experiment with 120 participants was conducted. Participants played an interactive game using an Amazon Echo. Team affiliation was manipulated by telling one group their game performance would be rated individually while telling the interdependent group that they are being assessed on their joint performance with the assistant. To operationalize reciprocity, we opted for a behavioral measurement to assess participants’ willingness to help the assistant: Participants were asked to name potential cities in which the game could take place. Participants were able to name any number of cities, assuming a relationship between the number of cities named and the level of cooperativeness towards the assistant.

       Results show a significant main effect of interdependence on the evaluation of Alexa as well as the number of cities named meaning participants did show reciprocal behavior towards the voice assistant.