Skip to main content
main-content

Über dieses Buch

This book constitutes the refereed proceedings of the 10th International Conference on Intelligent Technologies for Interactive Entertainment, INTETAIN 2018, held in Guimarães, Portugal, in November 2018. The 15 full papers were selected from 23 submissions and present developments in artificial intelligence for human interaction or entertainment; artificial intelligence in games, augmented reality and virtual reality; intelligent human-computer interaction; and other Intelligent interaction or entertainment applications covering a wide range of areas from smart cities to visual analytics and marketing.

Inhaltsverzeichnis

Frontmatter

Artificial Intelligence and Autonomous Systems

Frontmatter

Syntropic Counterpoints: Philosophical Content Generated Between Two Artificial Intelligence Clones

Abstract
In the project Syntropic Counterpoints, we are using discussions between Artificial Intelligence clones to generate creative content. Nevertheless, our focus is less on content analysis and more on the beauty of creation itself and given context by the machines. We are using a different recurrent neural network (RNN), and collective creativity approaches to support interactions between Artificial Intelligence clones and trigger a humanless creative process which should lead to unsupervised robot creativity. The robots are trained by using the publications of some of the greatest thinkers of their time such as Aristotle, Nietzsche, Machiavelli, Sun Tzu and confronted to the crucial questions related to humankind such as understanding of moral, aesthetic, ethic, strategy, politics, etc. Throughout this robot-robot interaction model, we are trying to investigate the possibilities and consider limitations of using artificial intelligence in context-based creative processes as well as to raise questions related to potential future phenomena of machines mindfulness.
Predrag K. Nikolić, Hua Yang

A Brief Overview on the Evolution of Drawing Machines

Abstract
Through the pictorial narratives engraved on the walls of the caves during prehistory, we are sure that Humans used drawing to express feelings and communicate, long before inventing writing. In the same way that utensils were used to help him, he also used several utensils to draw.
In the middle of the twentieth century, with all the technological evolution, we saw machines that helped artists in drawing and others that are extensions of the artist.
In a project seeking the development of a robotic system capable of drawing autonomously we were faced with the question for how long artists have used drawing machines for their aid or even their extension? In this work, we present a collection of artworks that demonstrates the use of drawing machines throughout history in the last 500 years and how they are being adapted and reinvented according to the most current and also developing technology. At present there is a vast field of experimentation of these machines with Interfaces and Sensors and Intelligent Human-Computer Interaction.
António Coelho, Pedro Branco, João Martinho Moura

Health-Centered Decision Support and Assessment Through Machine Reasoning

Frontmatter

Compression-Based Classification of ECG Using First-Order Derivatives

Abstract
Due to its characteristics, there is a trend in biometrics to use the ECG signal for personal identification. There are different applications for this, namely, adapting entertainment systems to personal settings automatically.
Recent works based on compression models have shown that these approaches are suitable to ECG biometric identification. However, the best results are usually achieved by the methods that, at least, rely on one point of interest of the ECG – called fiducial methods.
In this work, we propose a compression-based non-fiducial method, that uses a measure of similarity, called the Normalized Relative Compression—a measure related to the Kolmogorov complexity of strings. Our method uses extended-alphabet finite-context models (xaFCMs) on the quantized first-order derivative of the signal, instead of using directly the original signal, as other methods do.
We were able to achieve state-of-the-art results on a database collected at the University of Aveiro, which was used on previous works, making it a good preliminary benchmark for the method.
João M. Carvalho, Susana Brás, Armando J. Pinho

Predicting Postoperative Complications for Gastric Cancer Patients Using Data Mining

Abstract
Gastric cancer refers to the development of malign cells that can grow in any part of the stomach. With the vast amount of data being collected daily in healthcare environments, it is possible to develop new algorithms which can support the decision-making processes in gastric cancer patients treatment. This paper aims to predict, using the CRISP-DM methodology, the outcome from the hospitalization of gastric cancer patients who have undergone surgery, as well as the occurrence of postoperative complications during surgery. The study showed that, on one hand, the RF and NB algorithms are the best in the detection of an outcome of hospitalization, taking into account patients’ clinical data. On the other hand, the algorithms J48, RF, and NB offer better results in predicting postoperative complications.
Hugo Peixoto, Alexandra Francisco, Ana Duarte, Márcia Esteves, Sara Oliveira, Vítor Lopes, António Abelha, José Machado

A Many-Valued Empirical Machine for Thyroid Dysfunction Assessment

Abstract
Thyroid Dysfunction is a clinical condition that affects thyroid behaviour and is reported to be the most common in all endocrine disorders. It is a multiple factorial pathology condition due to the high incidence of hypothyroidism and hyperthyroidism, which is becoming a serious health problem requiring a detailed study for early diagnosis and monitoring. Understanding the prevalence and risk factors of thyroid disease can be very useful to identify patients for screening and/or follow-up and to minimize their collateral effects. Thus, this paper describes the development of a decision support system that aims to help physicians in the decision-making process regarding thyroid dysfunction assessment. The proposed problem-solving method is based on a symbolic/sub-symbolic line of logical formalisms that have been articulated as an Artificial Neural Network approach to data processing, complemented by an unusual approach to Knowledge Representation and Argumentation that takes into account the data elements entropic states. The model performs well in the thyroid dysfunction assessment with an accuracy ranging between 93.2% and 96.9%.
Sofia Santos, M. Rosário Martins, Henrique Vicente, M. Gabriel Barroca, Fernando Calisto, César Gama, Jorge Ribeiro, Joana Machado, Liliana Ávidos, Nuno Araújo, Almeida Dias, José Neves

Detecting Automatic Patterns of Stroke Through Text Mining

Abstract
Despite the volume increase of electronic data collection in the health area, there is still much medical information that is recorded without any systematic pattern. For instance, besides the structured admission notes format, there are free text fields for clinicians’ patient evaluation observation. Intelligent Decisions Support Systems can benefit from cross-referencing and interpretation of these documents. In the Intensive Care Units, several patients are admitted daily, and several discharge notes are written. To support real-time decision-making and to increase the quality of its process, is crucial to have all relevant patient clinical data available. Since there is no writing pattern followed by all medical doctors, its analysis becomes quite difficult to do. This project aims to make qualitatively and quantitatively analysis of clinical information focusing on the stroke or cerebrovascular accident diagnosis using text analysis tools, namely Natural Language Processing and Text Mining. Our results revealed a set of related words in the clinician’ patient diaries that can reveal patterns.
Miguel Vieira, Filipe Portela, Manuel Filipe Santos

A Preliminary Evaluation of a Computer Vision-Based System to Detect Effects of Aromatherapy During High School Classes via Movement Analysis

Abstract
In this paper we present a pilot study on non-intrusive visual observation and estimation of affective parameter using recorded videos (RGB). We aim at estimating student engagement analyzing upper-body movement comparing two different classroom settings: with the introduction of aromatherapy during the class vs standard lesson. Following previous studies on how aromatherapy can alter movement behaviour, we chose Lavender essential oil. We used computer vision techniques for pose estimation and developed software modules for the extraction of movement features from media data. Data show significant increases in overall velocity and acceleration when the participants are exposed to the aromatherapy condition. Significant decreases in neck flexion angle has been also observed, that shows students had a straighter head posture (i.e. sitting up straighter). No significant differences were observed for the overall kinetic energy of the joints and spinal extension.
Ksenia Kolykhalova, David O’Sullivan, Stefano Piana, Hyungsook Kim, Yonghyun Park, Antonio Camurri

Computational Inference Applied to Social Profiling

Frontmatter

Virtual Agents for Professional Social Skills Training: An Overview of the State-of-the-Art

Abstract
Training of interpersonal communication skills is typically done using role play, by practising relevant scenarios with the help of professional actors. However, as a result of the rapid developments in human-computer interaction, there has been an increasing interest in the use of computers for training of social and communicative skills. This type of training offers opportunities to complement traditional training methods with a novel paradigm that is more scalable and cost-effective. The main idea of such applications is that of a simulated conversation between a human trainee and a virtual agent. By developing the system in such a way that the communicative behaviour of the human has a direct impact on the behaviour of the virtual agent, an interactive learning experience is created. In this article, we review the current state-of-the-art in virtual agents for social skills training. We provide an overview of existing applications, and discuss various properties of these applications.
Kim Bosman, Tibor Bosse, Daniel Formolo

A Machine Learning Approach to Detect Violent Behaviour from Video

Abstract
The automatic classification of violent actions performed by two or more persons is an important task for both societal and scientific purposes. In this paper, we propose a machine learning approach, based a Support Vector Machine (SVM), to detect if a human action, captured on a video, is or not violent. Using a pose estimation algorithm, we focus mostly on feature engineering, to generate the SVM inputs. In particular, we hand-engineered a set of input features based on keypoints (angles, velocity and contact detection) and used them, under distinct combinations, to study their effect on violent behavior recognition from video. Overall, an excellent classification was achieved by the best performing SVM model, which used keypoints, angles and contact features computed over a 60 frame image input range.
David Nova, André Ferreira, Paulo Cortez

Detection and Prevention of Bullying on Online Social Networks: The Combination of Textual, Visual and Cognitive

Abstract
The adoption of online social platforms as a common space for the virtualisation of identities is also correlated with the replication of real-world social hazards in the virtual world. Bullying, or cyberbullying, is a very common practice among people nowadays, becoming much more present due to the increase of online time, especially in online social networks, and having more serious consequences among younger audiences. Related work includes the analysis and classification of textual characteristics that can be indicative of a bullying situation and even a visual analysis approach through the adoption of image recognition techniques. While agreeing that the combination of textual and visual analysis can help the identification of bullying practice, or the identification of bullies, we also believe that a part is missing. In this work, we propose a combination of textual and visual classification techniques, associated with a cognitive aspect that can help to identify possible bullies. Based on a previous model definition for a virtual social sensor, we propose the analysis of textual content present on online social networks, check the presence of people in multimedia content, and identification of the stakeholders on a possible bullying situation by identifying cognitive characteristics and similarities on the behaviours of possible bullies and/or victims. This identification of possible bullying scenario can help to address them before they occur or reach unmanageable proportions.
Carlos Silva, Ricardo Santos, Ricardo Barbosa

Exploring Novel Methodology for Classifying Cognitive Workload

Abstract
This paper describes our work in extracting useful cognitive load classification information from a relatively simple and non-invasive physiological measurement technique, with application in a range of Human Factors and Human-Computer Interaction contexts. We employ novel methodologies, including signal processing, machine learning and genetic algorithms, to classify Galvanic Skin Response/Electrodermal Activity (GSR/EDA) signals during performance of a customised game task (UAV Defender) in high- and low-workload conditions. Our results reveal that Support Vector Machine Linear was the most successful technique for classifying the level of cognitive load that an operator is undergoing during easy, medium, and difficult operation conditions. This methodology has the advantage of applicability in critical task situations, where other cognitive load measurement methodologies are problematic due to sampling delay (e.g. questionnaires), or difficulty of implementation (e.g. other psych-physiological measures). A proposed cognitive load classification pipeline for real-time implementation and its use in human factors contexts is discussed.
Seth Siriya, Martin Lochner, Andreas Duenser, Ronnie Taib

Virtual Environments, Entertainment and Games

Frontmatter

Scene Reconstruction for Storytelling in 360 Videos

Abstract
In immersive and interactive contents like 360-degrees videos the user has the control of the camera, which poses a challenge to the content producer since the user may look to where he wants. This paper presents the concept and first steps towards the development of a framework that provides a workflow for storytelling in 360-degrees videos. With the proposed framework it will be possible to connect a sound to a source and taking advantage of binaural audio it will help to redirect the user attention to where the content producer wants. To present this kind of audio, the scenario must be mapped/reconstructed so as to understand how the objects contained in it interfere with the sound waves propagation. The proposed system is capable of reconstructing the scenario from a stereoscopic, still or motion 360-degrees video when provided in an equirectangular projection. The system also incorporates a module that detects and tracks people, mapping their motion from the real world to the 3D world. In this document we describe all the technical decisions and implementations of the system. To the best of our knowledge, this system is the only that has shown the capability to reconstruct scenarios in a large variety of 360 footage and allows for the creation of binaural audio from that reconstruction.
Gonçalo Pinheiro, Nelson Alves, Luis Magalhães, Luís Agrellos, Miguel Guevara

User Behaviour Analysis and Personalized TV Content Recommendation

Abstract
Nowadays, there are many channels and television (TV) programs available, and when the viewer is confronted with this amount of information has difficulty in deciding which wants to see. However, there are moments of the day that viewers see always the same channels or programs, that is, viewers have TV content consumption habits. The aim of this paper was to develop a recommendation system that to be able to recommend TV content considering the viewer profile, time and weekday.
For the development of this paper, were used Design Science Research (DSR) and Cross Industry Standard Process for Data Mining (CRISP-DM) methodologies. For the development of the recommendation model, two approaches were considered: a deterministic approach and a Machine Learning (ML) approach. In the ML approach, K-means algorithm was used to be possible to combine STBs with similar profiles. In the deterministic approach the behaviors of the viewers are adjusted to a profile that will allow you to identify the content you prefer. Here, recommendation system analyses viewer preferences by hour and weekday, allowing customization of the system, considering your historic, recommending what he wants to see at certain time and weekday.
ML approach was not used due to amount of data extracted and computational resources available. However, through deterministic methods it was possible to develop a TV content recommendation model considering the viewer profile, the weekday and the hour. Thus, with the results it was possible to understand which viewer profiles where the ML can be used.
Ana Carolina Ribeiro, Rui Frazão, Jorge Oliveira e Sá

Virtual and Augmented Reality Interfaces in Shared Game Environments: A Novel Approach

Abstract
Augmented Reality (AR) and Virtual Reality (VR) have been usually addressed as two separated worlds and recent studies try to address the problem of merging the AR and VR applications into a single “environment”, providing a system that relies on both paradigms. The constant release of new hardware interfaces for both wearable AR and Immersive VR opens up new possibilities for the gaming area and many others. However, even if there are researches that explore the usage of AR and VR in the same application, videogames are deployed for one environment or the other depending on their strengths and flaws and the type of experience they can offer to the player, in order to exalt the peculiarities of the chosen medium. A novel approach would be to provide a multiplayer system that enables the users to play the same (or similar) experience through either an AR or VR interface: the player could freely choose the interface, based on several factors such as hardware availability, environment, physical limitations or personal preferences. In this paper, a preliminary study on a multiplayer game system for both AR and VR interfaces is proposed. A chess game experience is provided and a comparison through a System Usability Scale (SUS) questionnaire allowed to establish if both interfaces provided a satisfactory game experience and to highlight both hardware limitations and further interface enhancements.
Francesco De Pace, Federico Manuri, Andrea Sanna, Davide Zappia

Microbial Integration on Player Experience of Hybrid Bio-digital Games

Abstract
Hybrid bio-digital games physically integrate non-human, living organisms into computer gaming hardware and software. Whilst such type of game can add novelty value, the positive impact of the added biological element on player experience has not yet been verified quantitatively. We conducted a study involving two groups of 20 participants, to compare player experiences of two versions of a video game called Mould Rush, which relies on the growth patterns of micro-organisms commonly known as ‘mould’. Results from self-reporting Game Experience Questionnaire (GEQ) showed that the group who played the version of Mould Rush that integrated real mould, had produced significantly higher mean GEQ scores (p < .001) on the following dimensions: Positive Affect; Sensory and Imaginative Immersion; Positive Experience; and Returning to Reality. Furthermore, results from participant interviews indicated that the slowness of mould growth was enjoyed by those who played real-mould-integrated version of Mould Rush. Contrastingly, the slowness was perceived as a negative feature for those who played the game without integrated mould. We discuss the implications and limitations of all of our findings.
Raphael Kim, Siobhan Thomas, Roland van Dierendonck, Antonios Kaniadakis, Stefan Poslad

Backmatter

Weitere Informationen