Skip to main content
main-content

Über dieses Buch

This book constitutes the thoroughly refereed proceedings of the 22nd International Conference on User Modeling, Adaption and Personalization, held in Aalborg, Denmark, in July 2014. The 23 long and 19 short papers of the research paper track were carefully reviewed and selected from 146 submissions. The papers cover the following topics: large scale personalization, adaptation and recommendation; Personalization for individuals, groups and populations; modeling individuals, groups and communities; Web dynamics and personalization; adaptive web-based systems; context awareness; social recommendations; user experience; user awareness and control; Affective aspects; UMAP underpinning by psychology models; privacy; perceived security and trust; behavior change and persuasion.

Inhaltsverzeichnis

Frontmatter

Long Presentations

Modelling Long Term Goals

Goals have long been recognised as important in user modelling and personalisation. Surprisingly, little research has dealt with user model representations for people’s long term goals. This paper describes our theoretically-grounded design of user models for long term goals; notably, the theory points to the critical role of the user interface to this Goal Model, to enable people to set, monitor and refine their models over the long term. We report on a multi-study evaluation of the tightly coupled user model representation and Goal Interface, based on a preliminary lab study (16 participants), and a field trial (14 participants), starting with the lab study and then the in-the-wild use and the questionnaires. This provides multiple sources of evidence to validate the usefulness of our Goal Model to represent three long term health-related goals. It shows that the Goal Interface is usable and aids people in setting their long term goals.

Debjanee Barua, Judy Kay, Bob Kummerfeld, Cécile Paris

A Personalization Method Based on Human Factors for Improving Usability of User Authentication Tasks

Aiming to ensure safety of operation to application providers and improve the usability of human computer interactions during authentication, this paper proposes a two-step personalization approach of user authentication tasks based on individual differences in cognitive processing as follows: i) recommend a textual or graphical user authentication mechanism based on the users’ cognitive styles of processing textual and graphical information, and ii) recommend a standard or enhanced authentication key strength policy considering the users’ cognitive processing abilities. The proposed approach has been applied in a four month ecological valid user study in which 137 participants interacted with a personalized user authentication mechanism and policy based on their cognitive characteristics. Initial results indicate that personalizing the user authentication task based on human cognitive factors could provide a viable solution for balancing the security and usability of authentication mechanisms at the benefit of both application providers and end-users.

Marios Belk, Panagiotis Germanakos, Christos Fidas, George Samaras

The Magic Barrier of Recommender Systems – No Magic, Just Ratings

Recommender Systems need to deal with different types of users who represent their preferences in various ways. This difference in user behaviour has a deep impact on the final performance of the recommender system, where some users may receive either better or worse recommendations depending, mostly, on the quantity and the quality of the information the system knows about the user. Specifically, the inconsistencies of the user impose a lower bound on the error the system may achieve when predicting ratings for that particular user.

In this work, we analyse how the consistency of user ratings (

coherence

) may predict the performance of recommendation methods. More specifically, our results show that our definition of coherence is correlated with the so-called

magic barrier

of recommender systems, and thus, it could be used to discriminate between easy users (those with a low magic barrier) and difficult ones (those with a high magic barrier). We report experiments where the rating prediction error for the more coherent users is lower than that of the less coherent ones. We further validate these results by using a public dataset, where the magic barrier is not available, in which we obtain similar performance improvements.

Alejandro Bellogín, Alan Said, Arjen P. de Vries

Toward Fully Automated Person-Independent Detection of Mind Wandering

Mind wandering is a ubiquitous phenomenon where attention involuntary shifts from task-related processing to task-unrelated thoughts. Mind wandering has negative effects on performance, hence, intelligent interfaces that detect mind wandering can intervene to restore attention to the current task. We investigated the use of eye gaze and contextual cues to automatically detect mind wandering during reading with a computer interface. Participants were pseudo-randomly probed to report mind wandering instances while an eye tracker recorded their gaze during a computerized reading task. Supervised machine learning techniques detected positive responses to mind wandering probes from gaze and context features in a user-independent fashion. Mind wandering was predicted with an accuracy of 72% (expected accuracy by chance was 62%) when probed at the end of a page and an accuracy of 59% (chance was 50%) when probed in the midst of reading a page. Possible improvements to the detectors and applications are discussed.

Robert Bixler, Sidney D’Mello

Hybrid Recommendation in Heterogeneous Networks

The social web is characterized by a wide variety of connections between individuals and entities. A challenge for recommendation is to represent and synthesize all useful aspects of a user’s profile. Typically, researchers focus on a limited set of relations (for example, person to person ties for user recommendation or annotations in social tagging recommendation).

In this paper, we present a general approach to recommendation in heterogeneous networks that can incorporate multiple relations in a weighted hybrid. A key feature of this approach is the use of the

metapath

, an abstraction of a class of paths in a network in which edges of different types are traversed in a particular order. A user profile is therefore a composite of multiple metapath relations. Compared to prior work with shorter metapaths, we show that a hybrid composed of components using longer metapaths yields improvements in recommendation diversity without loss of accuracy on social tagging datasets.

Robin Burke, Fatemeh Vahedian, Bamshad Mobasher

Recommendation Based on Contextual Opinions

Context has been recognized as an important factor in constructing personalized recommender systems. However, most context-aware recommendation techniques mainly aim at exploiting item-level contextual information for modeling users’ preferences, while few works attempt to detect more fine-grained aspect-level contextual preferences. Therefore, in this article, we propose a contextual recommendation algorithm based on user-generated reviews, from where users’ context-dependent preferences are inferred through different contextual weighting strategies. The context-dependent preferences are further combined with users’ context-independent preferences for performing recommendation. The empirical results on two real-life datasets demonstrate that our method is capable of capturing users’ contextual preferences and achieving better recommendation accuracy than the related works.

Guanliang Chen, Li Chen

User Partitioning Hybrid for Tag Recommendation

Tag recommendation is a fundamental service in today’s social annotation systems, assisting users as they collect and annotate resources. Our previous work has demonstrated the strengths of a linear weighted hybrid, which weights and combines the results of simple components into a final recommendation. However, these previous efforts treated each user the same. In this work, we extend our approach by automatically discovering partitions of users. The user partitioning hybrid learns a different set of weights for these user partitions. Our rigorous experimental results show a marked improvement. Moreover, analysis of the partitions within a dataset offers interesting insights into how users interact with social annotations systems.

Jonathan Gemmell, Bamshad Mobasher, Robin Burke

Predicting User Locations and Trajectories

Location-based services usually recommend new locations based on the user’s current location or a given destination. However, human mobility involves to a large extent routine behavior and visits to already visited locations. In this paper, we show how daily and weekly routines can be modeled with basic prediction techniques. We compare the methods based on their performance, entropy and correlation measures. Further, we discuss how location prediction for everyday activities can be used for personalization techniques, such as timely or delayed recommendations.

Eelco Herder, Patrick Siehndel, Ricardo Kawase

A Two-Stage Item Recommendation Method Using Probabilistic Ranking with Reconstructed Tensor Model

In a tag-based recommender system, the multi-dimensional <user, item, tag> correlation should be modeled effectively for finding quality recommendations. Recently, few researchers have used tensor models in recommendation to represent and analyze latent relationships inherent in multi-dimensions data. A common approach is to build the tensor model, decompose it and, then, directly use the reconstructed tensor to generate the recommendation based on the maximum values of tensor elements. In order to improve the accuracy and scalability, we propose an implementation of the

n

-mode block-striped (matrix) product for scalable tensor reconstruction and probabilistically ranking the candidate items generated from the reconstructed tensor. With testing on real-world datasets, we demonstrate that the proposed method outperforms the benchmarking methods in terms of recommendation accuracy and scalability.

Noor Ifada, Richi Nayak

Time-Sensitive User Profile for Optimizing Search Personlization

Thanks to social Web services, Web search engines have the opportunity to afford personalized search results that better fit the user’s information needs and interests. To achieve this goal, many personalized search approaches explore user’s social Web interactions to extract his preferences and interests, and use them to model his profile. In our approach, the user profile is implicitly represented as a vector of weighted terms which correspond to the user’s interests extracted from his online social activities. As the user interests may change over time, we propose to weight profiles terms not only according to the content of these activities but also by considering the freshness. More precisely, the weights are adjusted with a temporal feature. In order to evaluate our approach, we model the user profile according to data collected from Twitter. Then, we rerank initial search results accurately to the user profile. Moreover, we proved the significance of adding a temporal feature by comparing our method with baselines models that does not consider the user profile dynamics.

Ameni Kacem, Mohand Boughanem, Rim Faiz

A Computational Model for Mood Recognition

In an ambience designed to adapt to the user’s affective state, pervasive technology should be able to decipher unobtrusively his underlying mood. Great effort has been devoted to automatic punctual emotion recognition from visual input. Conversely, little has been done to recognize longer-lasting affective states, such as mood. Taking for granted the effectiveness of emotion recognition algorithms, we go one step further and propose a model for estimating the mood of an affective episode from a known sequence of punctual emotions. To validate our model experimentally, we rely on the human annotations of the well-established HUMAINE database. Our analysis indicates that we can approximate fairly accurately the human process of summarizing the emotional content of a video in a mood estimation. A moving average function with exponential discount of the past emotions achieves mood prediction accuracy above 60%.

Christina Katsimerou, Judith A. Redi, Ingrid Heynderickx

Privacy and User Trust in Context-Aware Systems

Context-aware systems (CAS) that collect personal information are a general trend. This leads to several privacy considerations, which we outline in this paper. We present as use-case the SWELL system, which collects information from various contextual sensors to provide support for well-being at work. We address privacy from two perspectives: 1) the development point of view, in which we describe how to apply ‘privacy by design’, and 2) a user study, in which we found that providing detailed information on data collection and privacy by design had a positive effect on trust in our CAS. We also found that the attitude towards using our CAS was related to personal motivation, and not related to perceived privacy and trust in our system. This may stress the importance of implementing privacy by design to protect the privacy of the user.

Saskia Koldijk, Gijs Koot, Mark Neerincx, Wessel Kraaij

Hoeffding-CF: Neighbourhood-Based Recommendations on Reliably Similar Users

Neighbourhood-based collaborative filtering recommenders exploit the common ratings among users to identify a user’s most similar neighbours. It is known that decisions made on a naive computation of user similarity are unreliable, because the number of co-ratings varies strongly among users. In this paper, we formalize the notion of

reliable similarity

between two users and propose a method that constructs a user’s neighbourhood by selecting only those users that are reliably similar to her. Our method combines a statistical test and the notion of a

baseline user

. We report our results on typical benchmark datasets.

Pawel Matuszyk, Myra Spiliopoulou

Toward a Personalized Approach for Combining Document Relevance Estimates

A large body of work in the information retrieval area has highlighted that relevance is a complex and a challenging concept. The underlying complexity appears mainly from the fact that relevance is estimated by considering multiple dimensions and that most of them are subjective since they are user-dependent. While the most used dimension is topicality, recent works risen particularly from personalized information retrieval have shown that personal preferences and contextual factors such as interests, location and task peculiarities have to be jointly considered in order to enhance the computation of document relevance. To answer this challenge, the commonly used approaches are based on linear combination schemes that rely basically on the non-realistic independency property of the relevance dimensions. In this paper, we propose a novel fuzzy-based document relevance aggregation operator able to capture the user’s importance of relevance dimensions as well as information about their interaction. Our approach is empirically evaluated and relies on the standard TREC contextual suggestion dataset involving 635 users and 50 contexts. The results highlight that accounting jointly for individual differences toward relevance dimension importance as well as their interaction introduces a significant improvement in the retrieval performance.

Bilel Moulahi, Lynda Tamine, Sadok Ben Yahia

Adaptive Support versus Alternating Worked Examples and Tutored Problems: Which Leads to Better Learning?

Learning from worked examples has been shown to be superior to unsupported problem solving when first learning in a new domain. Several studies have found that learning from examples results in faster learning in comparison to tutored problem solving in Intelligent Tutoring Systems. We present a study that compares a fixed sequence of alternating worked examples and tutored problem solving with a strategy that adaptively decides how much assistance the student needs. The adaptive strategy determines the type of task (a worked example, a faded example or a problem to be solved) based on how much assistance the student received in the previous problem. The results show that students in the adaptive condition learnt significantly more than their peers who were presented with a fixed sequence of worked examples and problems.

Amir Shareghi Najar, Antonija Mitrovic, Bruce M. McLaren

Te,Te,Hi,Hi: Eye Gaze Sequence Analysis for Informing User-Adaptive Information Visualizations

Information visualization systems have traditionally followed a one-size-fits-all paradigm with respect to their users, i.e., their design is seldom personalized to the specific characteristics of users (e.g. perceptual abilities) or their tasks (e.g. task difficulty). In view of creating information visualization systems that can

adapt

to each individual user and task, this paper provides an analysis of user eye gaze data aimed at identifying behavioral patterns that are specific to certain user and task groups. In particular, the paper leverages the sequential nature of user eye gaze patterns through

differential sequence mining

, and successfully identifies a number of pattern differences that could be leveraged by adaptive information visualization systems in order to automatically identify (and consequently adapt to) different user and task characteristics.

Ben Steichen, Michael M. A. Wu, Dereck Toker, Cristina Conati, Giuseppe Carenini

Text-Based User-kNN: Measuring User Similarity Based on Text Reviews

This article reports on a modification of the user-kNN algorithm that measures the similarity between users based on the similarity of text reviews, instead of ratings. We investigate the performance of text semantic similarity measures and we evaluate our text-based user-kNN approach by comparing it to a range of ratings-based approaches in a ratings prediction task. We do so by using datasets from two different domains: movies from RottenTomatoes and Audio CDs from Amazon Products. Our results show that the text-based userkNN algorithm performs significantly better than the ratings-based approaches in terms of accuracy measured using RMSE.

Maria Terzi, Matthew Rowe, Maria-Angela Ferrario, Jon Whittle

Using DBpedia as a Knowledge Source for Culture-Related User Modelling Questionnaires

In the culture domain, questionnaires are often used to obtain profiles of users for adaptation. Creating questionnaires requires subject matter experts and diverse content, and often does not scale to a variety of cultures and situations. This paper presents a novel approach that is inspired by crowdwisdom and takes advantage of freely available structured linked data. It presents a mechanism for extracting culturally-related facts from DBpedia, utilised as a knowledge source in an interactive user modelling system. A user study, which examines the system usability and the accuracy of the resulting user model, demonstrates the potential of using DBpedia for generating culture-related user modelling questionnaires and points at issues for further investigation.

Dhavalkumar Thakker, Lydia Lau, Ronald Denaux, Vania Dimitrova, Paul Brna, Christina Steiner

Eye Tracking to Understand User Differences in Visualization Processing with Highlighting Interventions

We present an analysis of user gaze data to understand if and how user characteristics impact visual processing of bar charts in the presence of different highlighting interventions designed to facilitate visualization usage. We then link these results to task performance in order to provide insights on how to design user-adaptive information visualization systems. Our results show how the least effective intervention manifests itself as a distractor based on gaze patterns. The results also identify specific visualization regions that cause poor task performance in users with low values of certain cognitive measures, and should therefore be the target of personalized visualization support.

Dereck Toker, Cristina Conati

Evil Twins: Modeling Power Users in Attacks on Recommender Systems

Attacks on Collaborative Filtering Recommender Systems (RS) can bias recommendations, potentially causing users to distrust results and the overall system. Attackers constantly innovate, and understanding the implications of novel attack vectors on system robustness is important for designers and operators. Foundational research on attacks in RSs studied attack user profiles based on straightforward models such as random or average ratings data. We are studying a novel category of attack based explicitly on measures of influence, in particular the potential impact of high-influence

power users

. This paper describes our approach to generate synthetic attack profiles that emulate influence characteristics of real power users, and it studies the impact of attack vectors that use synthetic power user profiles. We evaluate both the quality of synthetic power user profiles and the effectiveness of the attack, on both user-based and matrix-factorization-based recommender systems. Results show that synthetic user profiles that model real power users are an effective way of attacking collaborative recommender systems.

David C. Wilson, Carlos E. Seminario

Personality Profiling from Text: Introducing Part-of-Speech N-Grams

A support vector machine is trained to classify the Five Factor personality of writers of free text. Writers are classified for each of the five personality dimensions as high/low with the mean personality score for each dimension used for the dividing point. Writers are also separately classified as high/medium/low with division points at one standard deviation above and below mean. The two-class average accuracy using 5-fold cross validation of 80.6% is much better than the baseline (pick most likely class) accuracy of 50%, but the 3-class accuracy is only slightly better (7.4%) than baseline because most writers fall into the medium class due to the normal distribution of personality values. Features include bag of words, essay length, word sentiment, negation count and part-of-speech

n

-grams. The consistently positive contribution of POS

n

-grams (averaging 4.8% and 5.8% for the 2/3 class cases) is analyzed in detail. The information gain for the most predictive features for each of the five personality dimensions are presented and discussed.

William R. Wright, David N. Chin

Collaborative Compound Critiquing

Critiquing-based recommender systems offer users a conversational paradigm to provide their feedback, named

critiques

, during the process of viewing the current recommendation. In this way, the system is able to learn and adapt to the users’ preferences more precisely so that better recommendation could be returned in the subsequent iteration. Moreover, recent works on experience-based critiquing have suggested the power of improving the recommendation efficiency by making use of relevant sessions from other users’ histories so as to save the active user’s interaction effort. In this paper, we present a novel approach to processing the history data and apply it to the compound critiquing system. Specifically, we develop a history-aware collaborative compound critiquing method based on preference-based compound critique generation and graph-based similar session identification. Through experiments on two data sets, we validate the outperforming efficiency of our proposed method in comparison to the other experience-based methods. In addition, we verify that incorporating user histories into compound critiquing system can be significantly more effective than the corresponding unit critiquing system.

Haoran Xie, Li Chen, Feng Wang

Sparrows and Owls: Characterisation of Expert Behaviour in StackOverflow

Question Answering platforms are becoming an important repository of crowd-generated knowledge. In these systems a relatively small subset of users is responsible for the majority of the contributions, and ultimately, for the success of the Q/A system itself. However, due to built-in incentivization mechanisms, standard expert identification methods often misclassify very active users for knowledgable ones, and misjudge activeness for expertise. This paper contributes a novel metric for expert identification, which provides a better characterisation of users’ expertise by focusing on the quality of their contributions. We identify two classes of relevant users, namely

sparrows

and

owls

, and we describe several behavioural properties in the context of the

StackOverflow

Q/A system. Our results contribute new insights to the study of expert behaviour in Q/A platforms, that are relevant to a variety of contexts and applications.

Jie Yang, Ke Tao, Alessandro Bozzon, Geert-Jan Houben

Short Presentations

Generalizability of Goal Recognition Models in Narrative-Centered Learning Environments

Recent years have seen growing interest in automated goal recognition. In user-adaptive systems, goal recognition is the problem of recognizing a user’s goals by observing the actions the user performs. Models of goal recognition can support student learning in intelligent tutoring systems, enhance communication efficiency in dialogue systems, or dynamically adapt software to users’ interests. In this paper, we describe an approach to goal recognition that leverages Markov Logic Networks (MLNs)—a machine learning framework that combines probabilistic inference with first-order logical reasoning—to encode relations between problem-solving goals and

discovery events,

domain-specific representations of user progress in narrative-centered learning environments. We investigate the impact of discovery event representations on goal recognition accuracy and efficiency. We also investigate the generalizability of discovery event-based goal recognition models across two corpora from students interacting with two distinct narrative-centered learning environments. Empirical results indicate that discovery event-based models outperform previous state-of-the-art approaches on both corpora.

Alok Baikadi, Jonathan Rowe, Bradford Mott, James Lester

Extending Log-Based Affect Detection to a Multi-User Virtual Environment for Science

The application of educational data mining (EDM) techniques to interactive learning software is increasingly being used to broaden the range of constructs typically incorporated in student models, moving from traditional assessment of student knowledge to the assessment of engagement, affect, strategy, and metacognition. Researchers are also broadening the range of environments within which these constructs are assessed. In this study, we develop sensor-free affect detection for EcoMUVE, an immersive multi-user virtual environment that teaches middle-school students about casualty in ecosystems. In this study, models were constructed for five different educationally-relevant affective states (boredom, confusion, delight, engaged concentration, and frustration). Such models allow us to examine the behaviors most closely associated with particular affective states, paving the way for the design of adaptive personalization to improve engagement and learning.

Ryan S. Baker, Jaclyn Ocumpaugh, Sujith M. Gowda, Amy M. Kamarainen, Shari J. Metcalf

Utilizing Mind-Maps for Information Retrieval and User Modelling

Mind-maps have been widely neglected by the information retrieval (IR) community. However, there are an estimated two million active mind-map users, who create 5 million mind-maps every year, of which a total of 300,000 is publicly available. We believe this to be a rich source for information retrieval applications, and present eight ideas on how mind-maps could be utilized by them. For instance, mind-maps could be utilized to generate user models for recommender systems or expert search, or to calculate relatedness of web-pages that are linked in mind-maps. We evaluated the feasibility of the eight ideas, based on estimates of the number of available mind-maps, an analysis of the content of mind-maps, and an evaluation of the users’ acceptance of the ideas. We concluded that user modelling is the most promising application with respect to mind-maps. A user modelling prototype – a recommender system for the users of our mind-mapping software

Docear

– was implemented, and evaluated. Depending on the applied user modelling approaches, the effectiveness, i.e. click-through rate on recommendations, varied between 0.28% and 6.24%. This indicates that mind-map based user modelling is promising, but not trivial, and that further research is required to increase effectiveness.

Joeran Beel, Stefan Langer, Marcel Genzmehr, Bela Gipp

iSCUR: Interest and Sentiment-Based Community Detection for User Recommendation on Twitter

The increasing popularity of social networks has encouraged a large number of significant research works on community detection and user recommendation. The idea behind this work is that taking into account peculiar users’ attitudes (i.e., sentiments, opinions or ways of thinking) toward their own interests can bring benefits in performing such tasks. In this paper we describe (i) a novel method to infer sentiment-based communities without the requirement of obtaining the whole social structure, and (ii) a community-based approach to user recommendation. We take advantage of the

SVO (sentiment-volume-objectivity)

user profiling and the Tanimoto similarity to evaluate user similarity for each topic. Afterwards we employ a clustering algorithm based on modularity optimization to find densely connected users and the Adamic-Adar tie strength to finally suggest the most relevant users to follow. Preliminary experimental results on Twitter reveal the benefits of our approach compared to some state-of-the-art user recommendation techniques.

Davide Feltoni Gurini, Fabio Gasparetti, Alessandro Micarelli, Giuseppe Sansonetti

Towards Identifying Contextual Factors on Parking Lot Decisions

The relevance of contextual factors that adapt in-car recommendations to the driver’s current situation is not yet fully understood. This paper presents a field study that has been conducted in order to identify relevant contextual factors of in-car parking lot recommender systems. Surprisingly, most contextual factors examined, i.e., weather, luggage, and traffic conditions, did not have a significant effect on the parking lot decision in the conducted field study. Only the urgency of the trip and the willingness to walk have significant effects on the decision outcome. Therefore, automobile manufacturers should focus on understanding the relevance of different contextual factors when developing user models for in-car recommender systems.

Klaus Goffart, Michael Schermann, Christopher Kohl, Jörg Preißinger, Helmut Krcmar

Trust-Based Decision-Making for Energy-Aware Device Management

Smart energy systems are able to support users in saving energy by controlling devices, such as lights or displays, depending on context information, such as the brightness in a room or the presence of users. However, proactive decisions should also match the users’ preferences to maintain users’ trust in the system. Wrong decisions could negatively influence users’ acceptance of a system and at worst could make them abandon the system. In this paper, a trust-based model, called User Trust Model (UTM), for automatic decision-making is proposed, which is based on Bayesian Networks. The UTM’s construction, the initialization with empirical data gathered in an online survey, and its integration in an office setting are described. Furthermore, the results of a user study investigating users’ experience and acceptance are presented.

Stephan Hammer, Michael Wißner, Elisabeth André

Doing More with Less: Student Modeling and Performance Prediction with Reduced Content Models

When modeling student knowledge and predicting student performance, adaptive educational systems frequently rely on content models that connect learning content (i.e., problems) with its underlying domain knowledge (i.e., knowledge components, KCs) required to complete it. In some domains, such as programming, the number of KCs associated with advanced learning contents is quite large. It complicates modeling due to increasing noise and decreases efficiency. We argue that the efficiency of modeling and prediction in such domains could be improved without the loss of quality by reducing problems content models to a subset of most important KCs. To prove this hypothesis, we evaluate several KC reduction methods varying reduction size by assessing the prediction performance of Knowledge Tracing and Performance Factor Analysis. The results show that the predictive performance using reduced content models can be significantly better than using original one, with extra benefits of reducing time and space.

Yun Huang, Yanbo Xu, Peter Brusilovsky

The Role of Adaptive Elements in Web-Based Surveillance System User Interfaces

In this paper we present an analysis of improvements to a web-based Graphical User Interface (GUI) for health surveillance systems. Such systems are designed to provide means to detect and suggest outbreaks and corresponding information about them from both formal (e.g., hospital reports) and informal (e.g., news sites) sources. However, despite the availability of different such systems, few studies have been carried out to discuss the elements of the system’s GUI and how it can support users in their tasks. To this end, we investigate techniques for adapting, structuring and browsing information in an intuitive and friendly way to the user, focusing on a transition from a static to a dynamic adapted web experience. We conduct a case study with health surveillance experts where we present a case for recommendations matching the user’s preferences within a system and discuss improvements to the presented GUI. We discuss improvements in the light of the feedback provided by these users, proposing how adapted elements of a GUI can be used to improve the user experience in a surveillance task.

Ricardo Lage, Peter Dolog, Martin Leginus

Uncovering Latent Knowledge: A Comparison of Two Algorithms

At the beginning of every course, it can be expected that several students have some syllabus knowledge. For efficiency in learning systems, and to combat student frustration and boredom, it is important to quickly uncover this latent knowledge. This enables students to begin new learning immediately. In this paper we compare two algorithms used to achieve this goal, both based on the theory of Knowledge Spaces. Simulated students were created with appropriate answering patterns based on predefined latent knowledge from a subsection of a real course. For each student, both algorithms were applied to compare their efficiency and their accuracy. We examine the trade-off between both sets of outcomes, and conclude with the merits and constraints of each algorithm.

Danny J. Lynch, Colm P. Howlin

Client-Side Hybrid Rating Prediction for Recommendation

The centralized gathering and processing of user information made by traditional recommender systems can lead to user information exposure, violating her privacy. Client-side personalization methods have been created as a mean for avoiding privacy risks. Motivated by limiting the exposure of user private information, we explore the use of a client-side hybrid recommender system placed on the online learning setting. We propose a prediction model based on an ensemble blender of an online matrix factorization CF model and a logistic regression model trained on item metadata with a probabilistic feature inclusion strategy. The final prediction is a blend of the two models on a weighted regret approach. We validate our approach with the Movielens 10M dataset.

Andrés Moreno, Harold Castro, Michel Riveill

Combining Distributional Semantics and Entity Linking for Context-Aware Content-Based Recommendation

The effectiveness of content-based recommendation strategies tremendously depends on the representation formalism adopted to model both items and user profiles. As a consequence, techniques for semantic content representation emerged thanks to their ability to filter out the noise and to face with the issues typical of keyword-based representations. This article presents Contextual eVSM (C-eVSM), a content-based context-aware recommendation framework that adopts a novel semantic representation based on distributional models and entity linking techniques. Our strategy is based on two insights: first, entity linking can identify the most relevant concepts mentioned in the text and can easily map them with structured information sources, easily triggering some inference and reasoning on user preferences, while distributional models can provide a lightweight semantics representation based on term co-occurrences that can bring out latent relationships between concepts by just analying their usage patterns in large corpora of data.

The resulting framework is fully domain-independent and shows better performance than state-of-the-art algorithms in several experimental settings, confirming the validity of content-based approaches and paving the way for several future research directions.

Cataldo Musto, Giovanni Semeraro, Pasquale Lops, Marco de Gemmis

IntelWiki: Recommending Resources to Help Users Contribute to Wikipedia

We describe an approach to facilitating user-generated content within the context of Wikipedia. Our approach, embedded in the

IntelWiki

prototype, aims to make it easier for users to create or enhance the free-form text in Wikipedia articles by: i) recommending potential reference materials, ii) drawing the users’ attention to key aspects of the recommendations, and iii) allowing users to consult the recommended materials in context. A laboratory evaluation with 16 novice Wikipedia editors revealed that, in comparison to the default Wikipedia design, IntelWiki’s approach has positive impacts on editing quantity and quality, and perceived mental load.

Mohammad Noor Nawaz Chowdhury, Andrea Bunt

Balancing Adaptivity and Customisation

In Search of Sustainable Personalisation in Cultural Heritage

Personalisation for cultural heritage aims at delivering to visitors the right stories at the right time. Our endeavour to determine which features to use for adaptation starts from acknowledging what forms of personalisation curators value as most meaningful. Working in collaboration with curators we have explored the different features that must be taken into account: some are related to the content (multiple interpretation layers), others to the context of delivery (where and when), but some are idiosyncratic (“match my mood”, “something that is relevant to my life”). The findings reveal that a sustainable personalization needs to accurately balance: (i) support to curators in customising stories to different visitors; (ii) algorithms for the system to dynamically model aspects of the visit and instantiate the correct behaviour; and (iii) an active role for visitors to choose the type of experience they would like to have today.

Elena Not, Daniela Petrelli

Who’s Afraid of Job Interviews? Definitely a Question for User Modelling

We define job interviews as a domain of interaction that can be modelled automatically in a serious game for job interview skills training. We present four types of studies: (1) field-based human-to-human job interviews, (2) field-based computer-mediated human-to-human interviews, (3) lab-based wizard of oz studies, (4) field-based human-to-agent studies. Together, these highlight pertinent questions for the user modelling field as it expands its scope to applications for social inclusion. The results of the studies show that the interviewees suppress their emotional behaviours and although our system recognises automatically a subset of those behaviours, the modelling of complex mental states in real-world contexts poses a challenge for the state-of-the-art user modelling technologies. This calls for the need to re-examine both the approach to the implementation of the models and/or of their usage for the target contexts.

Kaśka Porayska-Pomsta, Paola Rizzo, Ionut Damian, Tobias Baur, Elisabeth André, Nicolas Sabouret, Hazaël Jones, Keith Anderson, Evi Chryssafidou

Towards Understanding the Nonverbal Signatures of Engagement in Super Mario Bros

In this paper, we present an approach for predicting users’ level of engagement from nonverbal cues within a game environment. We use a data corpus collected from 28 participants (152 minutes of video recording) playing the popular platform game

Super Mario Bros

. The richness of the corpus allows extraction of several visual and facial expression features that were utilised as indicators of players’ affects as captured by players’ self-reports. Neuroevolution preference learning is used to construct accurate models of player experience that approximate the relationship between extracted features and reported engagement. The method is supported by a feature selection technique for choosing the relevant subset of features. Different setup settings were implemented to analyse the impact of the type of the features and the position of the extraction window on the modelling accuracy. The results obtained show that highly accurate models can be constructed (with accuracies up to 96.82%) and that players’ nonverbal behaviour towards the end of the game is the most correlated with engagement. The framework presented is part of a bigger picture where the generated models are utilised to tailor content generation to a player’s particular needs and playing characteristics.

Noor Shaker, Mohammad Shaker

Towards Personalized Multilingual Information Access - Exploring the Browsing and Search Behavior of Multilingual Users

The shift from the originally English-language-dominated web towards a truly global

world wide web

has generated a pressing need to develop novel solutions that address

multilingual user diversity

. In particular, many web users today are polyglots, i.e. they are proficient in more than one language. However, little is known about the browsing and search habits of such users, and even less about how to best assist their multilingual behaviors through appropriate systems and tools. In order to gain a better understanding, this paper presents a survey of 385 polyglot web users, focusing specifically on the relationship between multiple language proficiency and browsing/search language choice. Results from the survey indicate that polyglot users make significant use of multiple languages during their daily browsing and searching, and that contextual factors such as language proficiency, usage purpose, and topic domain have a significant influence on their language choice and frequency. The paper provides a detailed analysis regarding each of these factors, and offers insights about how to support multilingual users through novel

Personalized Multilingual Information

Access

systems.

Ben Steichen, M. Rami Ghorab, Alexander O’Connor, Séamus Lawless, Vincent Wade

Graph-Based Recommendations: Make the Most Out of Social Data

Recommender systems use nowadays more and more data about users and items as part of the recommendation process. The availability of auxiliary data, going beyond the mere user/item data, has the potential to improve recommendations. In this work we examine the contribution of two types of social auxiliary data – namely, tags and friendship links – to the accuracy of a graph-based recommender. We measure the impact of the availability of auxiliary data on the recommendations using features extracted from both the auxiliary and the original data. The evaluation shows that the social auxiliary data improves the accuracy of the recommendations, and that the greatest improvement is achieved when graph features mirroring the nature of the auxiliary data are extracted by the recommender.

Amit Tiroshi, Shlomo Berkovsky, Mohamed Ali Kaafar, David Vallet, Tsvi Kuflik

Fast Incremental Matrix Factorization for Recommendation with Positive-Only Feedback

Traditional Collaborative Filtering algorithms for recommendation are designed for stationary data. Likewise, conventional evaluation methodologies are only applicable in offline experiments, where data and models are static. However, in real world systems, user feedback is continuously being generated, at unpredictable rates. One way to deal with this data stream is to perform online model updates as new data points become available. This requires algorithms able to process data at least as fast as it is generated. One other issue is how to evaluate algorithms in such a streaming data environment. In this paper we introduce a simple but fast incremental Matrix Factorization algorithm for positive-only feedback. We also contribute with a prequential evaluation protocol for recommender systems, suitable for streaming data environments. Using this evaluation methodology, we compare our algorithm with other state-of-the-art proposals. Our experiments reveal that despite its simplicity, our algorithm has competitive accuracy, while being significantly faster.

João Vinagre, Alípio Mário Jorge, João Gama

When the Question is Part of the Answer: Examining the Impact of Emotion Self-reports on Student Emotion

A variety of methodologies have been put forth to assess students’ affective states as they use interactive learning environments (ILEs) and intelligent tutoring systems (ITS), such as classroom observations and subjective coding, self-coding by students after replays, as well as self-reports of student emotion as students are using the learning environment. Still, it is unclear what the disadvantages of each methodology are. In particular, does measuring affect by asking students to self-report alter student affect itself? The following work explores this question of how self-reports themselves can bias affective states, within one particular tutoring system, Wayang Outpost.

Michael Wixon, Ivon Arroyo

Doctoral Consortium

Enhancing Exploratory Information-Seeking through Interaction Modeling

With the explosive growth of information available in the web, locating needed and relevant information remains a difficult task, whether the information is textual or visual. Although information-retrieval algorithms have improved greatly in retrieving relevant information, exploratory information-seeking still remains difficult due to its inherently open-ended and dynamic nature. Modeling the user behavior and predicting dynamically changing information-needs in exploratory search is hard. Over the past decade there has been increasing attention on rich user interfaces, retrieval techniques, and studies of exploratory search. However, existing work does not yet support the dynamic aspects of exploratory search. The objective of this research is to understand how user interaction modeling can be applied to provide better support in exploratory information-seeking.

Kumaripaba M. Athukorala

Hybrid Solution of the Cold-Start Problem in Context-Aware Recommender Systems

A challenge of Context-Aware Recommender Systems (CARSs) is the cold-start problem, i.e., the usual poor recommendation of new items to new users in new contextual situations. In this research, we aim at solving this problem by developing a switching hybrid CARS, which exploits different context-aware recommendation techniques, each of which has its own strengths and weaknesses, and switches between these techniques depending on the current recommendation situation (i.e., new user, new item and/or new context).

Matthias Braunhofer

Improving Mobile Recommendations through Context-Aware User Interaction

Mobile recommender systems provide personalized recommendations to help deal with today’s information overload. However, due to spatial limitations in mobile interfaces and uncertainty of the user’s preferences in the beginning, the improvement of the user experience remains one of the main challenges when designing these systems and has not been investigated thoroughly. This paper describes the aim and progress of the author’s PhD studies on the user interaction, usability and accuracy of mobile recommender systems. The approach aims to combine different user interaction methods with context-awareness to allow user-friendly personalized mobile recommendations.

Béatrice Lamche

Personalized Cultural Heritage Experience Outside the Museum: Connecting the Museum Experience to the Outside World

Museums, as cultural heritage sites, have long been a primary showground for the exploration of new technologies. Recent new directions for research in this field, have concerned themselves with 1) expanding the on-site visit with prior and post experiences, primarily at a desktop computer at home, but not necessarily; 2) expanding the visit from a onetime experience to an experience that may repeat itself multiple times over a lifetime, including the reuse of personal information elicited from experience gained onsite (e.g. a user model) for providing personalized experience at multiple sites. The proposed third new direction for research in this field, the one which is focused on is: examining how to enhance other experiences outside the museum site, based on experiences at the museum site. By doing this one can begin to connect our cultural heritage experiences to our "daily" lives.

Alan J. Wecker

Personality Profiling from Text and Grammar

Personality assessment can be used to predict subjects’ use of products and services, thriving in academic programs, and performance in work environments. To avoid the costs and inconvenience of administering personality questionnaires, researchers have inferred author personality from their writings. Extending such methods will enable marketing, interface adaptation, and a variety of data mining applications. The proposed program of research examines elements of syntax, addressing the following questions: does authors’ usage of English grammatical structures reflect their personalities? What methodology extracts and predicts personality from grammar usage? Key to this approach is the use of locally defined grammatical structures as described by Part of Speech

n

-grams.

William R. Wright

Backmatter

Weitere Informationen