User Modeling, Adaptation, and Personalization

Following Aristotle, "Man is by nature a social animal; an individual who is unsocial naturally and not accidentally is either beneath our notice or more than human." This is more than an abstract philosophical statement if, twenty five centuries after, we observe that people have exactly the same social behavior whether they interact with a computer or with another person. Furthermore, there is evidence that users tend to appreciate more computers displaying social behaviors similar to those they appreciate in people. This body of evidence suggests that there is a gap between current, unsocial, computers and user expectations for social behavior.

Social Signal Processing (SSP) is the new, emerging, domain that aims at making computers as social as their human users by modeling people and groups involved in social interactions. The SSP approach focuses on analysis, understanding and synthesis of social signals, the complex aggregates of nonverbal behavioral cues through which people convey their attitude towards others (including machines) and social environments. The development of SSP technologies will help computers to adapt to users like people adapt to others (i.e. depending on the kind of interaction and social context), and to personalize their interface in terms of characteristics socially desirable for the users.

Search is the main entry point for an ever-increasing range of information, services, communications and entertainment. During the last decade, there have been tremendous advances in the scale of search systems and the diversity of available resources. Yet the methods used to represent searchers’ information needs have changed very little. Search interfaces today look much the same as they did a decade ago. Searchers type a few words into a search box, and the search engine returns a long list of results. When the results fail to satisfy the searcher’s information needs, they try again and again with little support from the search engine. In this talk I describe several efforts to improve easy and effectiveness of search by: modeling searchers’ interests and activities over time, representing non-content attributes of information such as time or genre, and developing interaction techniques that enable searchers to articulate their information needs more effectively.

Adaptive hypermedia and adaptive web research have been reasonable successful in researching personalisation in closed corpus content and to a much lesser extent in open corpus content. From a commercial perspective, web adaptivity has been more focused on adaptive content retrieval rather than adaptive content composition. However, personal use of the web extends far beyond just content, and encompasses many dimensions which need to be addressed concurrently e.g. tasks & activities, cultural preferences, and social interaction etc. We need to consider new directions and dimensions in personalised, adaptive web and how they can be addressed within the same personal experience. In this talk I will investigate key challenges involving integrated open corpus & service personalisation, cultural adaptivity (including multilingual personalisation), dialogue and simulation personalisation and the power of the crowd, which could greatly empower web users of the future. I will also consider emerging approaches to tackle these problems and examine what this might mean to current web based personalisation engines and platforms.

We present a model of user affect to recognize multiple user emotions during interaction with an educational computer game. Our model deals with the high level of uncertainty involved in recognizing a variety of user emotions by probabilistically combining information on both the causes and effects of emotional reactions. In previous work, we presented the performance and limitations of the model when using only causal information. In this paper, we discuss the addition of diagnostic information on user affective valence detected via an EMG sensor, and present an evaluation of the resulting model.

In this paper we describe an Integrated Development System for Instructional Model for E-learning (INDESIME) to create and to maintain instructional models using adaptive technologies and collaborative tools. An authoring tool has also been developed for helping to non-programming users to create Learning Management Systems (LMSs) courses that implement a specific instructional model. Data mining techniques are proposed to evaluate the e-learning courses generated from the model. We have tested the degree of effectiveness of our system using Moodle courses. The courses topics tested are based on the European Computer Driving Licence Foundation catalogue.

In this paper we explore findings from three experiments that use minimally invasive sensors with a web based geometry tutor to create a user model. Minimally invasive sensor technology is mature enough to equip classrooms of up to 25 students with four sensors at the same time while using a computer based intelligent tutoring system. The sensors, which are on each student’s chair, mouse, monitor, and wrist, provide data about posture, movement, grip tension, arousal, and facially expressed mental states. This data may provide adaptive feedback to an intelligent tutoring system based on an individual student’s affective states. The experiments show that when sensor data supplements a user model based on tutor logs, the model reflects a larger percentage of the students’ self-concept than a user model based on the tutor logs alone. The models are further expanded to classify four ranges of emotional self-concept including frustration, interest, confidence, and excitement with over 78% accuracy. The emotional predictions are a first step for intelligent tutor systems to create sensor based personalized feedback for each student in a classroom environment. Bringing sensors to our children’s schools addresses real problems of students’ relationship to mathematics as they are learning the subject.

This paper introduces two independent open learner models (learner models that are accessible to user viewing), which are deployed alongside university courses to facilitate self-assessment skills, planning and independent learning. OLMlets is used in specific courses, while UK-SpecIAL, a modular extension to OLMlets, draws on the OLMlets learner models to display progress towards achieving learning outcomes applicable across courses. User logs demonstrate usage of each system, and questionnaire responses provide insight into the reasons for user trust in the environments.

Long term group work by small teams is a central part of many learning and workplace activities. Widespread group support tools such as wikis, version control systems and issue tracking systems are an invaluable aid for groups. They also have the potential to provide evidence for valuable models of the group activity. This paper describes

Narcissus

, designed as a new way to improve group-work by exploiting evidence from use of such group-work tools, to create a visual presentation of a

group model

. The

Narcissus

models and interfaces were designed to help groups function more effectively. It helps individuals see how well they are contributing to the group. It enables groups to assess contributions relative to plans. And it helps facilitators identify problems. The

Narcissus

interface supports scrutability and control over its models. We report a four part evaluation of

Narcissus

:

individual level

with 23 students;

group level

by 5 groups;

facilitator level

with 5 facilitators; and

fine grained study

with 8 students. Results indicate that all these groups were able to understand and use

Narcissus

and that they considered it effective in modelling the group activity in useful ways. They particularly valued the support for scrutability. Key contributions of this work are the creation of a

scrutable

and user controlled group model to support group work and to provide a new form of navigation interface for a complex groupware site.

Navigating through the ever-changing information space is becoming increasingly difficult. Social navigation support is a technique for guiding users to interesting and relevant information by leveraging the browsing behavior of past users. Effect of social navigation support on users’ information seeking behavior has been studied mostly from conceptual basis or under natural experiments. In the current work, we have designed and conducted a controlled experiment to investigate the effect of social navigation support through a multifaceted method. This paper reports on the design of the study and the result of log data, subjective evaluation, and eye movement data analysis.

Reconciling personalization with privacy has been a continuing interest in the user modeling community. In prior work, we proposed a dynamic privacy-enhancing user modeling framework based on a software product line architecture (PLA). Our system dynamically selects personalization methods during runtime that respect users’ current privacy preferences as well as the prevailing privacy laws and regulations. One major concern about our approach is its performance since dynamic architectural reconfiguration during runtime is usually resource-intensive. In this paper, we describe four implementations of our system that vary two factors, and an in-depth performance evaluation thereof under realistic workload conditions. Our study shows that a customized version performs better than the original PLA implementation, that a multi-level caching mechanism improves both versions, and that the customized version with caching performs best. The average handling time per user session is less than 0.2 seconds for all versions except the original PLA implementation. Overall, our results demonstrate that with a reasonable number of networked hosts in a cloud computing environment, an internationally operating website can use our dynamic PLA-based user modeling approach to personalize their user services, and at the same time respect the individual privacy desires of their users as well as the privacy norms that may apply.

Knowledge about computer users is very beneficial for assisting them, predicting their future actions or detecting masqueraders. In this paper, an approach for creating and recognizing automatically the behavior profile of a user from the commands (s)he types in a command-line interface, is presented.

Specifically, in this research, a computer user behavior is represented as a sequence of UNIX commands. This sequence is transformed into a distribution of relevant subsequences in order to find out a profile that defines its behavior. Then, statistical methods are used for recognizing a user from the commands (s)he types. The experiment results, using 2 different sources of UNIX command data, show that a system based on our approach can efficiently recognize a UNIX user. In addition, a comparison with a HMM-base method is done.

Because a user profile usually changes constantly, we also propose a method to keep up to date the created profiles using an

age

-based mechanism.

We propose a novel approach for constructing statistical preference models for context-aware recommender systems. To do so, one of the most important but difficult problems is acquiring sufficient training data in various contexts/situations. Particularly, some situations require a heavy workload to set them up or to collect subjects under those situations. To avoid this, often a large amount of data in a supposed situation is collected, i.e., a situation where the subject pretends/imagines that he/she is in a specific situation. Although there may be difference between the preference in the real situation and the supposed situation, this has not been considered in existing researches. Here, to study the difference, we collected a certain amount of corresponding data. We asked subjects the same question about preference both in the real and the supposed situation. Then we proposed a new model construction method using a difference model constructed from the correspondence data and showed the effectiveness through the experiments.

In this paper we target the automatic prediction of two personality traits, Extraversion and Locus of Control, in a meeting scenario using visual and acoustic features. We designed our task as a regression one where the goal is to predict the personality traits’ scores obtained by the meeting participants. Support Vector Regression is applied to thin slices of behavior, in the form of 1-minute sequences.

Recent sensor technologies have enabled the capture of users’ behavior data. Given the large amount of data currently available from sensor-equipped environments, it is important to attempt characterization of the sensor data for automatically modeling users in a ubiquitous and mobile computing environment. As described herein, we propose a method that predicts a customer model using features based on customers’ behavior in a shop. We capture the customers’ behavior using various sensors in the form of the time duration and the sequence between blocks in the shop. Based on behavior data from the sensors, we design features that characterize the behavior pattern of a customer in the shop. We employ those features using a machine learning approach to predict customer attributes such as age, gender, occupation, and interest. Our results show that our designed behavior-based features perform with

F

-values of 70–90% for prediction. We also discuss the potential applications of our method in user modeling.

We investigate the utility of an eye tracker for providing information on users’ affect and reasoning. To do so, we conducted a user study, results from which show that users’ pupillary responses differ significantly between positive and negative affective states. As far as reasoning is concerned, while our analysis shows that larger pupil size is associated with more constructive reasoning events, it also suggests that to disambiguate between different kinds of reasoning, additional information may be needed. Our results show that pupillary response is a promising non-invasive avenue for increasing user model bandwidth.

The description of the user-system interaction plays a crucial role in adaptive interactive systems, since the adaptations depend on this description. User actions in interactive systems can be described as a sequence of events, which are created by input through input devices as well as by the system as reactions to these inputs. An interactive system can observe these events and thus extract information about the user’s behavior. This paper presents a two-step approach for describing user behavior from sequences of basic events. First, user actions are recognized in the sequence of interaction events by means of previously trained probabilistic automata. Second, a task model describes the higher-level user activity as a hierarchical composition of these actions. Different kinds of adaptive support can be derived from this description of user behavior, such as recommending next interaction steps to the user.

Words mean different things to different people, and capturing these differences is often a subtle art. These differences are often “a matter of perspective”. Perspective can be taken to be the set of beliefs held by a person as a result of their background, culture, tastes, and experience. But how can we represent perspective computationally?

In this paper, we present PerspectiveSpace, a new technique for modeling spaces of users and their beliefs. PerspectiveSpace represents these spaces as a matrix of users, and data on how people agree or disagree on assertions that they themselves have expressed. It uses Principal Component Analysis (PCA) to reduce the dimensionality of that matrix, discovering the most important axes that best characterize the space. It can then express user perspectives and opinions in terms of these axes. For recommender systems, because it discovers patterns in the beliefs about items, rather than similarity of the items or users themselves, it can perform more nuanced categorization and recommendation. It integrates with our more general common sense reasoning technique, AnalogySpace, which can reason over the content of expressed opinions.

An application of PerspectiveSpace to movie recommendation, 2-wit, is presented. A leave-one-out test shows that PerspectiveSpace captures the consistency of users’ opinions very well. The technique also has applications ranging from discovering subcultures in a larger society, to building community-driven web sites.

A key aspect to study in the field of interface agents is the need to detect as soon as possible the user intentions. User intentions have an important role for an interface agent because they serve as a context to define the way in which the agents can collaborate with users. Intention recognition can be used to infer the user’s intentions based on the observation of the tasks the user performs in a software application. In this work, we propose an approach to model the intentions the user can pursue in an application in a semi-automatic way, based on Variable-Order Markov models. We claim that with appropriate training from the user, an interface agent following our approach will be able both to detect the user intention and the most probable sequence of following tasks the user will perform to achieve his/her intention.

Adapting user interfaces to cultural preferences has been shown to improve a user’s performance, but is oftentimes foregone because of its time-consuming and costly procedure. Moreover, it is usually limited to producing one uniform user interface (UI) for each nation disregarding the intangible nature of cultural backgrounds. To overcome these problems, we exemplify a new approach with our culturally adaptive web application MOCCA, which is able to map information in a cultural user model onto adaptation rules in order to create personalized UIs. Apart from introducing the adaptation flexibility of MOCCA, the paper describes a study with 30 participants in which we compared UI preferences to MOCCA’s automatically generated UIs. Results confirm that automatically predicting cultural UI preferences is possible, paving the way for low-cost cultural UI adaptations.

The vast amount of information presented in museums is often overwhelming to a visitor, making it difficult to select personally interesting exhibits. Advances in mobile computing and user modelling have made possible technology that can assist a visitor in this selection process. Such a technology can (1) utilise non-intrusive observations of a visitor’s behaviour in the physical space to learn a model of his/her interests, and (2) generate personalised exhibit recommendations based on interest predictions. Due to the physicality of the domain, datasets of visitors’ behaviour (i.e. visitor pathways) are difficult to obtain prior to deploying mobile technology in a museum. However, they are necessary to assess different modelling techniques. This paper reports on a methodology that we used to conduct a manual data collection, and describes the dataset we obtained. We also present two collaborative models for predicting a visitor’s viewing times of unseen exhibits from his/her viewing times at visited exhibits (viewing time is indicative of interest), and evaluate our models with the dataset we collected. Both models achieve a higher predictive accuracy than a non-personalised baseline.

This paper examines the problem of uncertainty due to instrumentation in user modeling systems within spatial domains. We consider the uncertainty of inferring a user’s trajectory within a physical space combined with the uncertainty due to inaccuracies in measuring a user’s position. A framework for modeling both types of uncertainties is presented, and applied to a real-world case study from the museum domain. Our results show that this framework may be used to investigate the effects of layout in a gallery, and to explore the degradation in the predictive performance of user models due to measurement error. This information in turn may be used to guide the curation of the space, and the selection of sensing technologies prior to instrumenting the space.

User modeling systems have been influenced by the overspread of Web 2.0 and social networks. New systems aimed at helping people finding information of interest and including “social functions” like social networks, tagging, commenting, inserting content, arose. Such systems are the so-called “social recommender systems”. The idea at the base of social recommender systems is that the recommendation of content should follow user’s preferences while social network just represents a group of users joined by some kind of voluntary relation and does not reflect any preference. We claim that social network is a very important source of information to profile users. Moving from theories in social psychology which describe influence dynamics among individuals, we state that joining in a network with other people exposes individuals to social dynamics which can influence their attitudes, behaviours and preferences.

We present in this paper

SoNARS

, a new algorithm for recommending content in social recommender systems. SoNARS targets users as members of social networks, suggesting items that reflect the trend of the network itself, based on its structure and on the influence relationships among users.

Shopping lists play a central role in grocery shopping. Among other things, shopping lists serve as memory aids and as a tool for budgeting. More interestingly, shopping lists serve as an expression and indication of customer needs and interests. Accordingly, shopping lists can be used as an input for recommendation techniques. In this paper we describe a methodology for making recommendations about additional products to purchase using items on the user’s shopping list. As shopping list entries seldom correspond to products, we first use information retrieval techniques to map the shopping list entries into candidate products. Association rules are used to generate recommendations based on the candidate products. We evaluate the usefulness and interestingness of the recommendations in a user study.

Recent growing interest in predicting and influencing consumer behavior has generated a parallel increase in research efforts on Recommender Systems. Many of the state-of-the-art Recommender Systems algorithms rely on obtaining user ratings in order to later predict unknown ratings. An underlying assumption in this approach is that the user ratings can be treated as ground truth of the user’s taste. However, users are inconsistent in giving their feedback, thus introducing an unknown amount of noise that challenges the validity of this assumption.

In this paper, we tackle the problem of analyzing and characterizing the noise in user feedback through ratings of movies. We present a user study aimed at quantifying the noise in user ratings that is due to inconsistencies. We measure RMSE values that range from 0.557 to 0.8156. We also analyze how factors such as item sorting and time of rating affect this noise.

Recommender systems help users find personally relevant media content in response to an overwhelming amount of this content available digitally. A prominent issue with recommender systems is recommending new content to new users; commonly referred to as the cold start problem. It has been argued that detailed user characteristics, like personality, could be used to mitigate cold start. To explore this solution, three alternative methods measuring users’ personality were compared to investigate which would be most suitable for user information acquisition. Participants (

N

= 60) provided user ease of use and satisfaction ratings to evaluate three different interface variants believed to measure participants’ personality characteristics. Results indicated that the

NEO interface

and the

CFG interface

were promising methods for measuring personality. Results are discussed in terms of potential benefits and broader implications for recommender systems.

In this paper we address the problem of prioritising feedback on the basis of multiple heterogeneous pieces of information in exploratory learning. The problem arises when multiple types of feedback are required in order to address different types of conceptual difficulties, accommodate particular learning behaviours identified during exploration, and provide appropriate support depending on the learning mode (e.g. individual or collaborative learning) and/or the stage of the exploratory learning process. We propose an approach that integrates learners’ characteristics and context-related information through a Multicriteria Decision-Making formalism. The outcome is a context-aware mechanism for prioritising personalised feedback that is tested in an exploratory learning environment for mathematical generalisation.

Web search is the dominant form of information access and everyday millions of searches are handled by mainstream search engines, but users still struggle to find what they are looking for, and there is much room for improvement. In this paper we describe a novel and practical approach to Web search that combines ideas from personalization and social networking to provide a more collaborative search experience. We described how this has been delivered by complementing, rather than competing with, mainstream search engines, which offers considerable business potential in a Google-dominated search marketplace.

Collaborative filtering (CF) is at the heart of most successful recommender systems nowadays. While this technique often provides useful recommendations, conventional systems also ignore data that could potentially be used to refine and adjust recommendations based on a user’s context and preferences. The problem is particularly acute with mobile systems where information delivery often needs to be contextualized. Past research has also shown that combining CF with other techniques often improves the quality of recommendations. In this paper, we present results from an experiment assessing user satisfaction with recommendations for leisure activities that are obtained from different combinations of these techniques. We show that the most effective mix is highly dependent on a user’s familiarity with a geographical area and discuss the implications of our findings for future research.

Today’s mobile leisure guide systems give their users unprecedented help in finding places of interest. However, the process still requires significant user interaction, for example to specify preferences and navigate lists. While interaction is effective for obtaining desired results, learning the interaction pattern can be an obstacle for new users, and performing it can slow down experienced users. This paper describes how to infer a user’s high-level activity automatically to improve recommendations. Activity is determined by interpreting a combination of current sensor data, models generated from historical sensor data, and priors from a large time-use study. We present an initial user study that shows an increase in prediction accuracy from 62% to over 77%, and discuss the challenges of integrating activity representations into a user model.

This paper covers a comparison between two distinct approaches to neural network modeling. The first one is based on a developing of a single neural network model to predict bad debt events. The second one is based on combined models, building firstly a clustering model to recognize the pattern assigned to the customers, with a particular focus on the insolvency, and then developing several distinct neural networks to predict bad debt. In the second approach, for each group identified by the clustering model one neural network had been constructed. In that way, we turned the quite heterogeneous customer base more homogeneous, increasing the average accuracy for the predictive modeling once several straightforward models were built.

Current publication sharing systems weakly support creation and personalization of customized user concept spaces. Focusing the attention on the user, SharingPapers, the adaptive publication sharing system proposed in this paper, allows users to organize documents in flexible and dynamic concept spaces; to merge their concept map with a social network connecting people involved in the domain of interest; to support knowledge expansion generating adaptive recommendations. SharingPapers presents a multi-agent architecture and proposes a new way of representing user profiles, their evolution and views of them.

We report on results of a paper-based lab study that used information on task performance, self appraisal and personal learning need assessment to validate the adaptation mechanisms for a work-integrated learning system. We discuss the results in the wider context of the evaluation of adaptive systems where the validation methods we used can be transferred to a work-based setting to iteratively refine adaptation mechanisms and improve model validity.

The web has become a major source of information to learn about a topic. With the continuous growth of information and its high connectivity, it is hard to follow only the links that are relevant and not to get lost in hyperspace. Our aim is to support people who read documents in a highly connected information space, helping them remain on focus. Our contextually-aware in-browser text summarisation tool, IBES, does this by capturing users’ current interests and providing users with contextualised summaries of linked documents, to help them decide whether the link is worth following.

This paper focuses on the utilization of the history of navigation within recommender systems. It aims at designing a collaborative recommender based on Markov models relying on partial matching in order to ensure high accuracy, coverage, robustness, low complexity while being anytime.

Indeed, contrary to state of the art, this model does not simply match the context of the active user to the context of other users but partial matching is performed: the history of navigation is divided into several sub-histories on which matching is performed, allowing the matching constraints to be weakened. The resulting model leads to an improvement in terms of accuracy compared to state of the art models.

We are working on the problem of modeling an analyst’s intent in order to improve collaboration among intelligence analysts. Our approach is to infer the analyst’s goals, commitment, and actions to improve the effectiveness of collaboration. This is a crucial problem to ensure successful collaboration because analyst intent provides a deeper understanding of what analysts are trying to achieve and how they are achieving their goals than simply modeling their interests. The novelty of our approach relies on modeling the process of committing to a goal as opposed to simply modeling topical interests. Additionally, we dynamically generate a goal hierarchy by exploring the relationships between concepts related to a goal. In this short paper, we present the formal framework of our intent model, and demonstrate how it is used to detect the common goals between analysts using the APEX dataset.

Collaborative filtering (CF) techniques have proved to be a powerful and popular component of modern recommender systems. Common approaches such as

user-based

and

item-based

methods generate predictions from the past ratings of users by combining two separate ratings components: a

base estimate

, generally based on the average rating of the target user or item, and a

neighbourhood estimate

, generally based on the ratings of similar users or items. The common assumption is that the neighbourhood estimate gives CF techniques a considerable edge over simpler

average-rating

techniques. In this paper we examine this assumption more carefully and demonstrate that the influence of neighbours can be surprisingly minor in CF algorithms, and we show how this has been disguised by traditional approaches to evaluation, which, we argue, have limited progress in the field.

The complex nature of virtual environments customarily hinders users to interact in a natural, intuitive and optimal way. Different user characteristics are hardly taken into account when designing 3D user interfaces for virtual environments. We envision that user interaction in virtual environments can be enhanced by integrating adaptation and personalization into 3D user interfaces. Through our research, we aim to provide adaptive and personalized 3D user interfaces for enhancing user interaction in virtual environments. The establishment of a user model becomes an important first step to facilitate adaptation and personalization to the user. In order to partly construct the user model, we carried out an experiment on 3D target acquisition task with four user groups (differing in experience level and gender). In this paper, we present a general user model that will enable first-time users to benefit instantly from adaptation and personalization in virtual environments.

This paper describes an experiment in which we tried to predict the learner’s answers from his brainwaves. We discuss the efficiency to enrich the learner model with some electrical brain metrics to obtain some important information about the learner during a test. We conducted an experiment to reach three objectives: the first one is to record the learner brainwaves and his answers to the test questions; the second is to use machine learning techniques to predict guessed and random answers from the learner brainwaves; the third is to implement an agent that transmits the prediction results to an Intelligent Tutoring System. 21 participants were recruited, 45827 recording were collected and we reached a prediction accuracy of 96%.

Adaptive course generation becomes more appropriate for realistic usage scenarios and more flexible if it includes mechanisms deciding just-in-time which content, which exercises, which external resources, and which tools to include for an individual student. We developed such a just-in-time delivery framework (called Dynamic Items) that is used for enhancing the adaptivity of (educational) online material generated by the web-based platform

ActiveMath

. This paper describes the framework and discusses several new learning opportunities created by Dynamic Items for an individual student.

The web is increasingly used as an information source to gain new knowledge but the management of found web pages can be a challenging task. Often social tagging systems are used for resource management. Besides the obvious use of tags – organizing a collection of web resources – they support functionalities like sharing resources with other users and recommendation of further possibly relevant web pages. This paper describes a novel application based on an extended tagging concept that can improve resource management and recommendation. Adding semantic information to tags and tagging fragments of web pages instead of whole web pages enhance the possibilities of well-known tagging applications. Individual knowledge networks are the basis of this tagging concept. A first prototype is developed as proof of concept.

Working memory (WM) is a psychological construct that has a major effect on information processing, thus signifying its importance when considering individual differences and adaptive educational hypermedia. Previous work of the authors in the field has demonstrated that personalization on human factors, including the WM sub-component of visuospatial sketchpad, may assist learners in optimizing their performance. To that end, a deeper approach in WM has been carried out, both in terms of more accurate measurements and more elaborated adaptation techniques. This paper presents results from a sample of 80 university students, underpinning the importance of WM in the context of an e-learning application in a statistically robust way. In short, learners that have low WM span expectedly perform worse than learners with higher levels of WM span; however, through proper personalization techniques this difference is completely alleviated, leveling the performance of low and normal WM span learners.

In this paper, we present a novel approach to track user interaction on a web page based on JavaScript-events combined with the Semantic Web standard Microformats to obtain more fine-grained and meaningful user information. Today’s user tracking solutions are mostly page-based and lose valuable information about user interactions. To get an in-depth understanding of user’s interests and intentions from observing him while interacting on a website, interaction data needs to be tracked on an event rather than on a page basis enhanced with semantic knowledge to understand the user intention. Our goal is to create an easy-to-integrate user tracker that is capable of collecting tracking information of configurable depth and feeding a highly sophisticated user model needed to provide personalized services such as recommendation and search.

Workers from all sectors use software applications to complete day-to-day tasks. The mastery of new software applications can be frustrating to users who are otherwise job-experts and can temporarily decrease productivity. Job and task experts are not well served by tutoring approaches that combine instruction about the task with instruction about the tool. This work presents an architecture and prototype implementation that selects timely, task-appropriate hints for expert users as they work with an application to complete real tasks. The architecture maintains models of user and task, as well as a specialized model of tutoring-for-experts that was created by observing human tutors. This research shows that domain experts can be successfully scaffolded with adaptive hints while doing their work and that they endure less cognitive load than users for whom the scaffolding is not adapted to the task.

In this paper, we introduce a novel approach for modelling user interests. Our approach captures users’ evolving information needs, identifies aspects of their need and recommends relevant news items to the users. We introduce our approach within the context of personalised news video retrieval. A news video data set is used for experimentation. We employ a simulated user evaluation.

User behaviors on a system vary not only among individuals but also within the same user when he/she gains experience on the system. We empirically investigated how individual users changed their behaviors on the basis of long-term data, which were collected by our telephone-based spoken dialogue system deployed for the open public over 34 months. The system was repeatedly used by citizens, who were each identified by their phone numbers. We conducted an experiment by using these data and showed that prediction accuracy of utterance-understanding errors improved when the temporal change was taken into consideration. This result showed that modeling temporally changing user behaviors was helpful in improving the performance of spoken dialogue systems.

Ideally designed software allow users to explore and pursue interleaving plans, making it challenging to automatically recognize user interactions. The recognition algorithms presented use constraint satisfaction techniques to compare user interaction histories to a set of ideal solutions. We evaluate these algorithms on data obtained from user interactions with a commercially available pedagogical software, and find that these algorithms identified users’ activities with 93% accuracy.

Recommender systems have explored a range of implicit feedback approaches to capture users’ current interests and preferences without intervention of users’ work. However, current research focuses mostly on implicit positive feedback. Implicit negative feedback is still a challenge because users mainly target information they want. There have been few studies assessing the value of negative implicit feedback. In this paper, we explore a specific approach to employ implicit negative feedback and assess whether it can be used to improve recommendation quality.

This paper describes the evaluation of a Scrutable User Modelling Infrastructure. SUMI is intended to form a service to allow users to share their user models from social e-networking and e-commerce providers to educational systems. The model is scrutable, meaning users can inspect and correct the data that is held about them, and implements privacy policies so that users can control how their models are accessed by other users. This evaluation was conducted with 107 users, which were exposed to a prototype service, for determining whether the proposed scrutability and privacy privileges were acceptable to the users, whether the users were able to achieve the desired outcome, and whether they understood the consequences of their interactions with the system. The conclusions show that the users expressed their general approval of the proposed privileges while making useful suggestions regarding improvements to the presentation and interface to the system.

To provide personalized health recommendations concerning disabled persons, an adaptive system needs a detailed user model that can account for the peculiar aspects of the many existing disabilities. This paper describes how we built such a user model and illustrates the Web-based system that allows all interested stakeholders to access and provide user model data.

This paper proposes a novel trail sharing system for mobile devices that deals with context information collected by sensors, as well as users’ personal opinions (e.g., landscape beauty) specified by ratings. To help the user in finding trails that are more suited to her, the system exploits a collaborative filtering approach to predict the ratings users may give to untried trails, and applies a similar approach also to context information that can significantly vary among users (e.g., lap duration).

Clustering of search results has been shown to be advantageous over the simple list presentation of search results. However, in most clustering interfaces, the clusters are not adaptive to a user’s interaction with the clustering results, and the important question “how to optimize the benefit of a clustering interface for a user” has not been well addressed in the previous work. In this paper, we study how to exploit a user’s clickthrough information to adaptively reorganize the clustering results and help a user find the relevant information more quickly. We propose four strategies for adapting clustering results based on user actions. We propose a general method to simulate different kinds of users and linearize the cluster results so that we can compute regular retrieval measures. The simulation experiments show that the adaptation strategies have different performance for different types of users; in particular, they are effective for “smart users” who can correctly recognize the best clusters, but not effective for “dummy users” who follow system’s ranking of results. We further conduct a user study on one of the four adaptive clustering strategies to see if an adaptive clustering system using such a strategy can bring users better search experience than a static clustering system. The results show that there is generally no significant difference between the two systems from a user’s perspective.

When developing an

Adaptive Learning System

(ALS), users are generally consulted (if at all) towards the end of the development cycle. This can limit users’ feedback to the characteristics and idiosyncrasies of the system at hand. It can be difficult to extrapolate principles and requirements, common to all ALSs, that are rated highly by users. To address this problem, we have elicited requirements from learners and teachers across several European academic institutions through explorative, semi-structured interviews [1]. The goal was to provide a methodology and an appropriate set of questions for conducting such interviews and to capture the essential requirements for the early iterations of an ALS design. In this paper we describe the methodology we employed while preparing, conducting, and analyzing the interviews and we present our findings along with objective and subjective analysis.

Traditional websites have long relied on users revealing their preferences explicitly through direct manipulation interfaces. However recent recommender systems have gone as far as using implicit feedback indicators to

understand

users’ interests. More than a decade after the emergence of recommender systems, the question whether users prefer them compared to stating their preferences explicitly, largely remains a subject of study. Even though some studies were found on users’ acceptance and perceptions of this technology, these were general marketing-oriented surveys. In this paper we report an in-depth user study comparing Amazon’s implicit book recommender with a baseline model of explicit search and browse. We address not only the question “do people accept recommender systems” but also how or under what circumstances they do and more importantly, what can still be improved.

Web usability has been considered as a key issue to the success of the Web. However, Web readers typically face difficulties since Web pages are presented according to the local contexts of Web authors. Web authors and readers follow their own local contexts to represent and interpret Web contents as they originate from different communities. Hence, there is a need to transform Web contents created according to the authors’ contexts into the different contexts of their readers.

In this paper, we aim at presenting a solution that provides Web contents according to the reader’s context. Our solution is based on an explicit representation of the authors’ and readers’ local contexts. We rely on RDFa to annotate contents with the author’s context and we provide an adaptation process on the client-side that generates contextualized Web contents according to the readers’ contexts. We validate our approach through a Firefox extension.

Plan recognition of movement by car or foot is generally intractable because of the huge number of potential destinations and routes. However in restricted areas with limited ingress/egress and few places to go such as a military base, plan recognition of movement can be done. The ABM system uses RFID and Lidar to track the movement of vehicles and people, infer their plans/goals, and distinguish threat from normal behavior. ABM represents plans as a series of polygons that abstract important road/terrain features such as intersections and driveways. ABM’s keyhole plan recognition algorithm handles unobserved steps caused by insufficient data rates or deficient sensor coverage and handles position inaccuracies due to limited sensor precision or multi-path reflections from buildings. ABM guards privacy by storing only a person’s role (e.g., visitor, office worker, grounds keeper) on the military base.

Users have become accustomed to a web that is more than an interactive hypermedia but is a complex mix of rich multimedia services and hypermedia content. Users are now contributors and active participants on the web. However, Pesonalisation technologies, such as Adaptive Hypermedia, have so far focused almost exclusively on adaptive content delivery resulting in their failure to become a high impact technologies. The absence of rich multimedia services in the current generation of Adaptive Hypermedia Systems means that they do not live up to the expectations of users. By providing personalised web experiences that combine both services and content in a seamless environment such systems could not only live up to the expectations of users but could exceed them. This paper presents a system that supports the adaptive selection and sequencing of both content and services in a unified manner. By applying techniques used in content based Adaptive Hypermedia to services with making use of the state of the art in service composition, this system delivers personalised web experiences that combine adaptively selected and sequenced content and services. The integration of appropriate content with services can improve the experience of the user as well as making the activity more efficient.