The history of personalisation and recommender systems is, in large part, a web-tale: a story of sites and services that learn about users, in order to provide more tailored experiences. The rapid rise of mobile computing, combined with wearable sensors, and an increasingly connected IoT world, has begun to shift the potential for personalisation, from the virtual world of the web, to the physical world in which we live, work, and play. This talk will consider exciting new application opportunities for user modelling, personalisation, and recommendation in the area of personal health and fitness, with a particular emphasis on how these technologies can help people to exercise more effectively, and by drawing from recent results for marathon runners.
Commercial buildings consume a large portion of the total electricity in the United States. One method for energy saving in commercial buildings targets inefficiencies of unoccupied spaces by relaxing the setpoint temperature. However, energy savings are severely limited when occupants are assumed to be "immovable objects"; instead, by encouraging occupant participation in the optimization, a much greater amount of energy savings can be achieved. In this work, we build on this idea and introduce energy saving recommendations based on occupant location. We introduce two types of energy saving recommendations based on location: move recommendations, which recommends the occupant to move from one space to another, and shift schedule recommendations, which recommends the occupant to arrive or depart a set amount of time earlier or later. To investigate the effects of the energy saving recommendations, we introduced a tightly coupled system composing of a simulator and a recommender system. Simulations in our building testbed revealed that energy saving recommendations coupled with occupancy-based HVAC energy management saves 25% more energy than occupancy-based HVAC energy management alone.
Check-in prediction using location-based social network data is an important research problem for both academia and industry since an accurate check-in predictive model is useful to many applications, e.g. urban planning, venue recommendation, route suggestion, and context-aware advertising. Intuitively, when considering venues to visit, users may rely on their past observed visit histories as well as some latent attributes associated with the venues. In this paper, we therefore propose a check-in prediction model based on a neural framework called Preference and Context Embeddings with Latent Attributes (PACELA). PACELA learns the embeddings space for the user and venue data as well as the latent attributes of both users and venues. More specifically, we use a probabilistic matrix factorization-based technique to infer the latent attributes specific to users and locations in location-based social networks (LBSNs), considering the user visitation decisions that could be affected by area attraction, neighborhood competition, and social homophily. PACELA also includes a deep learning neural network to combine both embedding and latent features to predict if a user performs check-in on a location. Our experiments on three different real world datasets show that PACELA yields the best check-in prediction accuracy against several baseline methods.
The Location-Based Social Networks (LBSN) (e.g., Facebook, etc.) have many attributes (e.g., ratings, reviews, etc.) that play a crucial role for the Point-of-Interest (POI) recommendations. Unlike ratings, the reviews can help users to elaborate their consumption experience in terms of relevant factors of interest (aspects). Though some of the existing systems have exploited user reviews, most of them are less transparent and non-interpretable (as they conceal the reason behind recommendation). These reasons have motivated us towards explainable and interpretable recommendation. To the best of our knowledge, only a few of the researchers have exploited user reviews to incorporate the sentiment and opinions on different aspects for personalized and explainable POI recommendation. This paper proposes a model termed as ReEL (Review aware Explanation of Location Recommendation) which models the review-aspect correlation by exploiting deep neural network, formulates user-aspect bipartite relation as a bipartite graph, and models the explainable recommendation by using dense subgraph extraction and ranking-based techniques. The major contributions of this paper are: (i) it models users and POIs using the aspects posted on user reviews, and it provisions incorporation of multiple contexts (e.g., categorical, spatial, etc.) in POI recommendation, (ii) it formulates preference of users' on aspects as a bipartite relation, represents it as a location-aspect bipartite graph, and models the explainable recommendation with the notion of ordered dense subgraph extraction using bipartite cores, shingles, and ranking-based techniques, and (iii) it extensively evaluates the proposed models using three real-world datasets and demonstrates an improvement of 5.8% to 29.5% on F-score metric, when compared to the relevant studies.
The news you read is, of course, a highly individual choice and one for which substantial and successful news recommendation techniques have been developed. But as well as what news you read, the way you choose and read that news is also known to be highly individual. We propose a framework for extending the user profile of news readers with features of these interactions. The extensions are dynamic through monitoring an individual's reading and browsing activity. They include factors learned from the user's interaction log and also factors inferred from category level definitions contained in the framework. We report a study in which users' interaction logs with a news app are used to generate user profiles that are verified with self-reported questionnaire data about reading habits. We discuss the implications of our user modeling approach in news personalisation for both recommendation and user interface personalisation for news apps.
Constrained action-based decision-making is one of the most challenging decision-making problems. It refers to a scenario where an agent takes action in an environment not only to maximize the expected cumulative reward but where it is subject to certain action-based constraints; for example, an upper limit on the total number of certain actions being carried out. In this work, we construct a general data-driven framework called Constrained Action-based Partially Observable Markov Decision Process (CAPOMDP) to induce effective pedagogical policies. Specifically, we induce two types of policies: CAPOMDPLG using learning gain as reward with the goal of improving students' learning performance, and CAPOMDPTime using time as reward for reducing students' time on task. The effectiveness of CAPOMDPLG is compared against a random yet reasonable policy and the effectiveness of CAPOMDPTime is compared against both a Deep Reinforcement Learning induced policy and a random policy. Empirical results show that there is an Aptitude-Treatment Interaction effect: students are split into High vs. Low based on their incoming competence; while no significant difference is found among the High incoming competence groups, for the Low groups, students following CAPOMDPTime indeed spent significantly less time than those using the two baseline policies and students following CAPOMDPLG significantly outperform their peers on both learning gain and learning efficiency.
Guiding students to the learning activities that are most appropriate for their current level of knowledge is one of the goals that adaptive educational systems tried to achieve during the last decades. Recently, several attempts have been made to use Open Learner Models (OLM) as a tool for achieving this goal. While the original goal of OLM is to help students reflect about their own learning process, extending OLM with navigation support functionality enables students to take immediate actions towards improving their knowledge. In this work, we attempted to extend the navigation support functionality of OLM by developing a fine-grained OLM that offers student knowledge visualization on both topic and concept levels. The fine-grained OLM enables students to directly explore connections between their knowledge and available learning activities, making an informed decision about their next learning steps. To assess the impact of the new type of OLM, we evaluated several versions of it in a classroom study, while also comparing it with data from our earlier studies that featured a coarse-grained OLM. Our results suggest that the fine-grained OLM considerably impacts student choice of learning activities, making student learning more efficient. We also found that the specific design features of fine-grained OLM could affect students' confidence and persistence while selecting and attempting the learning activities.
Recent advances in eye-tracking technologies have introduced the opportunity to incorporate gaze into student modeling. Creating student models that leverage gaze information holds significant promise for game-based learning environments. This paper introduces a gaze-enhanced student modeling framework that incorporates student eye tracking to dynamically predict students' performance in a game-based learning environment for microbiology education, CRYSTAL ISLAND. The gaze-enhanced student modeling framework was investigated in a study comparing a gaze-enhanced student model with a baseline student model that does not utilize student eye-tracking. Results of a study conducted with 65 college students interacting with the CRYSTAL ISLAND game-based learning environment indicate that the gaze-enhanced student model significantly outperforms the baseline model in dynamically predicting student problem-solving performance. The findings suggest that incorporating gaze into student modeling can contribute to a new generation of student models for game-based learning environments.
We present a neural architecture to model student behavior in virtual educational environments using purely unsupervised information. A crucial part of this architecture is the optimization of a joint embedding function to represent both students and course elements into a single shared space. This joint representation is more adequate than disjoint representations because it elicits insights on the relations between students and contents. Moreover, the model is trained only with interactions of the student with online learning platforms, without requiring any additional manual labeling by experts. We obtain state-of-the-art results using this approach in two types of task: first, dropout prediction in online courses (MOOCs), and second Knowledge Tracing in Intelligent Tutoring Systems (ITS). We explore how the deep architecture is flexible enough to capture variables related to different phenomena, such as engagement or skill mastery.
Diversity has been identified as one of the key dimensions of recommendation utility that should be considered besides the overall accuracy of the system. A common diversification approach is to rerank results produced by a baseline recommendation engine according to a diversification criterion. The intent-aware framework is one of the frameworks that has been proposed for recommendations diversification. It assumes existence of a set of aspects associated with items, which also represent user intentions, and the framework promotes diversity across the aspects to address user expectations more accurately. In this paper we consider item-based collaborative filtering and suggest that the traditional view of item similarity is lacking a user perspective. We argue that user preferences towards different aspects should be reflected in recommendations produced by the system. We incorporate the intent-aware framework into the item-based recommendation algorithm by injecting personalised intent-aware covariance into the item similarity measure, and explore the impact of such change on the performance of the algorithm. Our experiments show that the proposed method improves both accuracy and diversity of recommendations, offering better accuracy/diversity tradeoff than existing solutions.
Collaborative filtering (CF) has made it possible to build personalized recommendation models leveraging the collective data of large user groups, albeit with prescribed models that cannot easily leverage the existence of known behavioral models in particular settings. In this paper, we facilitate the combination of CF with existing behavioral models by introducing Bayesian Behavioral Collaborative Filtering (BBCF). BBCF works by embedding arbitrary (black-box) probabilistic models of human behavior in a latent variable Bayesian framework capable of collectively leveraging behavioral models trained on all users for personalized recommendation. There are three key advantages of BBCF compared to traditional CF and non-CF methods: (1) BBCF can leverage highly specialized behavioral models for specific CF use cases that may outperform existing generic models used in standard CF, (2) the behavioral models used in BBCF may offer enhanced intepretability and explainability compared to generic CF methods, and (3) compared to non-CF methods that would train a behavioral model per specific user and thus may suffer when individual user data is limited, BBCF leverages the data of all users thus enabling strong performance across the data availability spectrum including the near cold-start case. Experimentally, we compare BBCF to individual and global behavioral models as well as CF techniques; our evaluation domains span sequential and non-sequential tasks with a range of behavioral models for individual users, tasks, or goal-oriented behavior. Our results demonstrate that BBCF is competitive if not better than existing methods while still offering the interpretability and explainability benefits intrinsic to many behavioral models.
The "black box'' nature of today's recommender systems raises a number of challenges for users, including a lack of trust and limited user control. Providing more user control is interesting to enable end-users to help steer the recommendation process with additional input and feedback. However, different users may have different preferences with regard to such control. To the best of our knowledge, no research has investigated the effect of personal characteristics on visual control techniques in the music recommendation domain. In this paper, we present results of a user study on the web using two different visualisation techniques (a radar chart and sliders) that allows users to control Spotify recommendations. A within-subject design withLatin Square counterbalancing measures was used for the study. Results indicate that the radar chart helped the participants discover a significantly higher number of new songs compared to the sliders. We also found that users' experience with Spotify had an influence on their interaction with different musical attributes. The participants who used Spotify frequently and users with a high individual musical sophistication interacted with the attributes significantly more with the radar chart compared to the sliders. Individual musical sophistication also had a significant impact on their interaction with the interaction techniques. The participants with high musical sophistication interacted significantly more with the radar chart in comparison to the sliders. Based on the feedback from our participants, we provide design suggestions to further improve user control in music recommendation.
Recommender systems have been widely applied in the literature to suggest individual items to users. In this paper, we consider the harder problem of package recommendation, where items are recommended together as a package. We focus on the clothing domain, where a package recommendation involves a combination of a 'top' (e.g. a shirt) and a 'bottom' (e.g. a pair of trousers). The novelty in this work is that we combined matrix factorisation methods for collaborative filtering with hand-crafted and learnt fashion constraints on combining item features such as colour, formality and patterns. Finally, to better understand where the algorithms are underperforming, we conducted focus groups, which lead to deeper insights into how to use constraints to improve package recommendation in this domain.
In this paper we present the results of a user study focusing on social relationships within small groups. The goal is to better understand how to incorporate the information about social relationships in group recommendation models. Our analysis, conducted on a data set of 150 participants in 41 groups deciding on a travel destination to visit together, brings out some intriguing outcomes. We demonstrate that social centrality is hardly an indicator of the social influence in the decision-making process of "equality matching" types of groups. However, socially central group members and socially close groups are significantly happier with group decisions than those who are loosely related. Moreover, in this paper we show that social relationships are indicators of other concepts relevant in group settings, therefore in group recommender systems as well.
The relationship between learners' cognitive and affective states has become a topic of increased interest, especially because it is an important component of self-regulated learning (SRL) processes. This paper studies sixth grade students' SRL processes as they work in Betty's Brain, an agent-based open-ended learning environment (OELE). In this environment, students learn science topics by building causal models. Our analyses combine observational data on student affect to log files of students' interactions within the OELE. Preliminary analyses show that two relatively infrequent affective states, boredom and delight, show especially marked differences among high and low performing students. Further analysis shows that many of these differences occur after receiving feedback from the virtual agents in the Betty's Brain environment. We discuss the implications of these differences and how they can be used to construct adaptive personalized scaffolds.
Research shows that Worked Examples (WE) and Erroneous Examples (ErrEx) provide learning benefits, particularly when presented alternatively with problems to solve. We previously proposed an adaptive strategy for selecting WE, ErrEx, and Problem Solving (PS) adaptively based on the student's problem-solving score and found that the adaptive strategy was beneficial for students in comparison to learning from a fixed sequence of alternating WE/PS pairs and ErrEx/PS pairs . Students who received learning activities adaptively achieved the same learning outcomes as their peers in a fixed condition, but with fewer learning activities . In this paper, we investigate a different adaptive strategy, which provides WEs and ErrExs to novices, and ErrEx and PS to advanced students. We found that the original adaptive strategy  is more effective than the new adaptive strategy. Furthermore, both novices and advanced students who learned with the original adaptive strategy demonstrated better performance on the post-test.
An important role of open learner models (OLMs) is to support self-reflection. We explore how to do this for an OLM based on fine-grained long term physical activity tracker data that many people are accumulating. We aim to tackle two well-documented challenges that people face, in making effective use of an OLM for reflection. 1. We created a tutorial to scaffold sense-making needed to understand the meaning of the OLM. 2. We integrated an interface scaffold to help users consider key questions for effective reflection. We report the results of a qualitative think-aloud lab study with 21 participants viewing their own long term OLM. To evaluate the tutorial scaffolding, we split participants into an experimental group, who did a tutorial before exploring the OLM and a control group which explored the interface without the tutorial. To evaluate the reflection scaffolding, all participants first explored the interface as they wished. We then provided goal prompts to scaffold reflection. Our study revealed that, under lab conditions, the tutorial scaffolding was not needed - all participants in both groups could readily understand the OLM. However, we found that several of the goal prompts were important to help participants consider key questions for effective reflection. Our key contribution is insights into the design of scaffolding for reflection in a life-long learning context of gaining insights and setting goals for physical activity.
Instagram is a popular social networking application that allows users to express themselves through the uploaded content and the different filters they can apply. In this study we look at personality prediction from Instagram picture features. We explore two different features that can be extracted from pictures: 1) visual features (e.g., hue, valence, saturation), and 2) content features (i.e., the content of the pictures). To collect data, we conducted an online survey where we asked participants to fill in a personality questionnaire and grant us access to their Instagram account through the Instagram API. We gathered 54,962 pictures of 193 Instagram users. With our results we show that visual and content features can be used to predict personality from and perform in general equally well. Combining the two however does not result in an increased predictive power. Seemingly, they are not adding more value than they already consist of independently.
One of the important aspects of movie-making is to trigger emotional responses in viewers. These emotional experiences can be divided into hedonic and eudaimonic. While the former are characterized as plain enjoyment, the latter deal with getting greater insight, self-reflection or contemplation. So far, modeling of user preferences about movies and personalization algorithms have largely ignored the eudaimonic aspect of the consumption of movies. In this paper we fill this gap by exploring what are the relationship between (i) eudaimonic and hedonic characteristics of movies, (ii) users' preferences and (iii) users' personality. Our results show that eudaimonic user profiling effectively divides users into pleasure-seekers and meaning-seekers.
Social media platforms such as blogs, wikis and file sharing have become very popular in enterprises. Despite their effectiveness in increasing collaboration in the organization, employees are overloaded with information originating from these many sources and find it hard to orient themselves in the stream of events occurring in their organizational news feed. In this paper we identify what makes an event in an organizational social media platform important to employees. Once important factors of an event to an employee are identified, the stream of events can be personalized and prioritized based on those and thus reduce the overload and assist in work efficiency. Through interviews and two extensive user surveys, the first hypothetical and the second empirical, we identified which factors of an event make it important and compare results from the hypothetical and empirical surveys.
As robots begin to integrate our world and invade our streets and homes, they must act as autonomous and intelligent beings. However, so far, they are deprived of our responsive and emotional capacities, lacking awareness of the social world we live in. In the future, robots should be able to take into account these distinctive dimensions of human social interactions to be able to act appropriately within such social contexts. To do this, they must adapt and embody the essence of social and emotional intelligence. This not only includes the ability to recognise human emotions and social interactions but also understand them, deliberate them and act accordingly. Lately, significant research has been carried out in an attempt to find ways to build social and emotional robots that are able to perceive the user's emotions, adapt to them, and react appropriately. This talk will, therefore, provide an overview of the area of emotions in social interactions established between humans and social robots. In this analysis, I will use scenarios from educational and entertainment robotics, outline the process of building emotional social robots and finally proceed to interpret the effect that such capabilities have on user's interactions, learning, motivation, relationship and trust. I believe that by studying and engineering emotional and social interactions "for" and "with" robots, we have the opportunity to build a new generation of natural, engaging, effective and, most importantly, "humane" AI.
Recommender systems are evaluated based on both their ability to create a satisfying user experience and their ability to help a user make better choices. Despite this, quantitative evidence from previous research in recommender systems indicate very high correlations between user experience attitudes and choice satisfaction. This might imply invalidity in the measurement methodologies of these constructs, whereas they may not be measuring what researchers think they are measuring. To remedy this, we present a new methodology for the measurement of choice satisfaction. Part of our approach is to measure a user's "ease of satisfaction," or that user's natural propensity to be satisfied, which is measured using three different approaches. An (N=526) observational study is conducted wherein users browse a movie catalog. A factor analysis is done to assess the discriminant validity of our proposed choice satisfaction apparatus from user experience. A statistical analysis suggests that accounting for ease-of-satisfaction allows for a model of choice satisfaction that is not only discriminant, but independent, from user experience. This enables researchers to more objectively identify recommender system factors that lead users to good choices.
Explanations can give credibility to recommendations and help users to make better choices. In current recommender systems, explanation is a step that comes after recommendation. In this paper, we describe an approach that turns recommender systems on their head. In our approach, which we call Recommendation-by-Explanation (hr-by-e ), the system constructs a reason, or explanation, for recommending each candidate item; then it recommends those candidate items that have the best explanations. By unifying recommendation and explanation, r-by-e finds relevant recommendations with explanations that have a high degree of fidelity. We present the results of an offline experiment using a movie recommendation dataset. We show that r-by-e achieves higher precision than a comparable recommender, while both produce recommendations with roughly equal levels of diversity and serendipity. We also present the results of deploying a web-based system through which we have conducted two user trials. In one trial, we evaluate recommendation quality. Participants in this trial found r-by-e's recommendations to be more diverse, serendipitous and relevant than those of the competitor system. In another trial, we evaluate explanation quality. We used a re-rating task: users rated recommendations initially in the case where they were given only the explanation and not the identity of the movie, and then re-rated in the case where they were given information about the recommended movie. We found a stronger correlation between the pairs of ratings in the case of r-by-e. This suggests that r-by-e's explanations allow users to make more accurate judgments about the quality of recommended items.
We present multi-modal adversarial autoencoders for recommendation and evaluate them on two different tasks: citation recommendation and subject label recommendation. We analyze the effects of adversarial regularization, sparsity, and different input modalities. By conducting 408 experiments, we show that adversarial regularization consistently improves the performance of autoencoders for recommendation. We demonstrate, however, that the two tasks differ in the semantics of item co-occurrence in the sense that item co-occurrence resembles relatedness in case of citations, yet implies diversity in case of subject labels. Our results reveal that supplying the partial item set as input is only helpful, when item co-occurrence resembles relatedness. When facing a new recommendation task it is therefore crucial to consider the semantics of item co-occurrence for the choice of an appropriate model.
Open data initiatives and policies have triggered a dramatic increase in the volume of available research data. This, in turn, has brought to the fore the challenge of helping users to discover relevant datasets. Research data repositories support data search primarily through keyword search and faceted navigation. However, these mechanisms may suit users, who are familiar with the structure and terminology of the repository. This raises the problem of personalized dataset recommendations for users unfamiliar with the repository or not able to clearly articulate their information needs. To this end, we present and evaluate in this paper a recommendation approach applied to a new task --- recommending research datasets. Our approach hybridizes content-based similarity with item-to-item co-occurrence, tuned to a feature weighting model obtained through a survey involving real users. We applied the approach in the context of a live research data repository and evaluated it in a user study. The obtained user judgments reveal the ability of the proposed approach to accurately quantify the relevance of datasets and they constitute an important step towards developing a practical dataset recommender.
Many state-of-the-art recommender systems are known to suffer from popularity bias, which means that they have a tendency to recommend items that are already popular, making those items even more popular. This results in the item catalogue being not fully utilised, which is far from ideal from the business' perspective. Issues of item exposure are actually more complex than simply overexposure of popular items. In this paper we look at the exposure of individual items to different groups of consumers, the item's audience, and address the question of whether recommender systems reach each item's potential audience. Thus, we go beyond state-of-the-art analyses that have simply addressed the extent to which items are recommended, regardless of whether they are reaching their target audience. We conduct an empirical study on the MovieLens 20M dataset showing that recommender systems do not fully utilise items' audiences, and existing sales diversity optimisers do not improve their exposure.
Behavior modeling has become a very important behavior change technique employed in most fitness apps. However, its effect as a persuasive strategy on users has not been well investigated. Consequently, we conducted an empirical study among 669 participants to uncover: (1) how the perceived persuasiveness of behavior model design influences three social cognitive theory (SCT) determinants of behavior: self-efficacy, self-regulation and outcome expectation; and (2) the moderating effect of gender-based personalization. We based our study on user evaluation of prototypes of behavior models performing push-up and squat exercise behaviors as a case study. Our results show that, overall, the perceived persuasiveness of behavior models significantly influences all of the three SCT factors. The effect of persuasiveness on self-regulation (β = 0.42, p < 0.001) and outcome expectation (β = 0.41, p < 0.001) is stronger than on self-efficacy (β = 0.13, p < 0.05). Moreover, the behavior model design has a stronger effect on females' self-efficacy and males' outcome expectation if personalized to their gender. We discuss the implication of our findings.
Personalizing persuasive technologies (PTs) is one of the hallmarks of a successful PT intervention. However, there is a lack of understanding of how Africans and North Americans differ or are similar in the susceptibility to persuasive strategies. To bridge this gap, we conducted a cross-cultural study among 284 subjects to investigate the moderating effect of culture on the susceptibility of users to Cialdini's principles of persuasion. Specifically, using Nigeria and Canada as a case study, we investigated how both groups vary in their levels of susceptibility to Authority, Commitment, Consensus, Liking, Reciprocity and Scarcity. The results of our analysis show that Nigerians are more susceptible to Authority and Scarcity than Canadians, while Canadians are more susceptible to Reciprocity, Liking and Consensus than Nigerians. However, both groups do not differ with respect to Commitment (the most persuasive strategy). Finally, we discussed our findings and mapped the most persuasive Cialdini's principles in each group to implementable persuasive strategies in the PT domain.
Group decision making is performed in real life for the purpose of selecting an optimal solution for the whole group. Decision making behavior of group members could be impacted by item domains and the chronological order in which decision tasks are presented to groups. In this paper, we analyze situations where group members could apply different decision strategies depending on the chronological order of decision tasks. The data analysis results confirm that item domains and the order of decision tasks have an impact on group decision strategies. This is especially the case if preferences of a minority of group members are significantly different from the other group members and when decision tasks related to high-involvement item domains are arranged before decision tasks related to low-involvement item domains. In addition, we also figured out that group members invest different amounts of time in making a decision task depending on its position in a sequence of decision tasks.
Adaptive user authentication policies are moving in the center of attention lately aiming to assist users in creating memorable and secure passwords. Focusing on graphical user authentication, state-of-the-art research has provided evidence that image-related attributes affect password memorability and security. Nonetheless, the effects of users' contemporary cultural-related memories towards password memorability and security have not been investigated so far, although it is known that user authentication is a cross-cultural task. Aiming to shed light on whether such effects exist, we conducted a study in which users created a graphical password with a contemporary culture-intensive vs. a culture-neutral image. Results indicate that image content related to one's cultural-related memories affects the interaction behavior during password composition, and consequently password memorability. Findings point towards a promising new direction for considering human contemporary cultural memories in the design of adaptive password policies to increase memorability and preserve security.
People typically eat what they shop for; if consumers shop for healthy foods, they will likely eat healthy foods. In order to influence healthier eating habits among consumers, it is important to identify the factors that influence them to shop for healthy foods. To contribute to ongoing research in this area, we explore the influence of commonly used e-commerce strategies: personality, persuasive strategies, social support, relative price, and perceived product quality on healthy shopping habits among e-commerce shoppers. Research has shown that personalizing these strategies makes them more effective in achieving the desired behavior change among users. Age and gender have been identified as factors that can be used for group-based personalization. We thus investigate the moderating effect of age and gender on the factors that influence healthy shopping habits in e-commerce shoppers. To achieve this, we carried out an online study of 244 e-commerce shoppers. Using partial least squares structural equation modeling (PLS-SEM), we developed a path model using the commonly used e-commerce factors: personality, persuasive strategies, social support, relative price, and perceived product quality. The result of our analysis suggests that social support, relative price and perceived product quality significantly influence healthy shopping habits in e-commerce shoppers. In addition, females are more influenced by social support to adopt healthy shopping habits compared to male e-shoppers. Furthermore, older shoppers are more influenced by social support to adopt healthy shopping habits, while the younger shoppers are more influenced by the relative price of products.
Given vast number of possible global travel destinations, choosing a destination has become challenging. We argue that traditional media are insufficient to make informed travel decisions, due to a lack of objectivity, a lack of comparability and because information becomes out of date quickly. Thus, travel planning is an interesting field for data-driven recommender systems that support users to master information explosion. We present unresolved research questions with working packages for a doctoral project that combines the fields of recommender systems and user modeling with data mining. The core contributions will be a framework that integrates heterogeneous data sources from the travel domain, novel user modeling techniques and constraint-based recommender algorithms to master the complexities of global travel planning.
Leisure activities constitute an important part of our life. Nowadays, the offer of activities to undertake is constantly growing. This can be easily seen not only by the increasing number of social events created and promoted on Facebook, Couchsurfing, etc., but also by the appearance of specialised online services and event-based social networks, such as Meetup, Eventbrite, etc. Moreover, multi-day events (e.g. conventions, festivals, cruise trips, exhibitions), to which we refer to as distributed events, attract thousands of participants. Their attendees are often overwhelmed with the amount of available options. Recommender systems appear as a common solution in such a context. In this project, we formulate the problem of recommendation of activity sequences and aim at providing an integrated support for users to create a personalised itinerary of activities in order to facilitate their decision making process which events to join. Such assistance is expected to bring a positive impact on well-being and satisfaction with life of individuals.
With the growing popularity of e-commerce, recommender systems play a critical role to enhance the user experience and increase sales revenue and profitability for a company. However, the accuracy of recommender systems is often suffering from data sparsity and the new user/item problems. A promising approach in solving the new user problem is stereotype based modeling. The proposed PhD research will go one step further and develop an item based recommender system employing the stereotype approach for item modeling. For the evaluation of our stereotype-based recommender system, the study will employ the MovieLens and IMDb datasets. These two datasets are integrated using the iSynchronizer tool that was developed by the researchers for tasks such as this. Early evaluation results demonstrate promising prediction accuracy with user-based stereotypes, especially for users with a small number of existing ratings.
The purpose of this research is to design practical link prediction models in signed social networks. Current works focus on the sign prediction, based on the assumption that it is already known whether there is a link between any two users. In other words, the no-relation status is ignored. Meanwhile, the strength of existing links are assumed to be equal, which is also not realistic. In this study, we will redefine the link prediction problem in signed networks and take a deep investigation on no-relation status. Then, we aim to propose a personalized ranking model from the individual's perspective. This research explores link prediction models in a more realistic scenario, and it will contribute to ongoing research in development of link prediction and recommendations in signed networks. Furthermore, our research will provide a better understanding on the link formation mechanism behind signed network evolution.
Wikis are attracting lots of attention for informal learning. The nature of wikis enables learners to freely navigate the learning environment and independently construct knowledge without being forced to follow a predefined learning path in accordance with the constructivist learning theory. Recommendation systems (RS) can provide useful content recommendations in different contexts. To our best knowledge, no effective personalized content recommendation approach has yet been defined to support informal learning in wikis. Therefore, we propose a personalized content recommendation framework to extrapolate topical navigation graphs from learners' free navigation and integrate them with fuzzy thesauri for automatic and adaptive personalized content recommendations to support informal learning in wikis.
Human Computer Interaction (HCI)1 is about information exchange between human and computers. Interaction between users and computers occurs at the User Interface (UI). Now, computers become pervasive, they are embedded in everyday things and UIs are the main value-added competitive advantages. UIs should be more natural for users. NUI (natural user interface) expands forms beyond formal input devices like the mouse and keyboard to more and more natural forms of interaction such as touch, speech, gestures, handwriting, and vision. Unlike speech, handwriting and vision, which have been researched for decades and put into practical use recently, touch and gestures are interaction tasks related, and yet lack of study. This talk will introduce methods of modelling user input action based on data with the random noise for fast touch input and natural gestures.
Psychological theory distinguishes between maximizing and satisficing decision-making styles. Maximizers tend to explore more or all alternatives when making a choice, while satisficers evaluate options until they find one that is good enough. There is limited research that examines how the existence of a recommender influences the choice process and decisions of different types of decision-makers. We report the results of a controlled study, in which we monitored the choice process of participants when provided with automated recommendations and different types of additional information regarding available options. Our analyses show that none of the differences that were expected based on the literature manifested itself in the experiment. Maximizers neither inspected more items, nor invested more time to study them. Instead, like satisficers, they mostly picked one of the top-ranked items recommended by the system, which emphasizes the value of recommenders in particular for maximizers, who would otherwise face a more challenging decision problem. The analysis of the preferences of participants over different types of additional information revealed that highlighting key pros and cons was perceived as particularly helpful for the maximizers, an insight that can be used for the design of explanation approaches for recommenders.
User system trust is critical to the uptake of recommendations, and several factors of trust have been identified and compared. In this paper we present a cross-cultural, crowdsourced study examining user perceptions of nine factors of trust and link the observed differences to trust development processes and cultural dimensions. While some factors consistently instil trust, others are preferred only in certain countries. Our findings and the discovered links are important for design of trusted recommender systems.
When recommendations become increasingly personalized, users are often presented with a narrower range of content. To mitigate this issue, diversity-enhanced user interfaces for recommender systems have in the past found to be effective in increasing overall user satisfaction with recommendations. However, users may have different requirements for diversity, and consequently different visualization requirements. In this paper, we evaluate two visual user interfaces, SimBub and ComBub, to present the diversity of a music recommender system from different perspectives. SimBub is a baseline bubble chart that shows music genres and popularity by color and size, respectively. In addition, ComBub visualizes selected audio features along the X and Y axis in a more advanced and complex visualization. Our goal is to investigate how individual traits such as musical sophistication (MS) and visual memory (VM) influence the satisfaction of the visualization for perceived music diversity, overall usability, and support to identify blind-spots. We hypothesize that music experts, or people with better visual memory, will perceive higher diversity in ComBub than SimBub. A within-subjects user study (N=83) is conducted to compare these two visualizations. Results of our study show that participants with high MS and VM tend to perceive significantly higher diversity from ComBub compared to SimBub. In contrast, participants with low MS perceived significantly higher diversity from SimBub than ComBub; however, no significant result is found for the participants with low VM. Our research findings show the necessity of considering individual traits while designing diversity-aware interfaces.
In online social networks (OSNs), e.g. Facebook, the relationship between users is binary, i.e., either friend (trust) or stranger (distrust). However, in real-world life, people always have different trust relationships with others (e.g., best friend, acquaintance, frenemy). For various applications such as social recommendation and semantic web, it is more worthwhile to know the trust strength between users. In this work, via a unique dataset obtained from a Facebook app and a carefully designed user study, we map trust values with users' online interactions, and thus build personalized trust models. For each individual, we learn her trust model via optimization on a ranking-oriented loss function. Experimental results demonstrate the superior of the proposed approach over state-of-the-art method and the good generalization ability of the approach.
Intentional engagement in positive activities, such as practicing kindness, showing generosity or expressing gratitude, can help people increase their happiness levels and improve their wellbeing. In this paper we explore how a gamified digital behaviour change intervention can be adapted to encourage people of different personality types to engage in simple acts of kindness. Participants were assigned 5 daily activities for 7 days, and asked to complete as many as possible by the end of each day. Participants received different persuasive notifications every day to encourage them to complete a higher number of activities. We investigated how participation levels are influenced by different personality types, different persuasive message types and different categories of activities. Furthermore, we analysed the influence of the intervention on participant behaviour and the effect on behavioural intention, by comparing pre-intention and post-intention to perform different kinds of positive activities. The findings from this study have implications for future work on personalising persuasive interventions to improve wellbeing and prevent mental health problems.
While much is known about how people tweet and interact on Twitter, surprisingly little is known about how the news items tweeted by journalists -- news tweets -- act as a distribution channel for the news that is spread by social media reading and sharing. This paper aims to fill this gap by analyzing the dynamics of news on Twitter, by revealing what drives users to consume news, and by developing a news consumption prediction model. We present the Twitter News Model (TNM), a computational data-driven approach to elucidate the dynamics of news consumption on Twitter. We apply the TNM to a dataset of interactions between users and journalists/newspapers to reveal what drives users' consumption of news on Twitter, and predictively relate users' news beliefs, motivations, and attitudes to their consumption of news. Our findings reveal that news motivations, followed by news attitudes and news beliefs, impact users' behavior of news consumption on Twitter.
Inferring socio-demographic attributes of users is an important and challenging task that could help with personalization, recommendation, advertising, etc. Sensor data collected from mobile devices can be utilized for inferring such attributes. Previous works have focused on combining different types of sensors, such as applications, accelerometer, GPS, battery, and many others, to achieve this task. In this study, we were able to infer attributes, such as gender, age, marital status, and whether the user has children, using solely the GPS sensor. We suggest a novel inference technique, which learns an embedding representation of preprocessed spatial GPS trajectories using an adaption of the Word2vec approach. Based on the embedding representation, we later train multiple classification models to achieve the inference goals. Our empirical results indicate that the suggested embedding approach outperforms a classification approach which does not take into consideration the embedding patterns. Experiments on real datasets collected from Android devices show that the proposed method achieves over 80% accuracy for various demographic prediction tasks.
In recent years, there has been a surge in the use of intelligent computer-supported collaborative learning (CSCL) tools to improve student learning. The aim of this project is to provide adaptive support for student collaboration. My work will focus on designing effective interventions to enhance peer-to-peer collaboration in the context of an Intelligent Tutoring System (ITS), as well as to promote socially-shared regulation of learning (SSRL). In online collaborative learning environments, it can be challenging to engage students. A lack of genuine interaction, social identity, background of students and user empowerment may have negative effects on the learning process. This research will focus on virtualization of the online collaboration, such that the system would help students collaborate as well as support self-regulation and group regulation to achieve higher learning gains.
Ontologies are recognized as a promising approach to support the reusability and interoperability of learners' preferences; which is useful for the optimization and flexibility of data and resources. However, little to no research on adaptive learning systems or semantic technologies explore personalized experiences based on the various out-of-school experiences and activities of the users. This research investigates the design, development, and evaluation of an ontology-based framework for students' interests in a math word problem generator that may be applied to various other learning systems and possibly other domains. The cohesiveness of the problems in addition to the usability, usefulness, and the short-term effectiveness of the derived technology will be investigated by comparing the generated questions to numerical and traditional Algebra I problems. We aim to better understand students' interests to identify the role that their interests can play in semantic technologies, further supporting the recent advances in ontology-based educational technologies and personalized math word problem generators.
Persistence is a non-cognitive attribute referring to one's disposition to attain a specific goal despite challenges and difficulties. Persistence is of particular interest and importance because of its relationship to students' academic achievement and other individual and societal outcomes. Despite claims that persistence is a highly valuable skill, quantitative studies on persistence in technology enhanced learning environments are limited. This study will attempt to build a quantitative model of persistence utilizing machine learning and related methodologies using features distilled from students' domain knowledge assessments and system logs from their use of an Intelligent Tutoring System. The results of the study could potentially provide additional insights on learners' persistence, the factors that influence student persistence, and what interventions to employ to induce learners to persist in tasks.
Collaborative filtering systems heavily depend on user feedback expressed in product ratings to select and rank items to recommend. These summary statistics of rating values carry two important descriptors about the assessed items, namely the total number of ratings and the mean rating value. In this study we explore how these two signals influence the decisions of online users based on choice-based conjoint experiments. Results show that users are more inclined to follow the mean indicator as opposed to the total number of ratings. Empirical results can serve as an input to developing algorithms that foster items with a, consequently, higher probability of choice based on their rating summarizations or their it explainability due to these ratings when ranking recommendations.
Learner models are a key element of many intelligent computer-based assessment and training approaches. Oftentimes, these models, the processes, and evidences are hidden from the learners and even teachers. Open learner models (OLM) are intended to make the underlying logic of a model as well as the evidences that contributed to a certain student appraisal transparent to its users. Furthermore, they are intended to support learning in a formative sense, aiming to help learners to self-monitor, plan, focus, and work independently as well as to communicate and negotiate appraisals with peers and teachers. We present a use case where Lea's OLM uses evidence from a speed-reading training application, to build the model. Learners use a persuasion function to interact with the system and maintain the OLM. The study showed that students are able to use it efficiently in order to make their model more accurate. We argue that persuasion, and other interactive maintenance features, are a strong approach to improve a transparent and accurate learner modelling, specifically when the underlying evidences are manifold and from multiple, partially unclear sources. Moreover, we found that the possibility to interact with the system on an individual basis and to have the possibility to intervene in the process is a strong means of making educational systems more personalized
Integrating information about the listener's cultural background when building music recommender systems has recently been identified as a means to improve recommendation quality. In this paper, we therefore propose a novel approach to jointly model users by the user's musical preferences and his/her cultural background. We describe the musical preferences of users by relying on the acoustic features of the songs the users have been listening to and characterize the cultural background of users by cultural and socio-economic features that we infer from the user's country. To evaluate the impact of the proposed user model on recommendation quality, we integrate the model into a culture-aware music recommender system. We show that incorporating both acoustic information of the tracks a user has listened to as well as the cultural background of users in the form of a music-cultural user model contributes to improved recommendation performance.
Recommender systems typically determine the items they should recommend by learning models of user-preferences. Most often, those preferences are modeled as static and independent of context. In real life however, users consider items in sequence: TV series are watched episode by episode and accessories are chosen after the main appliance. Unfortunately, since sequences are more complex to model, they are often not taken into account. We developed an efficient sequence-modeling approach based on Bayesian Variable-order Markov Models and combined it with an existing content-based system, the Ontology Filtering. We tested this approach through live evaluations on two e-commerce sites. It dramatically increased performance, more than doubling the CTR and strongly increasing recommendation-mediated sales. These tests also confirm that the technique works efficiently and reliably in a production setting.
In this paper, we propose a new factorization model that combines multi-view visual feature information with the implicit feedback data for prediction and ranking. The visual information is integrated into a collaborative filtering framework. The visual features of images are extracted by using a deep neural network. In order to conduct personalized recommendation better, the multi-view visual features are fused through user related weights. The user related weights reflect the personalized visual preference for items. They are different and independent between users. Experimental results show that our model with multi-view visual information achieves the better performance than models without or with only single-view visual information.
In order to sustain the user-base for a web service, it is important to know the return time of a user to the service. We propose a Bayesian point process, log Gaussian Cox process (LGCP), to model and predict return time of users. It allows encoding the prior domain knowledge and non-parametric estimation of latent intensity functions capturing user behaviour. We capture the similarities among the users in their return time by using a multi-task learning approach. We show the effectiveness of the proposed approaches on predicting the return time of users to last.fm music service.
In this paper we leverage the explicit user profile (relating to experience, knowledge, and self-regulation) to predict user engagement in active video watching. Data from two user studies for informal learning of presentation skills in a Higher Education context is used to develop and validate the prediction models. Our results show that these user characteristics can reasonably predict the overall engagement (inactive, passive and constructive learners). Our approach can be used to inform adaptive interventions that prevent disengagement and enhance the learning experience.
Studies have shown that contextual settings play an important role in users' decision processes of what to consume, but data supporting the investigation of context-aware recommender systems are scarce. In this paper we present a TV consumption dataset enriched with contextual information of viewing situations. The dataset is designed for studying the intrinsic complexity of TV watching activities, and hence we also evaluate the performance of predicting chosen genres given contextual settings, and compare the results to contextless predictions. The results suggest a significant improvement by including contextual features in the prediction.
Prediction of emotions is important for understanding human be-havior and modeling users in learning environments. In this paper,we present a deep multi-modal architecture for emotions predic-tion, which takes advantage of deep learning, user multimodal dataand the hierarchy of human memory. The architecture consists ofthe combination of Long Short-Term memory (LSTMs). One of thenovelty of our approach is that, we enhance the LSTM with anexplicit memory since in brain studies, the memory is often dividedinto two further main types: explicit (or declarative) memory andimplicit (or procedural) memory, the last one being the main pur-pose of LSTMs architectures. The resulting model has been testedon a public multi-modal dataset.
Eating activity monitoring using wearable sensors can potentially enable interventions based on eating speed for critical healthcare problems such as obesity or diabetes. We propose a novel methodology, IDEA that performs accurate eating action identification and provides feedback on eating speed. IDEA uses a single wristband with IMU sensors and functions without any manual intervention from the user. The F1 score for eating action identification was 0.92.
The majority of research works in the field of collaborative filtering recommender systems is based on the assumption that the input to the recommendation algorithms is a matrix containing user-item interactions. In reality, however, the input often is a sequence of various types of user-item interactions that are recorded over time and where we can have multiple data points per user-item pair. These sequential logs contain a variety of useful information that can be leveraged in the recommendation process, e.g., to predict the immediate next action of a user or to detect short-term trends in the community. In this tutorial we review what we call sequence-aware recommenders, i.e., approaches that aim to exploit the information in sequential interaction logs for a variety of different purposes. We in particular focus on sequential and session-based recommendation techniques and discuss algorithmic proposals as well as evaluation challenges.
We live in an age in which the dependency on technological tools is inescapable. At the same time, privacy-related issues are emerging in a way that we are at the breakpoint of losing control over our data. Information sharing by social-network, users can result in violations of privacy and security. For example, when a user is asking for a personalized service, he may find his contact details revealed, and may become the subject of harassment (cyber-bullying) or a potential victim of online deception or identity theft. Moreover, as Tim Berners-Lee stated, "The major players are making profit from our data. Therefore we lose out on the benefits we could realise if we had direct control over this data and choose when and with whom to share it". Today, more than ever, users need to keep control over their personal data when they ask for a personalized service. This tutorial, addresses how to reach this delicate balance between privacy and personalization.
Recommender systems for groups are becoming increasingly popular since many information needs originate from group and social activities, such as listening to music, watching movies, traveling, etc. There has been substantial progress on systems which recommend items to groups of users. However, many challenges remain. The goal of this tutorial is to introduce group recommendation and group modeling to the UMAP audience. First we will introduce the problem of making recommendations to groups and adapting to groups, and give an overview of the state-of-the art approaches to group recommendation. Next, we will also analyze more challenging topics, such as including different behavioral aspects into group modeling, and evaluation of group recommendations. Throughout, hands-on activities will be included. The tutorial will conclude with a summary of challenges and open issues.