OBJECTIVE: To develop and evaluate a time-aware data mining pipeline that integrates accelerometry-based Physical Activity (PA), static dietary intake, and sociodemographic factors to predict Healthy Weight Status (HWS), and identify features with predictive importance for HWS. METHODS: We propose TimePAD, Time-Based Physical Activity and Dietary Intake data mining, a three-stage prediction pipeline that integrates time-based PA representations, static dietary intake, and sociodemographic variables to predict HWS categories. Stage 1 derives intensity-specific hourly PA sequences and learns time-of-day PA embeddings using a Transformer encoder trained with masked Self-Supervised Learning (SSL) reconstruction. Stage 2 combines the learnt PA embeddings with static dietary variables and participant characteristics derived from the Food Frequency Questionnaire (FFQ). Stage 3 trains prediction models and ranks predictors by their predictive importance for HWS. TimePAD was evaluated on a real-world dataset of 206 adolescents (10-16 years) with 7-day continuous wrist-worn accelerometer data, FFQ records, and sociodemographic attributes, using 10-fold cross-validation and comparison to an ARIMA-based feature engineering baseline. RESULTS: TimePAD achieves an accuracy of 82.90% and an F1-score of 67.92% for HWS prediction, outperforming the best baseline (ARIMA-based PA features + diet, 77.51% accuracy). Across experiments, time-of-day light PA (LPA) features were consistently ranked as important predictors. Other key features include time-of-day Moderate-to-Vigorous Activity (MVPA) and Sedentary (SED) levels, Tanner stage, total weekly sleep time, age (in months), weekly fruit intake, weekly vegetable and legume intake, and intake of sugar-sweetened beverages. DISCUSSION: TimePAD contributes a pipeline for learning from the time-of-day structure in wearable PA time series and integrating it with static dietary and contextual data for prediction and feature analysis. The findings suggest that LPA is likely to have a significant association with HWS, calling for further attention and investigation to better understand the role of LPA in overall health outcomes. This illustrates the potential benefits of TimePAD in modelling PA with dietary intake context in shaping healthy behaviours.

Multimodal annotations provide important cues for understanding how a conversation proceeded, particularly in relation to affective factors. In this paper, we extend the automated conversation annotation system MONAH withpitchandvolumeannotations to encode new affective cues, making MONAHv3 the state-of-the-art automatic annotation system in terms of the number of automatically annotated aspects. MONAHv3 provides an automated solution that, while not a direct equivalent, offers performance that is competitive with the manually produced Jefferson transcription system. In automatic evaluations, the additions significantly improve supervised learning in ten out of eighteen experiments. In human evaluations of emotion recognition, the additions significantly outperformed the Jefferson transcription system. Usability studies further showed that MONAHv3 is much more user-friendly than Jefferson-style transcripts. Lastly, in human evaluations of paralinguistic cues (e.g., tone and volume), MONAHv3 achieved performance comparable to the Jefferson system, while remaining more competitive in kinesics (e.g., describing actions). This distinction between automated and manual systems is critical for appreciating the scalability and accessibility that MONAHv3 offers.

Obesity is a global health concern associated with various demographic and lifestyle factors including physical activity (PA). Research studies generally used self-reported PA data or, when accelerometer-based activity trackers were used, highly aggregated data (e.g., daily average). This suggests that the rich potential of detailed activity tracker data is largely under-exploited and that deeper analyses may help better understand such relationships. This is particularly true in children and adolescents who are distinct and engage more in bursts of PA. This article presents ABIPA, a machine learning-based methodology that integrates various aspects of accelerometer-based PA data into weight status prediction for adolescents. We propose a method to derive features regarding the structure of different PA time series using Auto-Regressive Integrated Moving Average (ARIMA). The ARIMA-based PA features are combined with other individual attributes to predict weight status and the importance of these features is further unveiled. We apply ABIPA to a dataset about young adolescents (N = 206) containing, for each participant, a 7-day continuous accelerometer dataset (60 Hz, GENEActiv tracker from ActivInsights) and a range of their socio-demographic, anthropometric, and lifestyle information. The results indicate that our method provides a practical approach for integrating accelerometer-based PA patterns into weight status prediction and paves the way for validating their importance in understanding obesity factors.

Understanding the relationship between technology and wellbeing is important in order to raise awareness and to improve interaction designs with digital technologies. Most studies used the time spent and frequency information of digital technology usage, very few explored the sequences and the patterns of how the activity occurs. We introduce the concept of “digital context,” a representation of activity data occurring in a short time-window. Using data from our study, we determined whether: (1) there are digital context patterns that are more frequent in a particular mood compared to other moods; and (2) in the case such patterns exist, whether they can be used to improve the performance of mood prediction models. Our results showed that a mood prediction model that include digital context features yielded an accuracy of 77.8%, which is an improvement compared with the models proposed in past studies.

Abstract Machine learning on trees has been mostly focused on trees as input. Much less research has investigated trees as output, which has many applications, such as molecule optimization for drug discovery, or hint generation for intelligent tutoring systems. In this work, we propose a novel autoencoder approach, called recursive tree grammar autoencoder (RTG-AE), which encodes trees via a bottom-up parser and decodes trees via a tree grammar, both learned via recursive neural networks that minimize the variational autoencoder loss. The resulting encoder and decoder can then be utilized in subsequent tasks, such as optimization and time series prediction. RTG-AEs are the first model to combine three features: recursive processing, grammatical knowledge, and deep learning. Our key message is that this unique combination of all three features outperforms models which combine any two of the three. Experimentally, we show that RTG-AE improves the autoencoding error, training time, and optimization score on synthetic as well as real datasets compared to four baselines. We further prove that RTG-AEs parse and generate trees in linear time and are expressive enough to handle all regular tree grammars.

Digital technology influences behaviours, moods and wellbeing. The relationships are complex, but users are increasingly interested in finding how to balance a digital life with psychological wellbeing. We present an approach for investigating the relationship between lifestyle aspects and digital technology usage patterns that combines MindGauge, a mobile app enabling users collect and analyse their moods and behaviours, with a productivity tool (RescueTime). We then report a 16-month study in which we collected computer and smartphone usage and self-reports from 72 participants. We present methods for analysing the relationship between productivity, task-switching, mood and lifestyle, and more specifically how digital technology usage associates with productivity and task-switching. Our study also investigates how lifestyle aspects (sleep quality, physical activity, workload, social interaction and alcoholic drink consumption) relate to mood, task-switching and productivity. Results show that more frequent task-switching is associated with negative moods. A few lifestyle aspects, such as sleep quality and physical activity, had a significant relationship with positive moods. We also contribute a mood detection model that utilise both digital footprints and lifestyle contexts, yielding an accuracy of 87 percent. The study provides evidence that such methods can be used to understand the impact of technology on wellbeing.

Joshua Y. Kim, Kalina Yacef, Greyson Kim, Chunfeng Liu, Rafael Calvo, Silas Taylor. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021.

Tree data occurs in many forms, such as computer programs, chemical molecules, or natural language. Unfortunately, the non-vectorial and discrete nature of trees makes it challenging to construct functions with tree-formed output, complicating tasks such as optimization or time series prediction. Autoencoders address this challenge by mapping trees to a vectorial latent space, where tasks are easier to solve, and then mapping the solution back to a tree structure. However, existing autoencoding approaches for tree data fail to take the specific grammatical structure of tree domains into account and rely on deep learning, thus requiring large training datasets and long training times. In this paper, we propose tree echo state autoencoders (TES-AE), which are guided by a tree grammar and can be trained within seconds by virtue of reservoir computing. In our evaluation on three datasets, we demonstrate that our proposed approach is not only much faster than a state-of-the-art deep learning autoencoding approach (D-VAE) but also has less autoencoding error if little data and time is given.

Learning to collaborate effectively requires practice, awareness of group dynamics, and reflection; often it benefits from coaching by an expert facilitator. However, in physical spaces it is not always easy to provide teams with evidence to support collaboration. Emerging technology provides a promising opportunity to make collocated collaboration visible by harnessing data about interactions and then mining and visualizing it. These collocated collaboration analytics can help researchers, designers, and users to understand the complexity of collaboration and to find ways they can support collaboration. This article introduces and motivates a set of principles for mining collocated collaboration data and draws attention to trade-offs that may need to be negotiated en route. We integrate Data Science principles and techniques with the advances in interactive surface devices and sensing technologies. We draw on a 7-year research program that has involved the analysis of six group situations in collocated settings with more than 500 users and a variety of surface technologies, tasks, grouping structures, and domains. The contribution of the article includes the key insights and themes that we have identified and summarized in a set of principles and dilemmas that can inform design of future collocated collaboration analytics innovations.

The increasing progress in ubiquitous technology makes it easier and cheaper to track students' physical actions unobtrusively, making it possible to consider such data for supporting research, educator interventions, and provision of feedback to students. In this paper, we reflect on the underexplored, yet important area of learning analytics applied to physical/motor learning tasks and to the physicality aspects of `traditional' intellectual tasks that often occur in physical learning spaces. Based on Distributed Cognition theory, the concept of Internet of Things and multimodal learning analytics, this paper introduces a theoretical perspective for bringing learning analytics into physical spaces. We present three prototypes that serve to illustrate the potential of physical analytics for teaching and learning. These studies illustrate advances in proximity, motion and location analytics in collaborative learning, dance education and healthcare training.

A large number of learning tools offering some sort of personalisation features rely mainly on the analysis of logged interactions between students and particular user interfaces. Much less attention has been given to the analysis of physical aspects so often present in 'traditional' intellectual tasks, although these are both important in the full development of a life-long learner. This paper (1) discusses existing literature focused on supporting learning using proximity and location analytics and sensors, and, based on this, (2) illustrates the feasibility and potential of these analytics for teaching and learning through an study in the context of proximity and location analytics in a team-based health simulation classroom.

Motor skill learning is an area where wearable technology and user modelling can be synergistically combined for providing support. In this paper, we explore how a simple accelerometer sensor can be used to capture motion data associated with critical aspects of learning in the context of social dancing. We developed a prototype mobile app that tracks students' motion data whilst they practise dance exercises. This paper describes a set of features, such as rhythm duration, consistency and body motion, which can be automatically tracked and included into a dance student model. These dancing features can be presented back to the students as feedback, in the form of i) summaries, ii) visualisations or iii) narratives. We illustrate the feasibility and potential of modelling these features through a study with beginner students taking dance classes during three weeks.

Designing, deploying and validating learning analytics tools for instructors or students is a challenge requiring techniques and methods from different disciplines, such as software engineering, human-computer interaction, educational design and psychology. Whilst each of these disciplines has consolidated design methodologies, there is a need for more specific methodological frameworks within the cross-disciplinary space defined by learning analytics. In particular there is no systematic workflow for producing learning analytics tools that are both technologically feasible and truly underpin the learning experience. In this paper, we present the LATUX workflow, a five-stage workflow to design, deploy and validate awareness tools in technology-enabled learning environments. LATUX is grounded on a well-established design process for creating, testing and re-designing user interfaces. We extend this process by integrating the pedagogical requirements to generate visual analytics to inform instructors' pedagogical decisions or intervention strategies. The workflow is illustrated with a case study in which collaborative activities were deployed in a real classroom.

The use of cloud computing writing tools, such as Google Docs, by students to write collaboratively provides unprecedented data about the progress of writing. This data can be exploited to gain insights on how learners' collaborative activities, ideas and concepts are developed during the process of writing. Ultimately, it can also be used to provide support to improve the quality of the written documents and the writing skills of learners involved. In this paper, we propose three visualisation approaches and their underlying techniques for analysing writing processes used in a document written by a group of authors: (1) the revision map, which summarises the text edits made at the paragraph level, over the time of writing. (2) the topic evolution chart, which uses probabilistic topic models, especially Latent Dirichlet Allocation (LDA) and its extension, DiffLDA, to extract topics and follow their evolution during the writing process. (3) the topic-based collaboration network, which allows a deeper analysis of topics in relation to author contribution and collaboration, using our novel algorithm DiffATM in conjunction with a DiffLDA-related technique. These models are evaluated to examine whether these automatically discovered topics accurately describe the evolution of writing processes. We illustrate how these visualisations are used with real documents written by groups of graduate students.

Learning about subject matter and about writing by collaboratively authoring an electronic document is an important variant of computer-supported collaborative learning. Collaborative writing is particularly often practiced in Higher Education. Our research has the goal to develop comprehensive software support tools for collaborative discipline-based writing, and to study how the team writing process is affected by the use of these tools. This paper describes initial tool developments that integrate computational document analysis methods with process mining methods into a comprehensive writing environment and reports first experiences gained in a undergraduate engineering course.

Reciprocal recommender is a class of recommender systems that is important for tasks where people are both the subject and the object of the recommendation; one such task is online dating. We have implemented RECON, a reciprocal recommender for online dating, and we have evaluated it on a major dating website. Results show an improved success rate for recommendations that consider reciprocity in comparison to recommendations that only consider the preferences of the users receiving the recommendations.

Group work is widespread in education. The growing use of online tools supporting group work generates huge amounts of data. We aim to exploit this data to support mirroring: presenting useful high-level views of information about the group, together with desired patterns characterizing the behavior of strong groups. The goal is to enable the groups and their facilitators to see relevant aspects of the group's operation and provide feedback if these are more likely to be associated with positive or negative outcomes and indicate where the problems are. We explore how useful mirror information can be extracted via a theory-driven approach and a range of clustering and sequential pattern mining. The context is a senior software development project where students use the collaboration tool TRAC. We extract patterns distinguishing the better from the weaker groups and get insights in the success factors. The results point to the importance of leadership and group interaction, and give promising indications if they are occurring. Patterns indicating good individual practices were also identified. We found that some key measures can be mined from early data. The results are promising for advising groups at the start and early identification of effective and poor practices, in time for remediation.

The XMLTutor is an authoring tool that generically delivers personalised teaching on an arbitrary domain, which must be specified in Extensible Markup Language (XML). The tool was built to explore the consequences of using XML as the main ontological structure in an intelligent tutoring system (ITS).

Traditionally, intelligent tutoring systems (ITS) are dedicated to learners. They help them learn at their own pace, following a curriculum tailored to their individual needs and receiving individualised feedback. Intelligent teaching assistant systems (ITAs) are dedicated both to learners and teachers. Their aim is to facilitate the whole teaching/learning process by helping the teacher as well as the student. There has been an increasing interest to integrate the teacher as an end-user of the ITS. This paper presents the characteristics and discusses the architecture of ITAs. We illustrate the concept of ITA with two systems that we designed.

No abstract available.