We present TutorChat, an intelligent chatbot conceived to be able support search and synthesis of information during a learning task accomplishment, in particular for dyslexic students. TutorChat is based on ChatGPT; it is able to support question/answer inter-activity of learners, and to generate concept maps on the topics at hand, with the possibility, beside analysis, to have such maps extended with additional sub-maps starting from a selected concept. We let TutorChat be used by a sample of dyslexic learners, coming from different educational levels. Then we collected encouraging sample’s feedback, through a questionnaire, about appreciation of the system’s services, and perception of the usefulness coming from its use.

We explore the classification and assessment capabilities of a selection of Open Source Small Language Models, on the specific task of evaluating learners' Descriptions of Algorithms. The algorithms are described in the framework of programming assignments, to which the learners in a class of Basics in Computer Programming have to answer. The task requires to 1) provide a program, in Python, to solve the assigned problem, 2) submit a description of the related algorithm, and 3) participate in a formative peer assessment session, over the submitted algorithms. Can a Language Model, be it small or large, produce an assessment for the algorithm descriptions? Rather than using any of the most famous, huge, and proprietary models, here we explore Small, Open Source based, Language Models, i.e. models that can be run on relatively small computers, and whose functions and training sources are provided openly. We produced a ground-truth evaluation of a large set of algorithm descriptions, taken from one year of use of the Q2A-II system. In this we used an 8-value scale, grading the usefulness of the description in a Peer Assessment session. Then we tested the agreement of the models assessments with such ground-truth. We also analysed whether a pre-emptive, automated, binary classification of the descriptions (as useless/useful for a Peer Assessment activity) would help the models to grade the usefulness of the description in a better way.

In recent years, also due to the covid-19 pandemic, the possibilities for distance learning have increased considerably, through web-based learning platforms, available on the Internet without space and time limits. As a result, the offer of courses and the number of enrolled students has grown exponentially. In order to be able to guarantee students a better learning support service, one of the proposals regards the intelligent Chatbots. These well known interactive applications are based mainly on machine or deep learning and in this paper we present Boulez, a system allowing the orchestration of a community of individual chatbots, each one with its algorithm and its private training dataset. We apply a technique called Federated Learning, where several individual chatbots, collaborate. In particular, here the approach is “centralized”, meaning that a main system orchestrates the collaboration of the federated systems. By addressing the communication inefficiencies and privacy issues of conventional federated learning, Boulez offers a more efficient and effective approach to chatbot interaction, ultimately leading to improved user experience. The paper presents the Boulez system, its operation principle, methods used, and potential benefits, along with a use case of its application.

Concept maps are graphic tools to organize, represent and share knowledge. In particular, a concept map can explicitly express the knowledge of a person or group, about a given domain of interest. Concept maps are used effectively to support learning of any topic, at any level: from Primary School to University, and to professional/vocational training, it can stimulate and unveil the occurrence of meaningful learning. In an educational context, having the possibility to compare Concept Maps coming from different students, also by means of an automated computation of map similarity, can reveal to be a great asset for a teacher. And this is so much more true when the number of students is very high, like in Massive Open Online Course. Here we propose a similarity measure based on two deep learning techniques that produce embeddings of the single structures that make up a concept map. We also report about a preliminary experiment, having encouraging results.

We discuss the problem of setting the best number of peers to which a given evaluation job should be assigned, in a Peer Assessment setting. The Peer Assessment is supposed to happen in a large scale class, such as in the case of Massive Open Online Courses. We use a dataset that simulate a large class (1000 students), based on Gaussian distributions of the Student Model features. Such features are related to the student's proficiency, and assessment capability. The number of peers assigned to the same evaluation job was controlled from 3 to 50 in 6 steps using 10-point scale. The abilities of participants were estimated using Item Response Theory. All parameters of IRT models, which is called as Generalized Partial Credit Model, such as "ability", "consistency", and "strictness", were estimated well using MCMC technique; their standard deviation errors gradually decrease with the number of peers. As a preliminary result of optimisation, an appropriate number of peers was 15 as comparing the stadardised errors across the conditions.

Massive Open Online Courses (MOOCs) are among the most popular online learning systems, with courses characterized by a very large number of attendees. Monitoring the students' learning process in a MOOC can be hard for the teacher, due to sheer numbers. Moreover, MOOC platforms produce large amounts of data related to the dynamics of the students community, which, if well used, can contribute to improving the educational offer. Here we present a web-based system, allowing the teacher to simulate a MOOC class of students, and experiment with it, by applying a pedagogic strategy based on Peer Assessment. The simulated students (peers) are supposed to produce answers to a question, and assessments of some other peers' answers, according to the models used to define the simulated class. Using our system the teacher can observe the dynamics of the simulated MOOC, based on a modified version of the K-NN algorithm, in line with the Teaching Analytics discipline. A first trial of the system produced promising results, showing its usefulness to support Teaching Analytics.

The users of the World Wide Web produce data continuously. This happens in varied areas such as trading on line, product ratings, support and use of services, and many more, comprising Distance Education. The ever increasing amount of such data can make analysis and extraction of meaningful information progressively harder, and sophisticated analysis techniques are to be used to extract added value from data. Many companies do collection and analysis of data with the purpose to develop their marketing strategies. In the field of education, and Distance Education in particular, data collected through online Learning Management Systems (LMSs) can provide a great resource, and a strong challenge, for the analysis of learning processes, the design of training paths, and the updating and personalization of learning environments. While, on the one hand, there is an increasing demand by educational institutions to measure, demonstrate, and improve the results achieved in distance learning, on the other hand the logic of traditional reporting included in LMS platforms does not satisfy that growing need. Learning Analytics is the answer to the need for optimization of learning through the techniques of analysis of data produced by learning processes, involving all stakeholders of the system. In this paper we show and discuss a brief state of the art of models of Learning Analytics presented in the literature.

The visualization of large graphs in interactive applications, specifically on small devices, can make harder to understand and analyze the displayed information. We show as simple topological properties of the graph can provide an efficient automatic computation of features which improves the "readibility" of a large graph by a proper selection of the displayed information. The connectivity (existence of a path) is a very intuitive structural property of a network; in this paper we propose an approach to the visualization of a network based on connectivity and related concepts as effective tools for visual analysis. In particular, given a root vertex r and a target vertex t, it is possible to check at a glance if there are some dominators, i.e., mandatory vertices that are on every path from r to t. Furthermore, using a recent graph algorithm from Georgiadis and Tarjan [19], [20], by selecting a target vertex it is possible to see two distinct paths from r to t: the paths are vertex-disjoints if there are no dominators from r to t, otherwise the paths have only the dominators in common. We conclude by presenting, as a relevant case study that motivated our work, as this approach improves a personalized eLearning application. In a framework, presented in [27], for dynamic configuration of paths of learning activities for both individual and group education, we can add visual analysis capabilities for both the final user/learner, and for the administrator of a repository.

We present a platform for Technology Enhanced Learning, allowing the learner to follow a path of game applications towards a learning objective. The path is determined by the learner, by selecting each time the next game of her choosing. The games are defined by teachers and experts, or even imported as web resources: they are associated to the system through suitable metadata, that express their pedagogical meaning. The system's interface allows the student to navigate the repository of learning games (organized as a graph) and see among them those that she can select to undertake, and those that are not yet affordable. Whether a game is affordable, at a given moment, is determined by comparing the Student Model with the game's specification. In conclusion, the path of learning activities followed by the learner is built interactively, by the learner, according to learner's choice and the system's pedagogical guidance. The system has not yet been experimented in a real class: we report about its design and implementation, and provide the reader with some simulated applications showing the system's behaviour.

Here we discuss a framework (OpenAnswer) providing support to the teacher's activity of grading answers to open ended questions. OpenAnswer implements a teacher mediated peer-evaluation approach: the marking results obtained from peer assessments are tuned by the grades explicitly assigned by the teacher, the teacher grades only a subset of the answers, suggested by the system. When a termination criterion is met, for the process managing the amount of teacher grading work, the remaining answers are automatically graded. A Bayesian Network is designed to represent the information related to students' models, peer assessments, and teacher's grading. The model parameters are many, here we report the results of investigations on a particularly tricky aspect of the framework, that is the modeling and optimization of the Conditional Probability Tables that are an important part of the Bayesian underlying model. In fact, they express the hypothesized relation between items of information that are relevant for evidence propagation through the network. Results suggest that this optimization improves OpenAnswer's performance, i.e. its capability to infer correct grades. We also show evidence of the influence of the teacher's assessing style on the grading process.

In this paper we experimentally investigate the influence of several factors on the final performance of an automatic grade prediction system based on teacher-mediated peer assessment. Experiments are carried out by OpenAnswer, a system designed for peer assessment of open-ended questions. It exploits a Bayesian Network to model the students' learning state and the propagation of information injected in the system by peer grades and by a (partial) grading from the teacher. The relevant variables are characterized by a probability distribution (PD) of their discrete values. We aim at analysing the influence of the initial set up of the PD of these variables on the ability of the system to predict a reliable grade for answers not yet graded by the teacher. We investigate here the influence of the initial choice of the PD for the student's knowledge (K), especially when we have no information on the class proficiency on the examined skills, and of the PD of the correctness of student's answers, conditioned by her knowledge, P(C|K). The latter is expressed through different Conditional Probability Tables (CPTs), in turn, to identify the one allowing to achieve the best final results. Moreover we test different strategies to map the final PD for the correctness (C) of an answer, namely the grade that will be returned to the student, onto a single discrete value. Copyright © 2016 by SCITEPRESS-Science and Technology Publications, Lda. All rights reserved.

Selecting and sequencing a set of Learning Objects (LOs) to build a course may turn out to be quite a challenging task. In this paper we focus on such an aspect, related to the verification and respect of the relationships of pedagogical dependence existing between two LOs added to a course (meaning that if a given LO has another one as "pre-requisite", then any sequencing of the LOs in the course will need to have the latter LO taken by the learners before of the former). In our approach the sequencing of LOs in the course can still be managed by the instructor, basing on her/his taste and preferences, yet s/he can also be helped by a set of suggestions, related to the pre-requisite relationships existing among the LOs selected for the course. Such suggestions (such relationships, in effect) can be computed automatically and provide the instructor with significant help and guidance. We show a light-weight formalization of the LO, and how it can be "represented" by a set of WikiPedia Pages ("topics"); then we show how such set of topics, together with a set of relevant hypotheses we previously defined, can help establish the dependence relationship existing between two LOs. In this endeavor we exploit the classification in categories available for the WikiPedia topics, and obtain interesting results for our framework, in terms of precision and recall of the dependence relationships.

The analysis of answers to open-ended questions provides greatly accurate assessment, being in turn demanding for the teacher. Here we show an approach exploiting peer assessment to partially relieve the teacher, and to provide information on the meta-cognitive ability of students of making correct evaluations on their peers. Open Answer handles a Bayesian model for each student, representing her/his learning state and judgment capability. The students' sub-networks are connected through peer-assessment. The process end up with a full set of grades for all students' answers, after the teacher had actually graded only part of them. We present experimental data and simulations aiming at identifying the best strategies to exploit the available information.

Open-ended questions are an important means to support analysis and assessment of students; they can be of extraordinary effectiveness for the assessment of higher cognitive levels of the Bloom's Taxonomy. On the other hand, assessing open answers (textual, freely shaped, answers to a question) is a hard task. In this paper we describe an approach to open answers evaluation based on the use of peer-assessment: in a social-collaborative e-learning setting implemented by the OpenAnswer web system, the students answer questions and rate others' (and may be own) answers, while the teacher marks a subset of the answers so to allow the system inferring the rest of the marks. The aim of our system is to ease the teacher's marking burden and allow for a more extensive use of open ended questionnaires in her/his teaching activity.

In this paper we present a comprehensive framework supporting the tasks of defining, retrieving, and importing Learning Objects (LOs) for personalized courses. It is partially implemented in a Moodle-based personalization system, where the instructional designer is guided through: 1) a theoretical specification of the needed LOs; 2) a retrieval function of actual LOs, by automatically querying standard-compliant repositories; 3) an analysis of such items, to import those selected by him, also adding metadata relevant to the personalization system, at hand. This work overcomes some well known shortcomings of the Moodle system in supporting retrieval of learning material in a personalization context.

The Open Answers module is designed to be integrated into the social collaborative reputation-based elearning system SocialX, to manage answers to open questions. In particular, its aim is to support personal evaluation of skills and knowledge of students, involved in peer-assessment-based learning activities, while mitigating the workload imposed upon the teacher, for analysis and correction of the answers. In brief, 1) students do answer open questions, to be evaluated by peers and teacher; 2) students peer-evaluate each other's answers; 3) the teacher grades only a subset of the whole answers corpus; 4) the system infers the assessment for the remaining answers, by exploiting the relations established through the web of the students' peer-assessments of those answers, and the personal evaluation maintained for each student. The peer-assessment data are analyzed through a constraint-logic-based model of student's possible behaviors, obtaining a (possibly big!) set of hypotheses on the answers' correctness. The teacher is proposed with a minimum set of answers to grade: by such grading (s)he helps narrowing the set of hypotheses. Testing of the constraint-logic-based analysis engine is ongoing by means of simulation devices that generate suitable sets of students' behaviors.

We investigate the integration of LECOMPS, a web-based e-learning environment for the automated construction and adaptive delivery of learning paths, and SOCIALX, a web-based system for shared e-learning activities, which exploits a reputation system to provide feedback to its participants. Our overall goal is the integration of personalized and collaborative learning to support the Vygotskij's educational theory of proximal development. Therefore we propose a two-way tunneling strategy: the LECOMPS student model is used to select the set of social activities (met in SOCIALX) according to the present individual learner state of knowledge, on the other hand, the solution of exercises, and the associated reputation derived in SOCIALX, is used to update the LECOMPS student model. In particular, we present a mapping between the student model and the definition of Vygotskij's concepts of Autonomous Problem Solving and Proximal Development regions, with the aim to provide the learner with better guidance during the taking of the course.