Catálogo de publicaciones - libros

In tutorial systems, effective progress in teaching the problem-solving target is frequently hindered by expressive sloppiness and low-level errors made by the student, especially in conventionalized expressions such as formulas. In order to improve the effectiveness of tutorial systems in teaching higher-level skills, we present a fault-tolerant formula interpreter that aims at finding plausibly intended, formally correct specifications from student statements containing formal inaccuracies. The interpretation consists of local changes based on categorization of errors, a fault-tolerant structure building, and testing contextually-motivated alternations. The error interpretation component is intended to enhance the analysis component of a tutorial system that teaches mathematical proving skills.

The provision of intelligent, user-adaptive, and effective feedback requires human tutors to exploit their expert knowledge about the domain of instruction, and to diagnose students’ actions through a potentially huge space of possible solutions and misconceptions. Designers and developers of intelligent tutoring systems strive to simulate good human tutors, and to replicate their reasoning and diagnosis capabilities as well as their pedagogical expertise. This is a huge undertaking because it requires an adequate acquisition, formalisation, and operationalisation of material that supports reasoning, diagnosis, and natural interaction with the learner. In this paper, we describe SLOPERT, a glass-box reasoner and diagnoser for symbolic differentiation. Its expert task model, which is enriched with buggy rules, has been informed by an analysis of human-human tutorial dialogues. SLOPERT can provide natural step-by-step solutions for any given problem as well as diagnosis support for typical student errors. SLOPERT’s capabilities thus support the generation of natural problem-solving hints and scaffolding help.

Students often use available help facilities in an unproductive fashion. To improve students’ help-seeking behavior we built the Help Tutor – a domain-independent agent that can be added as an adjunct to Cognitive Tutors. Rather than making help-seeking decisions for the students, the Help Tutor teaches better help-seeking skills by tracing students actions on a (meta)cognitive help-seeking model and giving students appropriate feedback. In a classroom evaluation the Help Tutor captured help-seeking errors that were associated with poorer learning and with poorer declarative and procedural knowledge of help seeking. Also, students performed less help-seeking errors while working with the Help Tutor. However, we did not find evidence that they learned the intended help-seeking skills, or learned the domain knowledge better. A new version of the tutor that includes a self-assessment component and explicit help-seeking instruction, complementary to the metacognitive feedback, is now being evaluated.

A major issue in Intelligent Tutoring Systems is off-task student behavior, especially performance-based gaming, where students systematically exploit tutor behavior in order to advance through a curriculum quickly and easily, with as little active thought directed at the educational content as possible. The goal of this research was to explore the phenomena of off-task gaming behavior within the system. Machine-learned gaming-detection models were developed to investigate underlying factors related to gaming, and an analysis of gaming within the Assistments system was conducted to compare some of the findings of prior studies.

It has been found in recent years that many students who use intelligent tutoring systems game the system, attempting to succeed in the educational environment by exploiting properties of the system rather than by learning the material and trying to use that knowledge to answer correctly. In this paper, we introduce a system which gives a gaming student supplementary exercises focused on exactly the material the student bypassed by gaming, and which also expresses negative emotion to gaming students through an animated agent. Students using this system engage in less gaming, and students who receive many supplemental exercises have considerably better learning than is associated with gaming in the control condition or prior studies.

In recent years, a number of systems have been developed to detect differences in how students choose to use intelligent tutoring systems, and the attitudes and goals which underlie these decisions. These systems, when trained using data from human observations and questionnaires, can detect specific behaviors and attitudes with high accuracy. However, such data is time-consuming to collect, especially across an entire tutor curriculum. Therefore, to deploy a detector of behaviors or attitudes across an entire tutor curriculum, the detector must be able to transfer to a new tutor lesson without being re-trained using data from that lesson. In this paper, we present evidence that detectors of gaming the system can transfer to new lessons without re-training, and that training detectors with data from multiple lessons improves generalization, beyond just the gains from training with additional data.

This paper introduces a domain-independent open learner model with multiple simple views on individual learner model data. Learners can also compare their knowledge level to their peer group, and to instructor expectations for different stages of their course. The aim is to help learners identify their knowledge, difficulties and misconceptions; prompt reflection on their knowledge and learning; and facilitate planning. We present a study of OLMlets in 4 university courses, with 114 users making over 20000 learner model inspections.

We present a student’s modeling process in algebra. This work is situated in the framework of the deployment of the Aplusix system, a learning environment for algebra. The process has two phases. The first phase is a local diagnosis where a student’s transformation of an expression A into an expression B is diagnosed with a sequence of rewriting rules. A library of correct and incorrect rules has been built for that purpose. The second phase uses conceptions for modeling students more globally. Conceptions are attributed to students according to a mechanism using the local diagnoses as input. This modeling process has been applied to data (log files) gathered in France and Brazil with 13-16 years old students who used the Aplusix learning environment. The results are described and discussed.

This paper explores the feasibility of using conversational agents, or chatbots, in negotiated learner modelling. This approach aims to combine the motivational, intuitive and domain-independent benefits of natural language dialogue using a chatbot, with the opportunities for learner reflection and increased model accuracy that can be achieved through negotiation of the learner model contents. A Wizard-of-Oz paradigm allowed investigation into the interactions between learners and their learner model in order to highlight key issues for the design of a chatbot for this purpose. Users appreciated interacting with a chatbot, and found it useable and an aid to negotiation. The study suggested many avenues for future investigation of the role of conversational agents in facilitating user-system dialogue about learner understanding.

This paper describes research to analyze students’ initial skill level and to predict their hidden characteristics while working with an intelligent tutor. Based only on pre-test problems, a learned network was able to evaluate a students mastery of twelve geometry skills. This model will be used online by an Intelligent Tutoring System to dynamically determine a policy for individualizing selection of problems/hints, based on a students learning needs. Using Expectation Maximization, we learned the hidden parameters of several Bayesian networks that linked observed student actions with inferences about unobserved features. Bayesian Information Criterion was used to evaluate different skill models. The contribution of this work includes learning the parameters of the best network, whereas in previous work, the structure of a student model was fixed.