Exploring increased personalization in automated cognitive tutors by adapting to learners' personal characteristics
Current cognitive tutors are generally effective in adapting learning activities to student knowledge, but they often fail to account for other characteristics, such as demographics, personality variables and preferences. The next step in Intelligent Tutoring Systems is to integrate these student characteristics with the current ability to direct instruction based on student knowledge.
An architecture, open-source tutoring engine, and plugins that are used to collect and analyze user data so that instructional systems can adapt to both learner knowledge and characteristics, including personality, mood, and demographics. The result is a framework that personalizes instructions more completely than current tutoring systems and supports best practices for instructional designers to better customize instructions to fit learner characteristics.
About the Project
The purpose of the Hyper-Personalized Intelligent Tutor (HPIT) project was to develop an architecture and proof of concept for a system that extends intelligent tutoring capabilities to allow for rapid integration of personalization via plug-ins. The HPIT architecture supports the activities of various tutoring systems, simultaneously, in order to manage and/or derive both cognitive and non-cognitive factors to enable personalization. This new technology aims to enable intelligent tutors to adapt to the personal characteristics of the individual learner and address the learner’s cognitive needs like a human tutor.
The core architecture of the HPIT system does not provide any specific tutoring capabilities. Instead, this schema-less, event-driven system manages the communication between various distributed tutoring systems. Tutoring systems interact with HPIT by sending messages, which consist of a named event and a data payload. Plugins are then used to listen to these events and perform actions as necessary, including submitting a response back to the HPIT architecture that will then be routed to the tutoring system. The HPIT architecture consists of several components: administrative tools, personalization/tutor engine to manage students’ skills, a data connector engine for setting and retrieving HPIT data, and a plug-in manager that provides a wide range of customization and extensibility to the architecture.
HPIT provides a robust framework for creating various plugin types to track or derive status values for cognitive and non-cognitive factors that could improve outcomes for students using a specific tutoring system. Several open-source plugins were developed to provide the infrastructure for a tutoring system to start when connecting to HPIT. These plugins include knowledge tracing (tracks what a student knows and does not know), automatic hints (generates hints and feedback on problems using data from students who have already attempted to solve the same problem), a general affect detector (boredom detection), and personalization (tracking any student data desired by a tutoring system). Each of these plugin components can be utilized from any tutoring system connected to the HPIT.
In May 2015, two pilot HPIT studies were conducted that engaged approximately 200 students:
A racecar-themed game app for iPad was designed to help students increase their math fluency with an architecture that allows for rapid prototyping of game features. The game was adaptive to both cognitive and non-cognitive factors, with multiple levels and a system for rewards. It was also configurable in both content and behavior, allowing educational researchers to quickly build a wide range of experiments. A survey to measure non-cognitive factors was deployed to the students before using the game app and HPIT directed individual student interventions based on those measured factors.
A handwriting-based equation solver for iPad was designed to mimic equation solving on paper. The Equation Solver app used the HPIT architecture to both evaluate student attributes and trace knowledge levels. It then applied adaptations to the user’s experience based on these two factors.
The successful use of HPIT by both the game and equation solver demonstrated that the architecture could successfully deploy individualized interventions to hundreds of students and proved the ability of HPIT to drive multi-dimensional, adaptive learning. The technology developed from the HPIT project has been incorporated into RoboTutor, one of the finalists selected for the Global Learning XPRIZE. The RoboTutor team used code developed from HPIT to implement an Android app, adapt the content to younger students, and translate to Swahili.
Although HPIT is no longer actively supported by the ADL Initiative, the technology is functional and can be applied to additional tutoring systems. Information made available in Github provides details on how to use the technology.
Ready to engage with HPIT? All of the documentation, terminology, API framework documentation, system requirements and more can be found on this designated GitHub page.
HPIT Python Client Libraries: Ready to create plugins for a HPIT project? Visit this GitHub site to learn how to install client libraries, run the test suite, register with HPIT, and follow a tutorial on creating plugins and plugin hooks.