AFFITS is a research project supported by Polish-Norwegian Research Programme operated by the National Centre for Research and Development under the Norwegian Financial Mechanism 2009–2016 in the frame of Project Contract no. Pol-Nor/209260/108/2015.

The AFFITS project aims at developing new methods and tools for affect recognition, representation and modeling for Intelligent Tutoring Systems (ITS).

The main research question addressed by AFFITS is: „How to effectively monitor, represent and interpret affect in Intelligent Tutoring Systems to facilitate emotional states that support learning process?”. In Intelligent Tutoring Systems user emotion recognition aims at distinguishing emotional states, that support or disturb learning processes. The mechanism provided for ITS must deal with uncertainty and fuzzy nature of emotion at two stages: emotion recognition and in control mechanisms. In this research an affectawareness framework is proposed that comprises: user emotional state elicitation (recognition), user affect interpretation and analysis, affect-aware intervention and control. In this framework emotional state elicitation and interpretation can be considered as testing a hypothesis: ‘user emotional state requires intervention’ against null hypothesis: ‘user emotional state does not require intervention’. In that context type 1 error indicates that a system recognizes a non-existent emotional state and makes unnecessary intervention, while type 2 error means that a counterproductive emotional state was not recognized and the system fails to make intervention when required. Although both errors should be minimized, it is sometimes impossible to minimize both at the same time and in that case the choice of the optimization purpose must be based on the real-life consequences of an error. In Intelligent Tutoring Systems it would be better when a system sometimes did not make required intervention than when it made them unnecessarily and perhaps instantly, which would be very disturbing.

Some detailed research questions include: How to monitor user emotions in e-learning environment? What is the best way of non-disturbing monitoring of computer user emotional states? How precise and reliable are measurements of computer user emotional states in ITS environment? How can affectawareness be built into Intelligent Tutoring Systems control mechanisms? How to reduce risk of unnecessary and disturbing intervention of intelligent applications that deal with user emotional states?

Due to time limits of the AFFITS project, the scope of the research was restrained to verification of the following hypothesis:

  • Research hypothesis 1: It is possible to propose a set of algorithms for non-disturbing monitoring of emotional states in ITS environment and to propose a set of criteria for choosing algorithms for affect-aware software.
  • Research hypothesis 2: Algorithm based on text processing, affect-tagged Polish dictionary and peripheral devices usage patterns allows to effectively recognize user emotional states in conversational Intelligent Tutoring Systems.
  • Research hypothesis 3: Reliable emotional state hypothesis can be obtained by the combination of multiple results from multimodal emotion recognition algorithms and the trustworthiness model, that deals with uncertainty of emotion recognition.
  • Research hypothesis 4: Construction of ITS based on consistent emotional state representation and control including uncertainty model (affect-awareness framework) allows to reduce the risk of unnecessary and disturbing application intervention.

The AFFITS project aims not only at providing some new methods in affect recognition, interpretation and expression but also combining them into one unified framework, providing that partial results work with each other to support affect-awareness in Intelligent Tutoring Systems. The proposed affect-awareness framework is not intended to be a complete affect-processing engine, but rather an extendable (modular and maintainable) structure to be applied in many domains and be modifiable with the progress of the affective learning domain, with new algorithms easily detached, exchanged or optimized.

Enhancement of software with affect-awareness mechanism is an ambitious task, as even psychologists tend to debate on the nature and causality of emotions, but it is a valuable effort, as affect-aware ITS are expected to address fluctuation in motivation and concentration and to deal with such emotional states like boredom, stuck or frustration. As a result drop-out rate may lower down, learning process effectiveness may increase, and a learner may be more motivated to follow his learning track and undertake new educational challenges.