Avatar Behavior Analysis based on kNowledge inTegration (ABANT)
In this project, different methods and data integration techniques will be developed that will allow the capture of information about the avatar behaviour in virtual worlds, as well as the analysis of this behaviour and its visualization by means of 3D interfaces.
Different tools and techniques will be used for the acquisition and later analysis of the collected data. Information about the avatar behaviour includes data about their movements (spatial position), the actions they carry out (their attention focus, for example), and their interaction with the environment (document exchange, conversations, ….). Different techniques for the integration of heterogeneous data will be used, amongst them: (1) techniques based on information theory (data compression) and natural language processing that will be used for text analysis, (2) techniques for movement capture and eye tracking, (3) evolutive algorithms for the optimum adjustment of the weight assigned to each information type within the global measurement, and (4) techniques based on regular expressions for the extraction of relevant data in textual information.
As a result of the data integration, a representation of the avatar behaviour will be obtained. These behaviours will be analised by using automatic classification techniques in order to identify groups, or clusters, that indicate the degree of closeness between behaviour patterns. Interfaces to facilitate the visualisation of this analysis results will be designed and implemented.
The project will start from the results previously obtained by the research team in the Virtual World area to design new models and algorithms for knowledge integration that will allow the knowledge modelling and analysis in an automated and user-independent manner. Specifically, the proposal will contribute with the design of new techniques and computational methods to the following basic research areas:
- Characterisation and identification of avatar behaviour patterns.
- Integration of heterogeneous data sources in virtual worlds.
- Filtering of relevant data sources by means of evolutive regular expressions.
- Analysis and design of synthetic metrics starting from heterogeneous data obtained by means of techniques based on eye-gaze tracking (avatar focus point), human tracking (avatar’s position), and natural language processing combined with information theory (applied to text processing).
- Automatic optimization of the considered metrics (by using evolutive techniques).
- Design of automatic classifiers that allow visualizing the observed behavior pattern in the virtual world.
- Analysis of the behavior patterns within three specific domains: education, users with special needs, and social micro-nets.
All these contributions will be integrated within a specific platform (V-LeaF: TIN2008-02729-E/TIN) that will allow not only the evaluation and testing of the above-mentioned techniques but also the real deployment in real scenarios.