Utilizando análise semântica para minerar implicações significantes em mapas conceituais

Abstract

Concept maps are forms of graphic representation that establish relationships between concepts. They can be used in education in different situations and purposes: as a learning resource, cognitive representation, means of evaluation, instructional organization and knowledge sharing. The knowledge evidenced in a conceptual map involves implications between meanings. Meanings are everything that can be said about an object, such as a description of its properties, as well as everything that we can observe in it. In addition, a meaning or implication also refers to everything we can think of objects (classifying them, establishing some kind of relationship, among others). In the classroom, concept maps can be applied as a way of mediating teaching-learning, stimulating meaningful learning. There are many challenges for using concept maps to be more effective in measuring learning and evaluating the learner’s cognitive processes and their interactions with other learning participants. The assessment of learning is a very complex task, especially when the objective is to automate the assessment process, which demands, among other obligations, a formalization of knowledge representation structures. Furthermore, the evaluation of a concept map becomes more complex when the author does not represent his knowledge, but the knowledge expressed in a text or in a map made by someone else. In this context, we carried out a literature review between the years 2015 and 2020 on the development of technological approaches that help or automate the process of evaluating concept maps in the educational environment, through the use of significant implications. We also sought to identify limitations and gather the best features of related works to propose our approach. Among the limitations found, we found that none of the approaches found applied semantic analysis in the evaluation of concept maps. The application of semantic analysis makes it possible to achieve a deeper perception of the supposed knowledge represented in the concept maps, seeking to transform evidence into evidence of the learning, in fact, carried out by the learner, helping the teacher in the evaluation of the concept maps. Furthermore, an activity that involves concept maps in a classroom would result in the construction of different maps, and with that, the evaluation would become costly and of great cognitive effort for the teacher. Therefore, the automation of this process is of great value. In order to develop a computational architecture capable of performing semantic analysis of concept maps through the significant implications defined by Piaget, this research resulted in a new API, called MAPimp 2.0, which aims to provide a deeper semantic dimension in the analysis. of significant implications and to extract more accurate information about the representation of knowledge and an individual’s understanding from the representation of their map. This new version brings the implementation of syntactic-semantic validations as a novelty. The architecture is capable of processing concept maps written in English and Portuguese. The development took place using natural language processing techniques and pre-trained neural network models with billions of texts for word prediction and semantic similarity calculation. In order to obtain a quantitative and qualitative analysis, the conceptual architecture was applied in a classroom environment, presenting satisfactory results.