Abstract:

Cultural aspects and linguistic diversity represent significant challenges in the production of digital educational materials that are truly inclusive and context-aware. Although digital textbooks expand access to information and integrate multimedia resources, many still present standardized content that does not adequately consider regional differences in language, culture, consumption habits, and everyday references experienced by students. This limitation may reduce engagement, weaken students’ identification with examples, and hinder meaningful knowledge construction.

In this context, this study proposes the development of a Hierarchical Multi-Agent System (HMAS) capable of extracting, analyzing, and adapting multimedia educational resources to generate culturally contextualized digital textbooks tailored to different regions of a target country. Brazil's National Textbook Program (PNLD) is adopted as the case study due to its strong cultural and linguistic diversity.

We develop a HMAS architecture responsible for different stages of educational content adaptation. At the lower level, multimedia extraction agents collect and analyze resources from educational repositories, including texts, images, videos, infographics, and interactive learning objects. These agents employ techniques such as Natural Language Processing, computer vision, and semantic analysis to identify key concepts, pedagogical contexts, and metadata associated with educational materials.

At the intermediate level, semantic and cultural interpretation agents process the extracted resources and perform sociocultural contextualization of the educational content. These agents map cultural elements, terminology, and examples present in the original material by comparing them with knowledge bases representing Brazilian regional characteristics, such as linguistic expressions, cultural references, consumption habits, and everyday examples relevant to different states or regions. To support this process, the system relies on cultural ontologies and regional dictionaries capable of capturing lexical and semantic variations across Brazilian regions.

At the highest level, pedagogical recommendation and adaptation agents transform multimedia resources into chapters or sections of culturally adapted digital textbooks. These agents apply pedagogical rules and content-based recommendation models to select examples, images, and narratives aligned with the cultural context of the target region. Adaptations may include replacing product examples, adjusting vocabulary in explanations, and incorporating local cultural references to facilitate comprehension.

The generation of the digital textbook is coordinated by an aggregation agent responsible for integrating the adapted resources into a pedagogically structured format composed of modules, sections, and interactive activities. The result is a dynamic educational material that preserves curricular objectives while presenting examples, vocabulary, and visual elements aligned with the regional context of the target audience.

The expected results suggest that HMAS can contribute to the cultural personalization of digital textbooks in multicultural environments. Furthermore, the approach highlights the potential of integrating artificial intelligence, multimedia content mining, and cultural knowledge representation to support the creation of more inclusive, contextualized, and pedagogically effective educational materials.

Keywords:

Adaptive Textbooks, Multi-Agent Systems, Cultural knowledge representation, Educational Content Adaptation, Multimedia Educational Resources.