Navigating educational data is a growing challenge. EDGE offers a fusion of large language models and knowledge graphs to enable intuitive, natural language-driven exploration, empowering educators, learners, and administrators with actionable insights for accessing and understanding educational ecosystems.

In an era where digital education platforms generate vast amounts of data, navigating and making sense of this information has become increasingly challenging for educators, learners, and administrators. EDGE (Educational Knowledge Graph Explorer) addresses these challenges by combining the power of large language models (LLMs) with the structural richness of knowledge graphs (KGs). Developed in team with Neda Afreen, Giacomo Balloccu, Gianni Fenu, Francesca Maridina Malloci, Mirko Marras, and Andrea Giovanni Martis, and presented in a demo paper at CIKM 2024, EDGE demonstrates how natural language interfaces can support access to educational knowledge bases.

EDGE

EDGE is a web-based conversational tool designed to facilitate the exploration of educational knowledge graphs. It allows users to interact with educational datasets using natural language queries, transforming these into structured Cypher queries, which are executed on a graph database. The results are then converted back into natural language responses, ensuring accessibility for users without technical expertise.

Features

1. Natural language queries. EDGE translates user queries into Cypher, the query language for graph databases, enabling intuitive access to complex data.
2. Domain-specific adaptation. Through a KG ontology mapping framework, EDGE converts educational datasets into semantically coherent structures tailored for natural language exploration.
3. Versatility: EDGE has been demonstrated in four educational contexts, including MOOCs, practice-oriented platforms, course marketplaces, and high-school transition platforms.

Architecture

EDGE employs a three-tier architecture:

1. Presentation layer. A responsive, React-based chat interface allows users to input natural language queries and view responses. Tailwind CSS enhances the interface’s aesthetics and cross-device compatibility.
2. Application layer. This layer handles query translation and response generation. Key components include:
   • Text2Cypher. Converts natural language queries into Cypher queries using OpenAI’s GPT-3.5 Turbo.
   • Cypher2Text. Transforms query results into natural language responses.
   • FastAPI. Provides API endpoints for seamless communication.
3. Data layer. KGs are stored and managed using Neo4j, leveraging its robust capabilities for handling graph-based data.

Use Cases

1. MOOCs

EDGE simplifies course discovery by allowing learners to ask, “Which courses cover the subject forecasting model?” It also assists educators and administrators in analyzing enrollment patterns and course performance.

2. Practice-oriented platforms

Platforms like Educoder benefit from EDGE’s ability to extract insights about course subjects, student demographics, and course reviews through simple queries.

3. Course marketplaces

EDGE enhances platforms like Udemy by enabling semantically complex searches, such as finding courses on math and science taught in Japanese.

4. High-school platforms

For high-school students exploring university options, EDGE provides a user-friendly way to navigate offerings, such as engineering bachelor’s programs.

Conclusions

EDGE represents a significant advancement in educational data exploration, combining the semantic depth of KGs with the intuitive interaction capabilities of LLMs.

In the future, we plan to improve EDGE by integrating additional features, such as multilingual support, automated KG ingestion, and multi-channel delivery (e.g., through Moodle or Telegram). Furthermore, comprehensive evaluations with diverse user groups will refine its accuracy and usability.

To learn more about EDGE or to view the demo, visit the project page.