Multi-Source Education Knowledge Graph Construction and Fusion for College Curricula

Zeju Li, Linya Cheng, Chunhong Zhang, Xinning Zhu, Hui Zhao

TL;DR

This paper introduces an automated framework for constructing and fusing knowledge graphs from diverse educational resources (textbooks, slides) for university curricula, specifically focusing on Electronic Information majors. The goal is to create a comprehensive, visual representation of course content to enhance student learning and aid in curriculum design.

What Problem Does This Paper Address?

University education, particularly in technical fields like Electronic Information, is characterized by a vast and complex landscape of specialized courses and learning materials. This makes it challenging for students to navigate, understand the interconnections between concepts, and achieve optimal learning outcomes. Existing approaches to knowledge representation in education are often resource-intensive, produce small-scale graphs, or focus on individual courses rather than the broader curriculum. This paper aims to bridge this gap by proposing an automated system to build and integrate knowledge graphs from multiple sources, providing a holistic view of the subject matter.

Methods

The authors propose a multi-stage framework for constructing and fusing knowledge graphs:

  1. Automated Knowledge Extraction and KG Construction:
    • Textbooks and course slides are converted into editable formats.
  2. Documents are segmented into hierarchical units ("Knowledge Units," "Knowledge Chapters," "Knowledge Blocks," "Knowledge Points") using NLP techniques like Named Entity Recognition (NER).
  3. These units are then imported into a Neo4j graph database to build large-scale knowledge graphs.
  4. Data Cleaning:
    • A language error correction model, enabled by deep learning, is used to identify and correct errors in extracted entities and attributes, addressing issues like text conversion errors, format encoding, and mixed languages.
  5. Knowledge Graph Fusion:
    • Entity Name Matching: String matching, literal similarity, and semantic similarity are used to identify entities referring to the same concept across different sources.
  6. Locating Coarse-grained Entities: A hierarchical approach is used to ensure that fused entities maintain their appropriate level of granularity.
  7. Taking Union of Attributes: Attributes from fused entities are combined to preserve as much information as possible.
  8. Preserving Original Edges: Existing relationships between entities are maintained after fusion to retain the topological structure.
  9. Fusion is performed both within a course (from different sources) and between courses.
  10. Auxiliary Information Statistics and Analysis:
    • Correlation Degree Calculation: A metric is developed to quantify the similarity between courses based on the overlap of knowledge concepts and their relationships. This involves calculating the proportion of equivalent relation edges between courses.
  11. Relationship Exploration: Techniques like spectral clustering are used to analyze the relationships between multiple courses, aiding in curriculum classification and design.
  12. Hot Knowledge Concept Identification: Frequency analysis and intersection of concepts across different sources are used to identify the most valuable and integral knowledge concepts.

Datasets

The paper utilizes educational materials from the Electronic Information major at Beijing University of Posts and Telecommunications, specifically:

  • Textbooks
  • Course slides
  • Course syllabi

These are treated as multi-source data for constructing the knowledge graphs. No specific public benchmark datasets are mentioned for training or evaluation, but the constructed knowledge graphs are substantial, covering over 60,000 entities and 80,000 triples.

Key Insights and Ideas

  • Automated KG Construction for Education: The paper demonstrates a practical, automated approach to building comprehensive knowledge graphs from diverse educational materials, overcoming the traditional labor-intensive nature of KG creation.
  • Multi-Source Fusion for Richer Representation: Fusing knowledge from textbooks, slides, and syllabi creates a more robust and complete representation of course content than any single source alone.
  • Quantifying Course Relationships: The developed correlation degree metric provides a data-driven way to understand how courses are related, which can inform curriculum design and student guidance.
  • Identifying Core Concepts: By analyzing concept frequency and intersection, the framework can highlight the most critical knowledge points within a curriculum, helping both students and instructors focus their efforts.
  • Hierarchical Knowledge Representation: The use of "Knowledge Units," "Chapters," "Blocks," and "Points" allows for a granular and structured representation of complex subject matter.

Key Results

  • The framework successfully constructs multi-source, large-scale knowledge graphs for hundreds of courses in Electronic Information, containing over 60,000 entities and 80,000 triples.
  • The proposed knowledge graph fusion method is effective in alleviating redundant knowledge nodes and improving data quality.
  • Analysis of course correlation reveals highly related courses, suggesting their potential as auxiliary learning resources. For example, "Communication Principles" shows high correlation with "Signals and Systems" and "Digital Circuits and Logic Design."
  • The identification of "most valuable" knowledge concepts within a course (e.g., 16 concepts for "Communication Principles" appearing across all sources) provides actionable insights for teaching and learning.

Limitations and Future Work

  • Generalizability: While the authors suggest their framework is applicable to other disciplines, the current implementation and evaluation are focused on the Electronic Information major. Further validation across different fields would be beneficial.
  • Data Quality: The effectiveness of the framework relies heavily on the quality of the input educational materials. Incomplete or poorly structured resources might still pose challenges.
  • Human Oversight: Although automated, the process of identifying and correcting errors, especially subtle ones, might still benefit from human review.
  • Future Work: The authors envision using the constructed knowledge graphs to enhance learning efficiency, explore new educational paradigms, and provide students with a comprehensive understanding of their field. Further development could involve building interactive learning platforms or personalized recommendation systems based on these graphs.

Summary updated

Abstract

The field of education has undergone a significant transformation due to the rapid advancements in Artificial Intelligence (AI). Among the various AI technologies, Knowledge Graphs (KGs) using Natural Language Processing (NLP) have emerged as powerful visualization tools for integrating multi-faceted information. In the context of university education, the availability of numerous specialized courses and complicated learning resources often leads to inferior learning outcomes for students. In this paper, we propose an automated framework for knowledge extraction, visual KG construction, and graph fusion, tailored for the major of Electronic Information. Furthermore, we perform data analysis to investigate the correlation degree and relationship between courses, rank hot knowledge concepts, and explore the intersection of courses. Our objective is to enhance the learning efficiency of students and to explore new educational paradigms enabled by AI. The proposed framework is expected to enable students to better understand and appreciate the intricacies of their field of study by providing them with a comprehensive understanding of the relationships between the various concepts and courses.
BibTeX
@inproceedings{li2023multi,
  author = {Zeju Li and Linya Cheng and Chunhong Zhang and Xinning Zhu and Hui Zhao},
  title = {{Multi-Source Education Knowledge Graph Construction and Fusion for College Curricula}},
  booktitle = {ICALT},
  year = {2023},
  url = {https://openreview.net/forum?id=ARGyzfruq2},
}
Figure