1. What is a Knowledge Graph
A knowledge graph (Knowledge Graph) is a series of various graphics that display the development process and structural relationships of knowledge. It uses visualization technology to describe knowledge resources and their carriers, excavating, analyzing, constructing, drawing, and displaying knowledge and their interconnections. Knowledge graphs integrate theories and methods from disciplines like artificial intelligence, applied mathematics, graphics, and information visualization technology, using visual graphs to vividly present the core structure, development history, cutting-edge fields, and overall knowledge architecture, achieving multidisciplinary integration in modern theory.
The construction of knowledge graphs lays the foundation for the current evolution of artificial intelligence from perceptual intelligence to cognitive intelligence. With technological advancements, artificial intelligence has reached or surpassed human levels in perceptual intelligence fields such as “listening, speaking, and seeing”; however, in cognitive intelligence fields that require external knowledge, logical reasoning, or domain transfer, it remains in the early stages. Cognitive intelligence will draw inspiration from cognitive psychology, brain science, and human social history, using knowledge graphs, causal reasoning, and continuous learning technologies to establish effective mechanisms for stable acquisition and expression of knowledge, enabling machines to understand and utilize knowledge, achieving a critical breakthrough from perceptual intelligence to cognitive intelligence.
2. What is a Course Knowledge Graph
A course knowledge graph is the structured and visualized result formed by using subject experts and AI technology to manually or automatically organize course content, extract course knowledge points, and establish a relational network structure between knowledge points and course resources.
As a refinement and summarization of course content, knowledge graphs have broad applications. Especially in online education, the combination of knowledge graphs and AI recommendation algorithms can provide personalized education for each student based on their mastery of knowledge points, plan learning paths, and display learning conditions, thus achieving tailored teaching for each individual. Specifically, it includes several typical application scenarios:
(1) Course knowledge point graph display
(2) Adaptive learning: learning path display, learning resource recommendation
(3) Learning condition reports
(4) Learning support robots (automated Q&A for courses)
(5) Intelligent lesson preparation
The course knowledge graph is a product of the combination of AI technology and education, and its various applications are the main battlefield of AI + education, especially AI + online education. The vast online education resources and adaptive learning and recommendations based on knowledge graphs represent a significant disruption to the traditional “one teacher for many students” educational model that has lasted for thousands of years.
3. OUPENG Course Knowledge Graph
As a leading enterprise in online education, OUPENG Education has thousands of course resources covering dozens of subjects and hundreds of majors. There is an urgent need for the construction of course knowledge graphs, and it possesses unique data advantages such as courseware and corpus resources required for this construction.
Since 2019, the engineering center has collaborated with the product center, student user division, and industry-education integration to conduct various attempts at the construction and application of course knowledge graphs. The construction and related applications of knowledge graphs for courses/majors are being increased in phases and batches.
4. Current Achievements and Capabilities
Through the continuous exploration, experimentation, and preliminary research of related technologies and algorithms by the AI innovation group of the engineering center, a semi-automated and semi-manual (90/10) method for extracting course knowledge points, identifying overall relationships between knowledge points, and associating knowledge points with resources (videos, documents, test questions) has gradually been developed.
The methods mainly include: video content analysis (OCR), audio content analysis (ASR), syllabus analysis (web scraping, data cleaning), knowledge point extraction (natural language processing), identification of overall relationships between knowledge points (natural language processing), and association of knowledge points with resources.
Based on the different styles of each course and different types of resources, AI technology automatically extracts and identifies, combined with manual corrections, forming a production line that allows the construction of a course knowledge graph to be completed within 1-2 person-days. Feedback from teachers in the product center last year indicated that manually constructing a course knowledge graph would take half a month to a month. It is evident that the AI technology’s automated extraction production line method greatly improves the efficiency of constructing course knowledge graphs.
Course Knowledge Graph Construction Production Line
By comparing the knowledge points extracted manually by a teacher and those extracted automatically by AI for the same course (“Introduction to the Basic Principles of Marxism”), it can be observed that the AI-extracted knowledge points have a finer granularity, with the number of extracted knowledge points being four times that of manually extracted points.
Comparison of manually extracted knowledge points (left) and AI-extracted knowledge points (right) for the same video segment
Hierarchy of knowledge points extracted by AI in the video segment above
AI-generated knowledge graph for the Fujian Normal University version of “Introduction to the Basic Principles of Marxism”
Total course duration is 14 hours and 29 minutes, with a total of 466 knowledge points extracted by AI
AI-generated knowledge graph for the Fujian Normal University version of “Introduction to the Basic Principles of Marxism”
Note: In the above image, the orange node represents a specific knowledge point, the green nodes represent associated test questions, and the yellow nodes represent associated videos. The “Teach” relationship between videos and knowledge points includes attributes such as startTime and endTime of the knowledge points in the video, which can be used for precise video recommendations..
As of now, OUPENG has 30 course knowledge graphs generated manually by teachers from the product center and 4 course knowledge graphs generated by the latest AI technology. Various application scenarios have been developed based on these.
(1) Course learning condition reports
(2) Test question recommendations based on incorrect answers (error notebook)
(3) Precise video recommendations based on incorrect answers
(4) Automated Q&A based on FAQs (in development)
In the future, while continuously promoting the construction of more course knowledge graphs, more application scenarios will be developed based on the existing knowledge graph content.
5. Future Plans
The construction and application of course knowledge graphs is an ongoing process. On one hand, the accuracy of AI algorithms and the level of automation in the production line can be further improved; on the other hand, the automatic discovery of more relationships between knowledge points (e.g., prerequisite relationships) will lay a solid disciplinary knowledge foundation for applications such as adaptive learning planning. The introduction of external knowledge (e.g., Baidu Encyclopedia, Chinese Wikipedia) will greatly enhance the accuracy of algorithms for the construction and application of knowledge graphs. The AI group will focus on the following areas to continuously improve the capabilities of knowledge graph construction and application:
Construction Aspects
(1) Introducing external encyclopedic professional knowledge (Baidu Encyclopedia, Chinese Wikipedia)
(2) Identifying more relationships between knowledge points (prerequisite relationships, equivalence relationships, etc.)
(3) Generating subject knowledge graphs (cross-course subject terminology graphs)
(4) OCR technology for automatic recognition of scientific formulas
(5) Automatic error correction of speech recognition results
(6) Online book knowledge graph construction platform
Application Aspects
(1) Automated Q&A based on knowledge graphs
(2) Applications of various inference and recommendation algorithms based on graph machine learning
(3) Collaborative filtering recommendation systems (based on XAPI user behavior collection)
(4) Intelligent lesson preparation
(5) Adaptive learning systems
6. Conclusion
The development and application of artificial intelligence technology is the most important and core technological trend of the 21st century. With the continuous release and maturation of related algorithms, artificial intelligence has found widespread applications in areas such as facial recognition, intelligent recommendations, and smart customer service.
AI + online education based on knowledge graphs will lead to profound changes in educational models and learning methods. The AI innovation team at the engineering center will delve into this blue ocean field of AI + education, continuously explore, and combine business innovation with technological innovation, using innovation to establish a unique technological moat for OUPENG online education.
Source: Internet
Reprinted from Teacher Training Alliance
