The development of new generation artificial intelligence technologies such as deep learning, knowledge graphs, and reinforcement learning is driving the “Internet + education” into a new era of “intelligent education”. Knowledge graphs, as a core driving force for the development of artificial intelligence, provide new empowering capabilities for education and teaching in the era of educational informationization 2.0. From the perspective of artificial intelligence research paradigms, knowledge graphs represent the evolution and development of the symbolic research paradigm in the era of big data and artificial intelligence; from the perspective of the stages of artificial intelligence development, knowledge graphs are an important foundation for the advancement of artificial intelligence from “perceptual intelligence” to “cognitive intelligence”. Therefore, from the perspective of “AI +”, educational knowledge graphs have broad application prospects in areas such as intelligent processing of educational big data, semantic aggregation of teaching resources, optimization of smart teaching, construction of learner profile models, adaptive learning diagnosis, personalized learning recommendations, and intelligent educational robots.
1. Introduction to Knowledge Graphs
A knowledge graph is a large-scale semantic network, a semantic representation of the real world. It represents entities as nodes, and the attributes of entities and the relationships between entities as edges, forming a web-like graph structure. This structured form is recognizable by humans and friendly to machines, facilitating machine understanding. The large-scale concepts, attributes, and relationships between entities in the graph provide rich semantic information and connections, naturally possessing various characteristics of graphs, allowing for relevant operations and applications. Once constructed, it can also serve as background knowledge for downstream applications. The concept of knowledge graphs was initially proposed by Google to improve search engine capabilities and enhance search quality. Due to the unique characteristics of knowledge graphs, they can play a significant role in multiple aspects of artificial intelligence.
2. The Connotation and Classification of Educational Knowledge Graphs
(1) The Connotation of Educational Knowledge Graphs
Currently, there is no unified understanding of the concept of educational knowledge graphs in academia, and different researchers have interpreted it from various research perspectives. We should recognize educational knowledge graphs from multiple dimensions: from the perspective of knowledge modeling, educational knowledge graphs are a method of modeling the ontology of subject knowledge; from the perspective of resource management, educational knowledge graphs are a way to semantically organize resources and their relationships in the educational field in the form of a “graph”; from the perspective of knowledge navigation, educational knowledge graphs can generate personalized learning paths aimed at learning goals with the support of technologies such as big data and artificial intelligence; from the perspective of learning cognition, based on knowledge graphs and the learner’s knowledge mastery state information, cognitive schemas of learners can be formed; from the perspective of knowledge bases, educational knowledge graphs are a structured semantic knowledge base that stores knowledge in the educational field in a way that can be “understood” by computers.
(2) Classification of Educational Knowledge Graphs
Educational knowledge graphs can be divided into two categories: (1) Static Knowledge Graphs (SKG), which are semantic networks formed by teaching elements involved in the teaching process as entity nodes and the logical relationships between teaching elements as edges. Here, teaching elements can refer to knowledge points in the subject (concepts, formulas, theorems, principles, etc.), as well as textbooks, courses, teaching resources, knowledge themes, knowledge units, teaching objectives, teaching questions, teaching participants, teaching contexts, etc.; (2) Dynamic Reason Graphs (DRG), which are multi-relational graphs represented by teaching events or activities, with logical relationships (sequential, causal, inverse, conditional, hierarchical, compositional, etc.) as edges. The following figure shows static knowledge graphs oriented towards knowledge points and dynamic reason graphs oriented towards activities.
3. Applications of Educational Knowledge Graphs Under the AI+ Perspective
(1) Supporting Intelligent Processing of Educational Big Data
Educational big data is the foundation for the application of artificial intelligence in education. The analysis and mining of educational big data is the extraction process from “data” to “knowledge”. Through knowledge graphs, we can integrate the vast and complex big data in the educational field into a semantic knowledge network, solving the problem of data aggregation and fusion, and optimizing the process of extracting data value, as shown in the figure below.
From the perspective of data aggregation and fusion, the low value density and sparse characteristics of educational big data require the integration of multi-source heterogeneous educational data to ensure the accuracy of analysis results. However, the current educational big data faces serious issues such as a lack of unified standards and specifications during the aggregation and fusion process, difficulties in dynamic changes of data patterns, and the semantic integration of multi-source heterogeneous data. Problems such as “data islands” and “data silos” remain severe, hence there is an urgent need for a flexible, scalable, and intelligently adaptive data model to achieve multi-dimensional deep integration of existing data. Knowledge graphs, as a lightweight data model with semantic associations and dynamic scalability, can achieve unified modeling and management of multi-source heterogeneous data to some extent. Firstly, subject experts and knowledge engineers need to define the standard model of knowledge graphs from three dimensions: teaching, management, and research; secondly, big data engineers need to map massive teaching log data, teaching resource data, learning behavior, and learning assessment data to the standard model; thirdly, through knowledge verification technologies, entities from different data sources need to be integrated; finally, entity linking technologies should be used to associate data with various resource libraries and knowledge bases.
(2) Supporting Semantic Aggregation of Teaching Resources
In recent years, with the emergence of ubiquitous learning environments and open educational resources such as MOOCs, SPOCs, and micro-courses, the methods of acquiring knowledge have shown characteristics of cross-end, cross-source, and cross-modal, facing serious issues such as decentralized disorder, knowledge fragmentation, difficulty in sharing, and lack of connections among learning resources. The semantic aggregation of learning resources has gradually become a hot topic in educational technology research.
Knowledge graphs, with their capabilities of semantic association and intelligent organization and aggregation, provide new ideas for machines to understand complex learning resources and build knowledge semantic networks. They are one of the key technologies for organizing, representing, and managing massive educational resources, achieving integration of educational resources.
(3) Supporting More Efficient Smart Teaching
Supporting “teaching based on learning” is the starting point and guiding principle of the teaching process. Its precise positioning determines the effectiveness of teaching design and process. Artificial intelligence technologies represented by knowledge graphs provide new technical means for the precise positioning of teaching objectives, mainly reflected in: (1) Based on subject knowledge graphs, it can accurately detect students’ mastery states of various knowledge points. With the help of educational big data collection technologies, intelligent learning systems can record students’ learning trajectories in assignments, exercises, exams, Q&A, and other aspects; combined with learning analytics technologies, the mastery level of students on knowledge points can be visualized in the form of knowledge graphs, accurately locating students’ learning weaknesses and fragile knowledge points. (2) Combining knowledge tracking technologies and relevant educational theories, it can dynamically predict changes in students’ mastery of various knowledge points. Each student’s learning state, progress, and knowledge level are different and dynamically changing, requiring artificial intelligence technologies for dynamic prediction.
(4) Empowering the Construction of Learner Profile Models
User profiling is a modeling method that describes users’ multi-dimensional characteristics based on user behavior big data, integrating technologies such as text mining, sentiment analysis, knowledge extraction, and data visualization, essentially “tagging” the subjects of the profile. Learner profiles are a special form of user profiles, primarily used to describe the individual characteristics of learners in areas such as subject knowledge, cognitive ability, subject literacy, learning style, and emotional state, which is the premise and foundation for providing personalized support services. The general process of learner profiling includes four stages: acquiring learning behavior data, analyzing learning behavior data, extracting user tags, and generating user profiles.
(5) Empowering Diagnostics for Adaptive Learning
Current smart education emphasizes a “learner-centered” teaching philosophy, yet assessing and diagnosing learners’ true mastery of knowledge and skills remains a challenge faced by global educational informationization. According to the cognitive theory perspective, learning is a process of forming and establishing cognitive structures through learners’ psychological processing and information handling in the face of real problem situations, while learning diagnosis is the process of assessing learners’ cognitive structures through diagnostic testing.
(6) Empowering Recommendations for Personalized Learning
Personalized learning is the essential pursuit and value orientation of educational development and the best practice for empowering education with artificial intelligence technologies. However, the exponentially growing learning resources exacerbate learners’ “cognitive load” and “learning disorientation” issues, and the contradiction between the vast enrichment of learning resources and insufficient personalized learning service supply is increasingly prominent. Personalized learning recommendations can recommend appropriate personalized learning resources and paths based on the current knowledge state of learners, becoming the key and foundation for achieving precise personalized learning.
(7) Empowering Intelligent Educational Robots
The Horizon Report has predicted for several years that teaching robots will become a key technology that continuously impacts the field of education; the Open University in the UK also pointed out in its 2019 Annual Report on Innovative Teaching that “robot-assisted learning” will become a potential “innovative teaching method” in the education field. Educational robots can act as learning assistants and intelligent companions for learners, providing intelligent services such as Q&A, navigation, recommendations, questioning, and social interaction, which play an important role in enhancing learning interest, stimulating learning motivation, and improving learning outcomes. At the same time, educational robots can also assist teachers in teaching monitoring, management, and automatic Q&A, extending teachers’ expressive capabilities, knowledge transmission abilities, and communication skills.
Currently, the robots applied in the education field are mainly conversational or chatbots. These robots use machine learning and knowledge graph technologies to enable machines to understand human language, responding to learners’ questions or completing specific tasks based on in-depth analysis of human intentions through simple logical reasoning and rule matching processes. For example, a conversational robot is a complex system that integrates language perception, speech recognition, intelligent decision-making, and automatic feedback, involving many artificial intelligence technologies such as natural language processing, knowledge graphs, knowledge reasoning, and reinforcement learning, where the knowledge graph plays a decisive role.
The following figure shows the basic structure of a conversational educational robot, which mainly includes four parts: natural language processing module, dialogue management module, knowledge graph module, and natural language generation module. The basic processing flow is as follows: the natural language processing module converts the text and voice data input by the learner into the machine’s internal representation, and with the assistance of the knowledge graph, processes the input for entity recognition, entity linking, anaphora resolution, and semantic understanding based on deep learning technologies, ultimately parsing it into a form of slot-value pairs; the dialogue management module comprehensively integrates the learner’s input data and information from the knowledge graph, generating answers to questions through knowledge reasoning, semantic disambiguation, context understanding, and semantic retrieval, and finally feedback is provided to the current learner by the natural language generation module. In this process, the knowledge graph serves as the memory system of the educational robot, storing a vast amount of common knowledge in the educational field, as well as data information such as learners’ emotional states, interests, and knowledge skills, enabling the robot to possess memory, reasoning, and judgment capabilities similar to the human brain.
Deep learning and knowledge graphs are the latest achievements of the symbolic and connectionist research paradigms in artificial intelligence. Integrating these two technologies deeply into education and teaching to improve the quality of precision teaching and the level of personalized services has become an inevitable requirement for promoting the development of educational informationization 2.0, and this is also an important way for intelligent education to move from theory to practice. However, relying solely on intelligent education based on deep learning technologies presents issues of black box and lack of interpretability, while knowledge graphs precisely bridge this gap, driving the development of intelligent education with knowledge at its core. As the foundation for the transition of intelligent education from the stage of “perceptual intelligence” to “cognitive intelligence”, knowledge graphs can provide technical support for intelligent processing of educational big data, semantic aggregation of teaching resources, optimization of smart teaching, construction of learner profile models, adaptive learning diagnosis, personalized learning recommendations, and educational applications of intelligent educational robots.
Note:The above images and text are valuable for sharing; copyright belongs to the original author and source. The content reflects the author’s views and does not represent the views of this public account or its responsibility for authenticity. If there are copyright issues, please contact us promptly to delete, contact email:[email protected].