Fuzzy knowledge graph system for artificial intelligence-based smart education

2.1 Data acquisition method

There are three main types of data sources for smart education field: the first type is unstructured data and the second type is semi-structured data, and the third type is structured data. Assuming that the entity sequence to be labeled is X and the state sequence is Y, the data acquisition is transformed into a state sequence in which T(x, y) takes the maximum value [9]. Under the condition that the random variable X takes the value x, the random variable Y takes the value y, the conditional probability calculation formula is: T ( x , y ) = log [ ∑ i , j x j y j ( x i - 1 y i - 1 , x i y i ) ] ∑ i = 1 n a i xy(1)

In this sequence, x₁, x₂, …, x _n are inputs, where each x _i is derived from an external input or a higher-level neuron, and a _i is the weight information corresponding to x _i and y _i . After the input information is weighted and summed and finally processed by the function f (t); the specific calculation process is as following: f ( t ) = T ( ρ ∑ i = 1 n s i x i , λ ∑ i = 1 n s i y i )(2)

Where s _I is the convolution kernel, ρ represents the input matrix vector, and λ is the bias parameter. The classification decision tree model is a tree structure for classifying instances, which is composed of nodes and directed edges and is a measure of uncertainty in a random variable. Assuming a discrete random variable with a finite number of values, and its probability distribution is: P ( x , y ) = f x ( t ) + f y ( t )(3)

The similarity algorithm between two users can simply and accurately evaluate the relationship between two users. The principle is to determine whether different users have the same preference by calculating the distance between two users in the scatter chart: K ( x , y ) = ∑ i = 1 n f ( i ) ( x n - y n ) 2(4)

Where K(x, y) represents the point set of the user in space, and f(i) is the user space distance. Relevance evaluation is another method of calculating the relationship between users. When the scoring data is not standardized, the relevance evaluation can give better results and the calculation formula is as following: Y ( X ) = α E ( XY ) - E ( X ) E ( Y ) E ( X 2 ) - E ( Y 2 ) E ( X 2 ) + E ( Y 2 )(5)

Where X and Y represent user variables, α is the linear correlation between users, and E is the mathematical expectation; the general fuzzy knowledge graph focuses on breadth and emphasizes the integration of more entities. Compared with the fuzzy industry knowledge graph, its accuracy is not high enough and is affected by the scope of this concept.

When calculating the similarity of words, this study considers the influence of the ambiguity of a single word on the calculation of word similarity, and considers that the maximum value of the mutual matching between individual concepts of words as the similarity between words reflects the local similarity of words [10]. For two words W₁ and W₂, assume that W₁ has n concepts s₁₁, s₁₂,  ...  , s_1n; W₂ has m concepts s₂₁, s₂₂,  ...  , s₂m, so the formula for calculating word similarity is as following: C ( W 1 , W 2 ) = 1 α ∑ i = 1 n s 1 i s 2 i(6)

Where C (W₁, W₂) is a commonly used set similarity measurement function, and its calculation idea is based on the ratio of the intersection and union of two sets, which is mainly used to measure the correlation of two sets. The cosine similarity calculation formula is as following: cos ( C 1 , C 2 ) = γ ∑ i = 1 C 1 s 1 i 2 + ∑ j = 1 C 2 s 2 j 2(7)

Where C₁ and C₂ represent two vector dimensions; therefore, a weight value γ is set in the similarity calculation for a weight assignment of the name and the relationship vector only selects the relationship of the important category. The similarity of two entities is combined into the root mean square through the combination of the two sets of vector similarity, and the calculation formula is as following: sin ( V 1 , V 2 ) = γ cos ( v 1 n , v 2 n ) 2(8)

Where V₁ and V₂ are two entities, v_1n represents the name and attribute vector; v_2n are the relation vector; γ is different according to the value set by different entities. Therefore, the advantage of using the process index to measure the development process of disciplines is scientific and comparable. The formula of process index is: I ( i ) = ∑ i = 1 n k cos ( C 1 i , C 2 i ) sin ( V 1 i , V 2 i )(9)

Where I is the advantage index; k is the number of years; the convolution operation is actually a process of extracting local features through a sliding window; the method of convolution operation is as following: M i = ∑ i = 1 n ∑ j = 1 n ( x m + i , y n + j )(10)

Where M _i is the width and height of the window, m and n are the indexes of the local area of matrix corresponding to sliding window, and i and j are indexes of sliding window. Another advantage of this is that the output of convolutional neural network can be easily added to another neural network for other operations.

3.1 Process design and information extraction

The construction methods of fuzzy knowledge graph include fully automatic, semi-automatic, and all manual. The fully automated construction method is not good enough due to the current natural language processing methods and the accuracy is not satisfactory. Although the completely manual construction method guarantees accuracy, it requires a huge amount of manpower and time. Therefore, how to coordinate the accuracy and efficiency in order to construct a fuzzy knowledge graph efficiently and accurately is a big problem that needs to be solved [11]. The construction of the AI-assisted smart education knowledge graph includes four challenges of accuracy, full subject, full coverage, and usability. Accuracy requires that the basic education knowledge in the graph must be accurate, and its source of knowledge must be authoritative resources such as textbooks; the full subject requires that the graph must cover all the main subject knowledge points of AI-assisted smart education. Full coverage requires that the knowledge of each subject must be covered all knowledge points specified in the subject curriculum standards; usability means that the efficiency of knowledge retrieval and access in the graph must be sufficiently high (Fig. 1). The technical route of the construction of the fuzzy knowledge graph of AI-assisted smart education combines the high accuracy of manual construction with the high coverage and easy maintenance of automatic construction, while avoiding the shortcomings of manual construction method update difficulty and insufficient automatic construction accuracy.

OPEN IN VIEWER

Fig. 1

Process design and information extraction for AI-assisted smart education.

By counting high-frequency keywords, the development of AI-assisted smart education, namely the technical dimension and the product dimension, can be taken into account. Key words such as deep learning, control systems, neural networks, and data mining indicate that the research energy of AI-assisted smart education in the technical dimension is mainly invested in the algorithm level, while the basic research on hardware is less invested. The application of AI is an educational scenario, involving education, learners, prediction, sentiment analysis, big data and other content. The application of deep learning in the field of AI-assisted smart education can optimize learning paths, reduce teachers’ repetitive work, and provide real-time feedback, so that teachers can grasp the learning status of learners in real time (Fig. 2). This includes building an intelligent analysis platform that includes learner emotion recognition, personalized resource recommendation, and learning behavior model will provide an important carrier for deep learning education applications. As the core dimension of deep learning research, mainstream algorithms focus on the technical layer, providing analysis techniques for intelligent analysis platforms through convolutional neural networks, recurrent neural networks, representation learning, migration learning to solve data circulation problems in complex education scenarios and provide technical support for the whole process of intelligent generation from data acquisition, information processing and knowledge construction, to intelligent output.

OPEN IN VIEWER

Fig. 2

Process design analysis of AI-assisted smart education.

In terms of coverage, fuzzy knowledge graphs can also be divided into general knowledge graphs and industry knowledge graphs. The general fuzzy knowledge graph focuses on breadth and emphasizes the integration of more entities. Compared with the fuzzy industry knowledge graph, its accuracy is not high enough and is affected by the scope of the concept, which is difficult to use the ontology library to support axioms, rules and constraints. Standardize its entities, attributes, and relationships between entities. The industry fuzzy knowledge graph usually needs to rely on the data of a specific industry to construct, which has specific industry significance. In the industry’s fuzzy knowledge graph, the attributes and data patterns of entities are often rich, and different business scenarios and users need to be considered. Through knowledge extraction technology, knowledge elements such as entities, relationships, attributes, etc. can be extracted from some public semi-structured and unstructured data. Through knowledge fusion, the ambiguity between reference items such as entities, relationships, attributes, and fact objects can be eliminated, and a high-quality knowledge base can be formed. Knowledge reasoning is to further explore the hidden knowledge on the basis of the existing knowledge base, thereby enriching and expanding the knowledge base. The comprehensive vector formed by the distributed knowledge representation is of great significance to the construction, reasoning, fusion and application of the knowledge base [12].

The fuzzy knowledge graph is a knowledge base composed of nodes and the relationships between nodes. The nodes are composed of knowledge points or teaching resources related to knowledge points. Each node has a globally unique identifier. The relationship between points expresses the relationship between knowledge points and knowledge points, between knowledge points and teaching resources, and between teaching resources. The fuzzy knowledge graph is constructed based on the curriculum standards and discipline teaching rules, and fully considers the sequence of learning knowledge points, and the different requirements of the same knowledge point in different learning stages. For the same knowledge point, there are big differences in the teaching goals of students of different ages, and these differences need to be reflected in the construction of the fuzzy knowledge graph [13]. Fuzzy knowledge graph is not only a stack of knowledge points or teaching resources, but also includes the connection between knowledge points, knowledge points and teaching resources, and teaching resources. Taking into account the cognitive differences of each learner and the differences in learning styles and other factors, in order to better provide learning services for learners, the resources in the fuzzy knowledge graph have attributes and not only include text-based learning resources, such as audio and pictures meet different learning needs of different learners (Fig. 3).

OPEN IN VIEWER

Fig. 3

Entity recognition and relationship mining of fuzzy knowledge graph system.

The size of the network node represents the frequency of keyword appearance, the connection represents the co-occurrence relationship of the keyword, and the thickness of the connection represents the closeness of the association. At the same time, in order to more accurately represent the status and relationship of keywords, the frequency and centrality of keywords can be counted and displayed in the form of tables. Keyword centrality represents the strength of the mediator’s ability of the keyword in the entire co-occurring network relationship. Generally, nodes with keyword centrality greater than 0.1 are more important in the network structure and play a role in the evolution of the keyword structure. Combining various factors, it is found that the keywords with the highest frequency are selected. Generally speaking, the keywords with higher frequency are not low in centrality because the more frequently the keyword appears, it co-occurs with other keywords. This also makes these keywords become hot vocabulary of the year (Fig. 4). According to the knowledge sharing process of AI-assisted smart education, the members of AI-assisted smart education are divided into knowledge providers, knowledge receivers, and knowledge assessors and the characteristics of the knowledge sharing process of the three types of AI-assisted smart education subjects are analyzed and the relationship is a nonlinear complex system.

OPEN IN VIEWER

Fig. 4

Data extraction analysis of AI-assisted smart education.

The method of constructing fuzzy knowledge graph based on external knowledge base mainly uses data from online search engine websites as the main source of research data. These data contain a large amount of high-quality common-sense knowledge, are slow to update, and have a certain format. They take the interactive search engine as an example, extract various related entities through abstracts, and obtain the upper and lower relationships of various related entity concepts through the anchor links corresponding to the sub-categories in the directory module [14]. They extract the attribute value pairs and entity relationships corresponding to the entities through the information module. Similarly, data can be extracted from multiple other online search engine knowledge websites to make up for the lack of knowledge extracted from a single search engine knowledge website. Co-word analysis is a content analysis technique that analyzes the co-occurrence forms of items in the same text topic, confirms the relationship between related topics in the subject field represented by the text, and explores the development of the subject. The main function of co-word analysis is to discover the analysis hotspots and main content of the research object through the clustering of high-frequency topic words reveal the corresponding research structure in depth, and then systematically explore its research dimensions, disciplinary background and theoretical basis, with a view to further grasp its research status, academic hotspots and development trends.

4.1 High-quality resource sharing and personalized service

Knowledge graph construction mainly studies how to extract specified entities from massive data, and build structured entity-to-entity relationships based on entity information. The key to the construction of educational knowledge graph lies in the use of knowledge extraction, knowledge fusion and other techniques to extract entities in the education field. This model can help the effective implementation of smart teaching. The key to the implementation of smart teaching lies in accurately determining the teaching objectives and screening, and optimizing the group learning path suitable for classroom teaching. A learning path is a sequence of concepts and activities selected or selected by learners in the learning process. The relationship between ability and ability mainly refers to the relationship between problem-solving ability and critical thinking ability and the relationship between problem-solving ability and creative thinking ability. The problem-solving ability is the key ability in the subject ability and learners need to reflect on their own problem-solving solutions based on the learner’s perspective (Fig. 5). Relation extraction is the process of extracting various dependencies from the knowledge base in the education field. This model can also provide an important reference for the construction of the educational resource system. The resource system is the structure of resource organization and storage and high-quality personalized learning cannot do without the strong support of high-quality educational resources [15].

OPEN IN VIEWER

Fig. 5

Cluster analysis results of high-frequency keywords for AI-assisted smart education.

The design of fuzzy knowledge graph is mainly based on the theory of scientific development model, scientific frontier theory, structural theory of social network analysis, information foraging theory of scientific communication and the theory of knowledge unit discrete and reorganization. The significance of these theoretical foundations is to strengthen the interpretability of the graph, the rationality and correctness of the interpretation, and realize the two major functions of the theory through the interpretation of the graph, that is, the interpretation function of the field’s current situation and the predictive function of the field’s future prospects [16]. This AI-assisted smart education development model can more profoundly explain the formation, accumulation, diffusion, and conversion process of a subject field citation cluster on the fuzzy knowledge graph, and reveal the emergence and evolution of a frontier of AI-assisted smart education research. The core function of the model is to use a visualized graph of the evolution of AI-assisted smart education to reflect and approach the law of scientific development in a specific field more deeply as a whole with a higher level of abstraction of AI-assisted smart education and more vivid and intuitive visual images (Fig. 6). Therefore, the mechanisms of scientific discovery can be explained by the knowledge unit about cohesion and dissociation based on network structure and information changes which is similar to the aforementioned scientific discovery mechanism.

OPEN IN VIEWER

Fig. 6

Entity recognition analysis of AI-assisted smart education.

The realization of the knowledge graph is based on many key technologies and its information extraction task is to obtain entities and relationship sets from a large amount of basic data and basic corpus through natural language processing such as word segmentation and part-of-speech tagging. In order to be able to construct a high-quality knowledge graph and perform accurate logical reasoning, the types of the acquired entities and relationship sets should be extracted and corresponded to the ontology model. In the process of type analysis and correspondence with ontology, there may be multiple related entries representing the same concept, or one entry may be related to multiple concepts, so entity alignment, semantic fusion and semantic disambiguation processing are required. After completing the construction of the knowledge graph, in order to enrich and optimize the semantic relations of the knowledge graph, it is also necessary to clean the knowledge graph and the relationship reasoning work. The knowledge graph has a wide range of uses and the question and answer process mainly include extracting the keyword vectors in the question. Based on the understanding of the question content, the sub-graph matching operation is carried out in the knowledge graph, which is to search the network with different scales and different matching degrees. Community structure with different internal connection tightness and the entity object set contained in community structure generates answers or retrieves answers in the knowledge base.

With the increasingly complex autonomous control of AI systems, these systems themselves should also be aware of the educational implications of their behavior. Some scholars believe that the educational issues of complex intelligent systems need to be considered clearly before they are put into use, that is, educational principles and rules should be introduced into AI systems. AI agents should be constructed so that they can make judgments and make educational decisions similar to human beings based on external environmental information. Fuzzy knowledge graph is based on the literature of a specific field as the research object, through the parameterization and graph analysis of the literature information to graph the relationship of the knowledge context of a specific subject within a period of time. Its research function has the dual attributes and characteristics of graph and spectrum and it is both a visual knowledge graph shape and a serialized knowledge pedigree (Fig. 7). Therefore, it is increasingly realized that the autonomous decision-making of AI can not only be limited to its originally designed work service goals, but also needs to be able to judge self-decision, behavior and consequences, so that their behavior is in line with the maximization of human interests. But this process has caused some new educational problems and AI and robotics are applied in the field of engineering manufacturing. They are mainly pre-programmed by human instructors, and then assist humans in their work according to the programmer’s intentions [17].

OPEN IN VIEWER

Fig. 7

Structure of automatic knowledge acquisition and integration.

Fuzzy knowledge graph is a grand data model, which can build a huge knowledge network, including the entities composed of all things in the world and the relationships between them, showing all aspects of knowledge with pictures and texts, allowing people to obtain information and find what they want more conveniently. The fuzzy knowledge graph intelligently connects people and knowledge, and it can be described as an important technical cornerstone. First, the entities in the knowledge graph are divided into two independent perspectives: semantic feature perspective and structural feature perspective, and based on these two perspectives, two different classification models are obtained: semantic-based entity alignment model and structure-based entity alignment model. In the training process, put the best results obtained from each model classification into the labeled data, and iteratively train the models of the two perspectives, so that the models of the two perspectives complement each other until they converge. The core of AI is to observe and perceive the world through data, to quickly and automatically acquire knowledge through data, to make predictions, to automate, and to provide end-to-end intelligent services through knowledge, which creates infinite opportunities and ultimately achieves goals. Fuzzy knowledge graph technology includes knowledge acquisition, organization, application and inheritance and this will be its core basic ability.

Is necessary to integrate multi-source heterogeneous education data to ensure the accuracy of analysis results. As a lightweight data model, knowledge graph has the ability of semantic association and dynamic scalability, which can realize the unified modeling and management of multi-source heterogeneous data to a certain extent. With the help of educational big data collection technology, the intelligent learning system can record the learning trajectory of students in all aspects of homework, exercises, and examinations [18]. Combined with learning analysis technology, it can visually display the degree of mastery of students’ knowledge points in the form of knowledge graphs and accurately locate students’ learning shortcomings and weak knowledge points (Fig. 8). Each student’s learning status, learning progress, and knowledge level are different and change dynamically, and AI technology is needed for dynamic prediction. Knowledge tracking technology is an emerging predictive method in the current intelligent education field. It predicts the learner’s internal and implicit knowledge and skill state through the learner’s external and explicit learning performance or behavior sequence. In smart teaching, knowledge structure visualization tools provide learners with cognitive scaffolding to effectively organize and construct knowledge.

OPEN IN VIEWER

Fig. 8

Information graph of fuzzy knowledge graph system for AI-assisted smart education.

5.1 Data source

This research is based on the Web of Science data source, and retrieves the journal literature from 2009 to 2019 in the SCI-EXPAND-ED database with the subject term “AI-assisted smart education”. According to research needs, the Web of Science category is narrowed down to a few related to education within the scope of the category, and further selected the document type “Education” as the statistical object of this research, the final logical search sentence formed is:

Theme=(”AI-assisted smart education”);

Document type=(Education) AND Web of Science category=(EDUCATION SCIENCE AI OR EDUCATION SCIENCE INTERDISCIPLINARY APPLICATIONS OR EDUCATION SCIENCE THEORY METHODS OR EDUCATION SCIENCE INFORMATION SYSTEMS OR EDUCATION SCIENCE SOFTWARE ENGINEERING OR EDUCATION SCIENCE CYBERNETICS OR EDUCATION SCIENCE HARDWARE ARCHITECTURE);

Time span = 2009–2019;

Database = SCI-EXPANDED;

According to the above logical expression, 5284 records were obtained after retrieval.

This study selects keywords with a frequency of not less than 3 as high-frequency keywords. After repeated inspection and proofreading, 33 high-frequency keywords are finally determined, hoping to fully reflect the hot issues in the field of AI-assisted smart education, and the keywords are ranked according to their frequency. Based on bibliometrics, this research uses mathematical statistics to objectively describe and summarize the current research status and future development trends of AI-assisted smart education. First, SPSS software was used to conduct descriptive statistics on each year’s journal articles from 2009 to 2019 to reveal the relationship and trend of the number and time of publication of AI-assisted smart education papers in recent years; CiteSpace, a visual analysis tool for citation network, was also used to perform visual analysis and co-citing documents, co-authors, co-authoring institutions, co-authoring countries, co-citation analysis of authors, co-word analysis of subject terms and keywords. Drawing co-citation graphs can dynamically identify co-citation clusters, and perform emerging words detection; then using authors, institutions, journals, etc. as research variables to draw a fuzzy knowledge graph to show the current development of AI-assisted smart education research, including the co-occurrence relationship of authors of AI-assisted smart education research, co-occurrence relationship of research institutions, keyword co-occurrence, keyword clustering and keyword mutation and other graphs, by analyzing the graphs and supplemented by statistical or charting software, discover the research progress and frontier hotspots in this field.

5.2 Result analysis

Judging from the content of each cluster, the human-computer relationship in the context of AI education essentially reflects the characteristics of digitalized apprenticeship: first, intelligence, AI researchers use big data and algorithms to build and design Intelligent education applications to improve the effectiveness and accuracy of education services; second, ubiquity, the Internet and virtual communities on which intelligent education applications rely makes individual learning ubiquitous and ubiquitous; third, human-machine collaboration AI-assisted smart education applications can independently cooperate with individual teaching and learning, and adapt to the educational environment independently. For example, teachers and managers can use creative tools to design AI-assisted smart education applications [19]. At the same time, students can learn through the intelligent guidance system or through it can teach robots to learn. fourth, the personalized, intelligent education applications can record and monitor learners’ learning behaviors, and intelligently analyze their learning characteristics and learning styles to provide suitable learning content, methods and models, achieve personalized learning. In order to further clarify the clustering of AI education research and the key node literature in the relationship between the knowledge structures, this research discusses machine education, intelligent guidance system and human education from three levels (Fig. 9).

OPEN IN VIEWER

Fig. 9

Construction and application of fuzzy knowledge graph system for AI-assisted smart education.

Using high-frequency keyword co-occurrence network conditions for visual analysis, it is easier to observe the research hotspots of AI and education and the relationship between them, which can be seen that the more prominent keywords in the literature are the learning process. Keywords such as intelligent education, educational applications, intelligent machines, machine learning, deep learning, and information technology are closely related to other keyword networks, indicating that these keywords are under research [20]. The attention of the authors is relatively high, and it is also a research hotspot. It is particularly important that the structured knowledge graph, as the knowledge base of the question answering system, provides strong support for the realization of cognitive intelligence and the provision of intelligent knowledge question answering (Fig. 10). The retrieval system returns results based on the keywords entered by the user. In the question and answer system, what the user enters is not a keyword but a question expressed in natural language. The question and answer system directly respond to the precise answer to the question ant the user profile gives a recommendation explanation based on the relationship. For example, the reason for recommending a resource to a user is that someone in the user’s scientific research team uses the resource, and combining specific learning, scientific research and other scenarios to give the recommended reason to make the user understand the recommended resource relationship with the topic of knowledge and its role in the knowledge chain to promote knowledge sharing, knowledge transfer, knowledge understanding and knowledge innovation [21].

OPEN IN VIEWER

Fig. 10

Multi-dimensional analysis results for construction and application of AI-assisted smart education.

AI-assisted smart education is an innovative paradigm for educational applications. Under the guidance of the concept of cross-media intelligence, it can collect learning environment data, individual cognitive data, behavioral data, etc. from different information sources, and conduct multi-dimensional learning analysis and cross-media. The generalized reasoning reveals the mental activities, cognitive mechanisms, behavior patterns, etc. of the learning process, conducts in-depth analysis of learning disabilities and their causes, makes reasonable decisions, and makes scientific recommendations [22]. Keyword emergence refers to the sudden increase or decrease of the keyword cited within a certain period of time, which can reflect the frontier and development trend of a certain field. The AI-based student diagnosis, assistance, and evaluation system can monitor students’ comprehension capabilities, evaluate students’ prior knowledge, establish learner profiles, provide personalized assistance, and perform quantitative and qualitative evaluations of learners’ performance. Cloud computing-based learning services use AI technology to aggregate and analyze the data presented by users on different social networks, infer their knowledge and interests evaluate users’ knowledge levels on different topics, and recommend their relevant research content to help users gain satisfactory work. AI-assisted smart education can also expand the existing learning space, carry out stretching experiments, break the traditional laboratory space, and realize boundless learning.