A Method of Ontology Integration for Designing Intelligent Problem Solvers

Nowadays, designing knowledge-based systems which involve knowledge from different domains requires deep research of methods and techniques for knowledge integration, and ontology integration has become the foundation for many recent knowledge integration methods. To meet the requirements of real-world applications, methods of ontology integration need to be studied and developed. In this paper, an ontology model used as the knowledge kernel is presented, consisting of concepts, relationships between concepts, and inference rules. Additionally, this kernel is also added to other knowledge, such as knowledge of operators and functions, to form an integrated knowledge-based system. The mechanism of this integration method works upon the integration of the knowledge components in the ontology structure. Besides this, problems and the reasoning method to solve them on the integrated knowledge domain are also studied. Many related problems in the integrated knowledge domain and the reasoning method for solving them are also studied. Such an integrated model can represent the real-world knowledge domain about operators and functions with high accuracy and effectiveness. The ontology model can also be applied to build knowledge bases for intelligent problem solvers (IPS) in many mathematical courses in college, such as linear algebra and graph theory. These IPSs have great potential in helping students perform better in those college courses.


Introduction
Nowadays, the knowledge from several sources needs to be integrated in order for machines to accomplish different tasks in a more intelligent way than conventional systems [1]. Knowledge integration is important in intelligent software development [2]. In order to achieve this, knowledge

Related Works
There are various ontology-based methods for knowledge integration, most of which focus on basic kinds of ontology and are mainly used for information searching. They have not yet met the requirements of knowledge integration.
The Semanticscience Integrated Ontology (SIO) is an ontology for facilitating biomedical knowledge discovery [14]. SIO provides an ontological foundation for the Bio2RDF linked data for the life sciences project and is used for semantic integration and discovery for SADI-based semantic web services (Semantic Automated Discovery and Integration -SADI). However, the concepts of SIO are only basic information for searching.
Ontology-based knowledge integration is also used for semantic web services. Ontology WSMO is built based on the Web Service Modeling Framework (WSMO) [15]. WSMO defines four top level elements as the main concepts which have to be described in order to describe the semantic web services: ontologies, services, mediators, and goals. These methods only solve the integration of ontology as information but cannot support solving decision problems.
Fuzzy ontology integration is used for the representation of uncertain knowledge on the semantic web [16,17]. The author in [16] used description logic and fuzzy set theory to represent fuzzy logic and reason on it. The study in [17] presented a method to integrate fuzzy ontology based on consensus theory. Nonetheless, those methods are just theoretical and cannot be applied in the complex knowledge domains in practice.
Ontology COKB (Computational Objects Knowledge Base) is a useful ontology to represent complex knowledge domains [12]. This ontology can be used to describe many kinds of knowledge, such as knowledge about relationships, operators, and functions. It can be applied to build intelligent educational systems [18]. However, ontology COKB is too general to represent a specific knowledge domain, so it is very difficult to apply. Furthermore, the combination problems on the knowledge components in COKB have not yet been mentioned.
Ontology is also a technique model for information retrieval via the processing and translation of ontological knowledge into database search requests [19]. It is also used to compare the existing ontology-to-database transformation and mapping approaches in terms of loss of data, domain knowledge applicability, and structural mapping. Nonetheless, the mentioned knowledge in this study is simple, and the ontology integration has not yet been studied.

Structure of the Kernel
The model of the knowledge kernel in this paper was an ontology which was built based on the object-oriented approach. The model of the knowledge kernel was a tube: (C, R, Rules ), in which C is a set of concepts, each concept is a class of objects, R is a set of binary relationships between concepts in C , and Rules is a set of inference rules. Figure 1 shows the structure of the knowledge kernel. It is a tube (C ,R ,Rules ) as part (A). The structure of each component is as Table 1. Each concept in C is a class of objects. An object has a ributes and internal relationships between them, it also has behaviors to solve the problems on it as part (B). This kernel had two kinds of problems: problems on an object and general problems (Section 3.2). The model of general problems is as part (C).

C R Facts
-Each concept c ∈ C is a class of objects, it has an instance set Ic including objects.
• Fundamental concepts are the default concepts of the knowledge domain. A set of fundamental concepts is denoted: Co -Set C is a set of concepts, the structure of an object in a concept is a tube: (Attrs, Facts, RulObj), in which: ◊ Attrs is a set of attributes: Attrs = AttrsVar ∪ AttrsList An attribute in AttrsVar is a variable which has the kind in Co: AttrsVar ⊆ {x|x:c, c ∈ Co}. An attribute in AttrsList is a list of variables which have the same kind in Co: AttrsList ⊆ {[x1, …, xn]|n ∈  and xi: c, c ∈ Co}.
◊ Facts is a set of facts on attributes in Attrs.
Facts ⊂ {f|f is a fact, var(f) ⊆ Attrs} ◊ RulObj is a set of deductive rules of the concept: In the case of ci = cj, the properties of a relationship Φ are considered: reflexive, symmetric, asymmetric, and transitive. * Relationships "is-a"  ∈ R: Let c1, c2 ∈ C: c2  c1 ⇔ c2 is a sub-concept of c1

6/Relationship between objects
Specification: x Φ y In the case of ci = cj, the properties of a relationship Φ are considered: reflexive, symmetric, asymmetric, and transitive. * Relationships "is-a" < ∈ R : Let c 1 , c 2 ∈ C : , o i ∈ I c , c ∈ C (1 ≤ i ≤ n) 4/Determination of an object by a value or a constant expression Specification: o = <const> Condition: o ∈ I c , c ∈ C <const>: constant 5/Equality on objects Specification: x = y Condition: x,y ∈ I c , c ∈ C or x = [x 1 , . . . , x n ] and y = [y 1 , . . . , y n ] with x i , y i ∈ I c , c ∈ C (1 ≤ i ≤ n) 6/Relationship between objects Specification: x Φ y Condition: Φ ∈ R , x ∈ I cx , y ∈ I cy , cx ∈ C , cy ∈ C * Kind(f): kind of fact f.

Definition 1.
Reference [20]: The closure of a set of facts Let Obj = (A rs, Facts, RulObj) be an object of concept in C , F be a set of facts. The closure of a set of facts F by Obj, Obj.Closure(F), is a maximum extension of F by combining facts in F and Obj.Facts and reasoning on rules in Obj.RulObj.

Definition 2.
Given two facts f 1 and f 2 . They are unified, , when they satisfy following conditions: 1.
f 1 and f 2 have them same kind of k; 2.
And if k = 1, 2, it means f 1 and f 2 are facts of information about object kind or determination of an object; Then f 1 = f 2 ; Or else if k = 3, it means f 1 and f 2 are facts of determination of a list of objects; Then NumberOfElements(f 1 ) = NumberOfElements(f 2 ); and f 1 Or else if k = 4, it means f 1 and f 2 are facts of determination of an object by a value or a constant expression; Then left(f 1 ) left(f 2 ) and compute(right(f 1 )) = compute(right(f 2 )); Or else if k = 5: it means f 1 and f 2 are facts of equality on objects; Then (left(f 1 ) left(f 2 ) and right(f 1 ) right(f 2 )); Or (left(f 1 ) right(f 2 ) and right(f 1 ) left(f 2 )); Or else if k = 6, it means f 1 and f 2 are facts of relationships between objects; Then NameOfRelationship(f 1 ) ≡ NameOfRelationship(f 2 ) and Property(f 1 ) ≡ "symmetric" and (left(f 1 ) left(f 2 ) and right(f 1 ) right(f 2 )); or (left(f 1 ) right(f 2 ) and right(f 1 ) left(f 2 )); Or NameOfRelationship(f 1 ) ≡ NameOfRelationship(f 2 ); and (left(f 1 ) left(f 2 ) and right(f 1 ) right(f 2 )); For which: • compute(expr): compute the value of the expression expr.
• NameOfRelation(f ): return the name of relation in fact f that is kind 6. • left(f ), right(f ): return the left side, right side of expression f.

Definition 3.
Reference [20]: Let x be a fact, A and B be sets of facts, the relationships between them have been defined as followed:

Rules -Set of Rules
A rule r ∈ Rules is one of three kinds, as follows: • r is a deductive rule, it has the form u(r)→ v(r), where u(r), v(r) are sets of facts; • r is a deductive rule for generating a new object, it has the form u(r)→ v(r), where u(r) and v(r) are sets of facts, they satisfy the conditions: ∃ object o, o ⊙ v(r) and not(o ⊙ u(r)); • r is an equivalent rule: h(r), u(r) ↔ v(r), where h(r), u(r) and v(r) are sets of facts, they satisfy the conditions: h(r), u(r)→ v(r), and h(r), v(r)→ u(r) are true.

Problems and Reasoning Methods on the Kernel Model
The kernel of knowledge model has two kinds problems: problems on an object and general problems with the kernel. Problems with an object relate to only the behaviors of an object. Each object of a concept in C uses its structure to solve the problems on itself. General problems with the kernel combine the knowledge of all objects and the kernel. When solving a general problem, the kernel has to apply the ability to solve problems with objects and the reasoning method on the knowledge of R and Rules .

Problem on an Object
An object of the Kernel Model has basic behaviors for solving problems on it. It is equipped abilities to solve problems such as: (i) Determine the closure of a set of a ributes; (ii) Determine the closure of a set of facts; and (iii) Execute deduction and give solutions for problems. In this section, the algorithm for solving the problem (ii) is presented. The other problems can be solved by using this algorithm.  The general algorithm for solving these problems was presented in [20]. It uses forward chaining reasoning for solving problems and applies heuristics rules as sample problems [21], arranging the priority to use inference rules. Each object a ends the reasoning process by solving problems on itself as Algorithm 1.

Knowledge Components of Operators and Functions
The knowledge of operators and functions is necessary to represent actual knowledge, especially the knowledge domains about computation. The knowledge of operators, called Ops -set, represents the computing operations over the objects in the knowledge domain. The knowledge of functions, called Funcs -set, represents how objects depend on each other, described as a set of conditions or a procedure over objects, in the knowledge domain. The integration of these components makes the knowledge base more powerful for representation and reasoning.

Principle to Integrate the Kernel and Components of Operators and Functions
The knowledge kernel can combine with other components, such as operators and functions, to represent real-world knowledge domains more accurately. When integrating new knowledge with the kernel (C, R, Rules) , some aspects have to be considered:

•
Building the structure of the additional component: This structure has to be built upon the kernel such that the integration of new concepts is seamless and consistent. Any additional component should have specification language similar to that of the kernel; • Adding more kinds of facts: Facts in any additional component need to represent the real knowledge and connect with former kinds of facts in the kernel. This is a study about the unification of facts; • Adding more kinds of rules from the additional component: More kinds of rules have been studied based on the integrated knowledge and need to be merged into the kernel; • Proposing the problems with the reasoning by integrating the kernel knowledge and additional knowledge.
In this paper, we present a method to integrate the additional knowledge of operators and functions as Figure 2. Appl. Sci. 2019, 9,

Structure of the Ops-Set
Ops is a set of operators between concepts in Co and C. Its structure is presented as Table 2: Table 2. Structure of the Ops-set.

Structure of the Ops -Set
Ops is a set of operators between concepts in C o and C . Its structure is presented as Table 2: Table 2. Structure of the Ops -set. (2) In which: + Set of unary operators O (1) :

Structure of the Ops -Set Additional Facts Additional Kind of Rules
Each operator is checked for its properties: commutation, association, identity.

Problems and Reasoning Methods on the Knowledge of Operators
The model of problems has the form (O, F, E) → G [22], where O is the set of objects, F is the set of facts, E = {e 1 , e 2 , . . . , e k } is the set of equations, and G is: KEYWORD": f } with "KEYWORD" is a keyword of the goal and f is a fact. Besides keywords as Section 3.2.2, "KEYWORD" may be as follows: - Compute": to determine the value of f when f is an expression; -Transform": transform an object into an expression between certain objects. Step 1. Record elements in hypothesis and goal; Classify kinds of facts in F and E.
Step 2. Check G; If G is obtained, then go to step 6.
Step 3: Use objects in set O and the set of facts in F and E to determine the closure of each object by Algorithm 1; Step 4: Use the equations in E to generate the new facts as relationship form; Use the relationships in F to generate the equations; Update KnownFacts and Sol.
Step 5: Select a rule in the set Rules for producing new facts or new objects; While (flag! = false) and not (G is determined) do Search r in the Rules set, which can be applied to KnownFacts.

Case: r is an equation rule;
if (r has form: g = h), then: r can generate a set of new facts.

Example 2. In the knowledge domain about direct current electrical circuits,Ops is the set of the knowledge of operators being series connection (+) and parallel connection (//) between two circuits. We considered the problem as follows:
Problem P1: Three resistors are connected like the Figure 3. Resistor T1 has value 45 Ω, Resistor T3 has value 90 Ω. The total current is 0.5 A, current though T3 is 0.3 A.
knowledge of operators being series connection (+) and parallel connection (//) between two circuits. We considered the problem as follows: Problem P1: Three resistors are connected like the Figure 3. Resistor T1 has value 45 Ω, Resistor T3 has value 90 Ω. The total current is 0.5 A, current though T3 is 0.3 A.
What is the current thought T1?  Table 3. ;

Fu.2. Dependence of an object on a function.
r is an equation rule, it has the form: where f, g are functions or expressions.
A function in the Funcs -set is specified as followed:

Structure of the Funcs -Set Additional Facts Additional Kinds of Rules
Let F = {f |f is a function}: α: F → N: assigning the number of arguments to a function.
Each function is checked for its properties: commutation.
Fu. 1. Determination of a function.
Dependence of an object on a function.
Fu. 3. Dependence of a function on an expression.  Step 2. Check G.
If G is obtained, then: Go to step 6.
Step    Set F is split into two sets:

Heuristic Rules
For making the inference process faster and more effective, in Algorithms 2-4, some heuristic rules can be used to search the inference rules, as follows:

Hypothesis 1. Restrict the set of rules that can be applied to the problem.
Based on the facts in the hypothesis of the problem, the reasoning part of the algorithms can reduce the set of rules being searched for the current problem, thus making it faster to search for rules. This heuristic rule is used after the step of recording the hypothesis of the problem in algorithms.

Hypothesis 2. Arrange the rules in order of priority.
For each kind of practical problem, inference rules are applied in order. Hence, these rules can be arranged according to their priority to speed up the inference process. This heuristic rule is applied when solving sub-problems in algorithms.

Application for Intelligent Systems in Education
Intelligent systems in education need useful knowledge bases to organize the complete knowledge in academic courses fully and accurately. The kernel ontology is used to represent the knowledge of matrixes and vectors, such as the structure of matrixes, the relationships between them, and the rules on matrixes. In this section, the method for integrating this knowledge kernel and the knowledge of operators and functions is applied to construct the knowledge bases of linear algebra and graph theory. These knowledge bases are the foundations for designing the intelligent problem solver (IPS) for the corresponding courses, which satisfy the criteria of IPS in mathematics education [23].

Representation of the Knowledge of Matrixes Using the Kernel Model
The basic knowledge of Matrixes in [13,24] can be represented by the kernel model. It includes the field of real numbers Ñ, the concepts of kinds of matrixes and vectors, and the relationships between these concepts and rules about transforming a matrix: (C kernel , R kernel , Rules kernel ), R kernel -set of relationships between concepts in C kernel : Relationship "is-a" between the kinds of matrixes, such as SquareMatrix is-a Matrix.
Rules kernel -set of inference rules on matrixes.

Design Knowledge Base of Linear Algebra
Using the Integrating Model (C, R, Rules ) + Ops.

Knowledge Base of Linear Algebra
The knowledge about linear algebra was collected from [13,24]. A part of this knowledge domain, about chapters Matrixes-Vectors, Linear Equations System, and Vector Space, can be represented by the model for integrating the kernel and knowledge of operators: (C, R, Rules ) + Ops.
C -set of concepts. Besides concepts in the kernel, C also includes concepts about the Linear Equations System, and Vector Space: R -set of relations. Besides the relations in the kernel, R also includes the relationships in the Linear Equations System and Vector Space:

Ops is the set of operators between matrixes and vectors, operators in vector space.
Rules is the set of rules on the knowledge domain about linear algebra. It includes rules for generating a new object, equation rules, and equivalent rules in linear algebra.
The knowledge base of this knowledge domain is represented in Appendix A.
Step 7: From B is a spanning-set of W, B is linearly independent. Then, B is a based-set of W

Knowledge Base of Graph Theory
The knowledge about graph theory was collected from [25][26][27]. In this section, we only cover the knowledge about undirected graphs. The knowledge base of this knowledge domain can be represented by the method for integrating the kernel and knowledge of functions: (C, R, Rules ) + Funcs .
C -set of concepts. It includes concepts in the kernel and concepts about the knowledge of the graph: R -set of relationships. Besides relationships in the kernel, R also includes relationships between graphs: Rules -set of rules on the knowledge domain about graph theory. It includes deductive rules and equivalent rules.
The knowledge base of this knowledge domain is represented in Appendix B.

Kinds of Exercises in Graph Theory
Based on the knowledge base and the inference engine in the above section, an intelligent system for solving problems in graph theory was designed. This system can solve many types of problems that are in the curriculum of this course. They are classified in the three following types [18]: • Type 1: find the shortest paths between nodes in a graph, find a minimum spanning tree for a weighted undirected graph. These problems also have additional conditions; • Type 2: Determine a graph from an expression of graphs; • Type 3: Prove a property of a graph or a relationship between two graphs.

Example 6.
Given two simple graphs G 1 , G 2 , as in Figure 5, prove G 1 and G 2 are isomorphic.
• Type 1: find the shortest paths between nodes in a graph, find a minimum spanning tree for a weighted undirected graph. These problems also have additional conditions; • Type 2: Determine a graph from an expression of graphs; • Type 3: Prove a property of a graph or a relationship between two graphs.

Intelligent Problem Solver in Linear Algebra
The integrating model between the kernel ontology and the knowledge of operators was used to design the knowledge base of IPS in linear algebra, as in Section 5.2. The knowledge base of this system satisfied the criteria of the knowledge representation for an IPS in education [27]. It can be used to support the study of Linear Algebra at university.
Symbolab is a website which can automatically solve mathematical problems [28]. It can solve several problems about matrixes and equation systems, and produce step-by-step solutions. Mathway is another website to solve linear algebra problems automatically [29]. It plays as a chatbot to interact with the user. The test results of Symbolab, Mathway, and our system are shown in Table 4 and Figure 6. Table 4. The results of testing of exercises in [13]. Symbolab is a website which can automatically solve mathematical problems [28]. It can solve several problems about matrixes and equation systems, and produce step-by-step solutions. Mathway is another website to solve linear algebra problems automatically [29]. It plays as a chatbot to interact with the user. The test results of Symbolab, Mathway, and our system are shown in Table  4 and Figure 6.  The knowledge bases of Symbolab and Mathway are organized as frames. They only can solve some types of exercises, which have already been set up. The interface of Mathway is a chatbot for showing the step-by-step solutions. Besides this, the user can input some simple problems by images to that program. In contrast, the knowledge base of our program was organized as a complete system, representing the knowledge similarly to the knowledge acquisition of human. Its solutions were clear, step-by-step, and similar to those of students.

Number of Problems that Can Be Solved
The IPS in linear algebra was also examined by the students in two universities. In the curriculum for information technology from these two universities, linear algebra takes place in the second semester of the first academic year. This survey focused on four requirements of an IPS in education [27,30]: sufficient knowledge base, the ability to solve common exercises, pedagogy, usefulness. Each criterion can be evaluated from 1-5, from very bad to very good. The meaning of The knowledge bases of Symbolab and Mathway are organized as frames. They only can solve some types of exercises, which have already been set up. The interface of Mathway is a chatbot for showing the step-by-step solutions. Besides this, the user can input some simple problems by images to that program. In contrast, the knowledge base of our program was organized as a complete system, representing the knowledge similarly to the knowledge acquisition of human. Its solutions were clear, step-by-step, and similar to those of students. Appl. Sci. 2019, 9, 3793 20 of 27 The IPS in linear algebra was also examined by the students in two universities. In the curriculum for information technology from these two universities, linear algebra takes place in the second semester of the first academic year. This survey focused on four requirements of an IPS in education [27,30]: sufficient knowledge base, the ability to solve common exercises, pedagogy, usefulness. Each criterion can be evaluated from 1-5, from very bad to very good. The meaning of each level was described in [30].
There were 87 students in this survey, who were either: • First-year students (63 students~72%): students who took this course the first time; • The other students (24 students~28%): students who took this course again because they had failed before.
Each student was prompted to choose three problems from a list of 149 exercises that could be solved by the program. After seeing the program-generated solutions for those problems, they were requested to input three more problems using specification language and then check the solutions produced by the program. Based on these results, the students evaluated this program by four criteria, on a scale of 1 to 5. The results of this survey are shown in Table 5.

Intelligent Problem Solver in Graph Theory
The integrating model between kernel ontology and the knowledge of functions was used to design the knowledge base of IPS in graph theory, as in Section 5. 3. The knowledge base of this system meets requirements of an IPS in education [23]. The students used this system to support their learning of graph theory at university.
VisuAlgo is a learning website [31] that can visualize algorithms in graph theory, such as graph traversal, finding the minimum spanning tree, shortest paths, graph matching, traveling salesman. Maple software is an essential mathematical tool for education [32]. It includes many packages for studying mathematics. Maple has a network package to represent the knowledge of graph theory. Although it can solve some problems about graphs, Maple only shows their results, and not step-by-step solutions. Table 6 and Figure 7 compare the ability for solving exercises in graph theory [27] between VisuAlgo, Maple, and our system.   With exercises of type 1, our system can solve problems with additional conditions, while VisuAlgo and Maple can only solve algorithms which do not contain conditions. Exercises of type 2 and type 3 are difficult for students. VisuAlgo only solves some exercises in type 3 about determining the ismorphism between two graphs, but our system could solve exercises of all types. Maple only shows the results of exercises and not the detailed solutions. It is a useful tool to support learning, but it does not meet the requirements of an IPS in graph theory.
Similar to the IPS in linear algebra, the IPS in graph theory was also examined by the students in two universities, based on four criteria: sufficient knowledge base, the ability to solve common exercises, pedagogy, and usefulness. In the curriculum for information technology from these two universities, graph theory took place in the first semester of the first academic year.
A total of 95 students joined this survey, and were again classified into two types: students who were studying this course for the first time (73 students~77%), and students who were studying the course for the second time (22 students~23%).
Each student was again prompted to pick three exercises in the list of 51 exercises that could be solved by the program. After seeing the program-generated solutions for those problems, they were asked to input three more problems using the specification language and then check the solutions produced by the program. Based on these results, the students evaluated this program by four criteria on a scale of 1 to 5. Results of this survey are shown in Table 7:

Conclusions and Future Works
Knowledge integration is an important method for developing intelligent software. It makes the system more intelligent than conventional systems. Systems with knowledge integration can simulate the knowledge organization and solve problems on the knowledge domains in a similar way to the human method. In this paper, a method for knowledge integration based on ontology was presented. This method is effective at representing the real knowledge domain. It meets the requirements of knowledge integration. The integrating models have a kernel consisting of concepts, relationships, and inference rules, which can be extended with other knowledge by integrating more components, kinds of facts, and kinds of rules into the kernel. Hence, a method for integration between the knowledge kernel and the knowledge of operators and functions was proposed. These models can represent real knowledge domains more clearly, naturally, and accurately, and therefore are applicable in practice.
These integrating models were applied to build IPSs in mathematical courses at universities, such as linear algebra and graph theory. The kernel of these models represented the concepts of matrixes and vectors, relationships, and inference rules on them. Based on this kernel, the integrating model with knowledge of operators was used to represent the knowledge base of linear algebra, consisting of additional knowledge about matrix operators, linear equations systems, and vector spaces. Besides this, this kernel was also integrated with the knowledge of functions and operators to represent and organize the knowledge base of graph theory. These knowledge bases were applied to build the IPSs for these courses. These systems supported students well in their study of these courses. They could solve common exercises in each course, and their reasoning and solutions were readable, step-by-step, and similar to human methods.
Our method for integrating the knowledge based on ontology can be applied to other knowledge domains, especially knowledge in natural sciences, such as mathematics, physics, and chemistry. Depending on the knowledge domains, knowledge bases may vary in their structure, as long as the structure is bounded to these following five elements: concepts, relationships, inference rules, operators, and functions. This setup makes the integration of multiple knowledge bases seamless and consistent.
In the future, we will research and develop more methods and techniques of knowledge integration that are applicable in more complex real-world knowledge domains. Problems using heuristics and sample problems [21] for reasoning on the integrated structure will also be studied. Additionally, with the IPS in education, after finding the solution of a problem, the system will omit the redundant steps, which are unnatural for humans, and wrap the solution steps into tutoring steps. This stage of the system uses the meta-knowledge of humans, especially the knowledge of the learner. In the next study, some meta-knowledge will be researched to process the final solutions of the system. This will make the solutions more natural and tend to the instructions for solving a problem.
Nowadays, there are many knowledge domains collected from many sources. Hence, a model to represent the combination of multiple knowledge domains is very necessary for knowledge integration, and a method to combine multiple knowledge models will have many applications in practice.

Conflicts of Interest:
The authors declare no conflict of interest.

Appendix A
The knowledge about linear algebra, including the chapters Matrixes-Vectors, Linear Equations System, and Vector Space was collected from [13,24]. It was represented by the model for integrating the kernel and knowledge of operators: (C, R, Rules ) + Ops C -set of concepts. Besides concepts in the kernel, C also includes concepts about Linear Equations System, and Vector Space: The structure of the EquationsSystem concept: • CoorMatrix V : 2 I Vector × 2 I Vector → I SquareMatrix (B 1 , B 2 ) # M CoorMatrix V is an operator to determine the matrix for converting the coordinate in a vector space V from based-set B1 to based-set B2: • Coor V : I Vector × 2 I Vector → I Vector ; ( Coor V is an operator to determine the coordinate of a vector v with based-set B in a vector space V.
Rules -set of rules on the knowledge domain about linear algebra.
• Rules for generating a new object:

Appendix B
The knowledge about undirected graphs in graph theory was collected from [25][26][27]. This knowledge base is represented by the method for integrating the kernel and knowledge of functions: (C, R, Rules ) + Funcs C -set of concepts: