KR20120082620A - Method for aligning ontology and ontology alignment system thereof - Google Patents

Abstract

PURPOSE: An ontology alignment method and an ontology alignment system thereof are provided to directly compare graph structures of an entity included in ontology, thereby improving performance through a decrease of information loss. CONSTITUTION: A service UI(User Interface) unit receives two ontologies. A graph extracting unit(210) extracts ontology graphs about the two ontologies. An ontology kernel(220) decides a similarity of the two ontologies by using the ontology graph and aligns the two based on the similarity. The graph extracting unit extracts sub graphs including structure information of entities of the two ontologies. The ontology kernel decides the similarity by measuring the similarity with the number of part structures shared by the sub graph of the two.

Description

Ontology Alignment Method and Ontology Alignment System Applying the Method {Method for Aligning Ontology and Ontology Alignment System}

The present invention relates to an ontology alignment method and an ontology alignment system using the same, and more particularly, to an ontology alignment method for aligning two ontology objects and an ontology alignment system applying the same.

In 1989 Tim Berners-Lee first proposed the World Wide Web. In particular, the client-server structure and the markup language of HTML (HyperText Markup Language), which are widely used, can upload personal information to the web anywhere in the Internet environment and access the information through a browser. We have a shared infrastructure for sharing information.

As a result, a great deal of information has been put on the Internet and distributed, and a great deal of information has existed on the Internet. The sharing of this information promotes social and technological development, and as a result, leads the innovation of the information society. Became.

However, the enormous amount of information has led to an increasing amount of effort and time to find the information desired by the user, and the emergence of various applications and services using the web has enabled this to be used effectively and appropriately. Many difficulties have arisen. In particular, the existing web-based retrieval method is mainly a keyword search, the method of determining the priority of the web document mainly using the frequency of words or vocabulary information. Therefore, there is a limit in finding a web document desired by a user. In particular, it has been very difficult to expand, integrate and share related web documents.

This problem occurs because the existing web and markup languages are human-centered and focus on the expression technology of the web browser for humans to see and understand. As a result, the existing web is a human-centered information processing technology that does not provide enough functions for a computer to effectively extract, interpret, and process necessary information on its own.

The Semantic Web is a technology that can extend the existing web to realize semantic interoperability based on well-defined meanings that computers can understand and to build an effective cooperative system between humans and computers. This appeared.

The semantic web is not a concept of a new web that is completely different from the existing web, but it is a program that empowers the current web to give well-defined meaning to information on the web, which enables computers and people to work collaboratively. Play a role. The semantic web is not only for humans but also for machines (computers) to grasp the meaning of information on the web to provide intelligent services that meet users' needs, or to facilitate collaboration between people and machines or machines and machines. It is a web with automatic service.

In addition, the semantic web is a next-generation web technology that allows computers to understand, automate, integrate, and reuse the meaning of information resources, such as ontology, semantic annotated web, and agent. Consists of

Ontology, on the other hand, is a formal specification system for shared conceptualization and provides semantic information of domain vocabulary. Ontology is a kind of knowledge expression, and the computer can understand the concept represented by the ontology and process the knowledge.

This ontology requires applications on the web to use the same ontology to communicate with each other. However, even applications belonging to the same domain use ontologies built individually. Therefore, in order to increase interoperability among applications, there is a necessity of combining ontologies separately constructed, which is called ontology alignment.

In the existing ontology alignment method, only part of a predefined feature or structure is extracted and used. In this case, there is a disadvantage that causes loss of information and is not easy to apply to all the ontology, this disadvantage caused a degradation of performance.

Therefore, the search for an ontology alignment method that can minimize the loss of structural information is required.

The present invention has been made to solve the above problems, an object of the present invention by directly comparing the graph structure of the ontology objects, ontology alignment method and minimized ontology alignment method to minimize the loss of structure information when ontology alignment and ontology alignment applied thereto In providing a system.

Ontology alignment method according to an embodiment of the present invention for achieving the above object, the step of receiving two ontologies; Extracting an ontology graph for the two ontologies; Determining similarity of the two ontologies by using the extracted ontology graph; And arranging the two ontologies based on the determined similarity.

The extracting may include extracting a subgraph including structure information of the entities of the two ontology.

In the determining, the similarity may be determined by measuring the similarity with the number of substructures shared by the subgraphs of the two ontologies.

The substructure may be a node-edge-node sequence having a specific length shown in the graph.

The determining may further include comparing whether two nodes or edges of the two ontology are of the same type.

And, in the sorting step, when the objects of the two ontologies are 1: 1, the objects having the highest similarity value are sorted, and the objects of the two ontology are 1: n (where n is a natural number other than 1). ), All objects with a value higher than the similarity value entered by the user can be sorted.

The method may further include outputting an alignment result of the two ontologies in an RDF form.

On the other hand, the ontology alignment system according to an embodiment of the present invention for achieving the above object, the service UI unit for receiving two ontologies; A graph extraction unit for extracting an ontology graph for the two ontologies; And an ontology kernel unit that determines the similarity of the two ontologies using the extracted ontology graph and aligns the two ontology based on the determined similarity.

The graph extractor may extract a subgraph including structure information of the entities of the two ontology.

The ontology kernel unit may determine the similarity by measuring the similarity with the number of substructures shared by the subgraphs of the two ontologies.

The substructure may be a node-edge-node sequence having a specific length shown in the graph.

In addition, the ontology kernel unit may determine whether two nodes or edges of the two ontology are the same type to determine the similarity between the two ontology.

The ontology kernel unit aligns the entities having the highest similarity value when the entities of the two ontology are 1: 1, and the entities of the two ontology are 1: n (where n is a natural number other than 1). In the case of, you can sort all objects having a value higher than the similarity value entered by the user.

The apparatus of claim 8, further comprising a result generator outputting the alignment results of the two ontology in RDF form.

According to various embodiments of the present invention described above, by directly comparing the graph structure of the objects included in the ontology, information loss can be reduced, thereby helping to improve performance.

1 is a simplified block diagram of an ontology alignment system according to one embodiment of the invention,
2 is a block diagram of an ontology alignment module according to an embodiment of the present invention.
3 is a diagram illustrating an ontology graph according to an embodiment of the present invention, and
4 is a flowchart illustrating an ontology alignment method according to an embodiment of the present invention.

Hereinafter, with reference to the drawings will be described in detail with respect to the present invention.

1 is a simplified block diagram of an ontology alignment system 10, in accordance with an embodiment of the present invention. As shown in FIG. 1, the ontology alignment system 10 according to an exemplary embodiment of the present invention includes a service UI unit 100 of a client part and an ontology alignment module 200 of a network and a server part. do.

The service UI 100 receives a user input of an internet service to which an ontology sorting method using an ontology kernel is applied. In particular, the service UI unit 100 receives two ontologies that are the ontology alignment targets from the user and transmits them to the ontology alignment module 200.

In addition, the service UI unit 100 serves to present the alignment result transmitted from the ontology alignment module 200 to the user.

In addition, the service UI unit 100 may be implemented as a touch screen. However, this is only one embodiment and may be implemented by other input methods.

The network connects the service UI unit 100 of the client part with the ontology alignment module 200 of the server part. The network may be implemented with the Internet. This is because the ontology is basically defined so that it can be operated in a web environment where the Internet address can be easily entered.

The ontology sorting module 200 receives two ontology inputs by the user through the service UI unit 100, extracts a graph, measures similarities between objects through the extracted graph, sorts them, and generates a final result. It serves to output to the service UI unit 100 again. Hereinafter, the ontology alignment module 200 according to an embodiment of the present invention will be described with reference to FIG. 2.

2 is a block diagram of the ontology alignment module 200 according to an embodiment of the present invention. As shown in FIG. 2, the ontology sorting module 200 includes a graph extractor 210, an ontology kernel unit 220, and a result generator 230.

The graph extractor 210 extracts graph information of two input ontology. In particular, the graph extractor 210 extracts a subgraph including the structural information of the objects of the two ontology. At this time, the entity of the ontology may be a concept or a property. In detail, the graph extractor 210 extracts the ontology graph as shown in FIG. 3. In this case, the extracted graph includes a concept 310 and a property 320 which are entities of the ontology. The graph extractor 210 extracts the subgraph 330 which is structural information of the specific entity.

The ontology kernel unit 220 determines the similarity of the two ontologies using the extracted ontology graph and arranges the two ontology based on the determined similarity.

In detail, the ontology kernel unit 220 uses an ontology kernel to determine the similarity of the ontology. At this time, the ontology kernel is a kind of convolution kernel proposed to measure the similarity between the graphs of the ontology. When the ontology graph is compared, the ontology kernel measures the similarity by the number of sub-structures shared by the two ontology graphs. In particular, the present invention uses a random walk as a partial structure. In this case, the random walk refers to a node-edge-node sequence of a specific length shown in the graph.

If this is expressed as an equation, Equation 1 below.

As shown in Equation 1, it can be seen that similarity is obtained by comparing all random walks for a given graph G1 and G2. At this time, the function K _walk for comparing the random walks compares the random walks by dividing them into steps, which are smaller structures. The step means a random walk of length 1 composed of node-edge-node. Equation 2 shows a function for comparing similarities in a step.

In Equation 2 above, n means the length of the random walk. The function K _step compares the similarity between two given steps. One step is represented by the i th node and the i + 1 th node in the equation. In this case, the similarity between steps is obtained by comparing the edges between the two nodes forming the step. Equation 3 below represents a function for comparing the similarity by comparing the edge with the two nodes.

In Equation 3, K _node and K _edge are functions for comparing a node and an edge. In general graph kernel, these node and edge comparison functions are defined as string comparison functions, but this does not take into account the characteristics of the ontology. In particular, nodes and edges of the ontology graph have both structure information and language information, so both of them must be considered.

Hereinafter, a kernel for comparing a node and an edge is defined as in Equation 4.

는 그래프 커널에 수학식 4의 노드와 엣지 비교함수를 적용한 온톨로지 커널을 의미한다. 특히, 노드와 엣지를 비교할때, 구조 정보를 적용하면 모든 컨셉간 혹은, 프로퍼티간의 유사도를 한번에 계산할 수 없다. 따라서, 이를 해결하기 위해 t번동안 반복하여 계산하게 된다. 먼저 t=0인 시작시에는 기존의 스트링 비교함수를 이용하는 그래프 커널을 사용하여 유사도를 계산하고 계산된 유사도가

가 된다. 이를 바탕으로 t=1, t=2, ..., t=T까지 계산하여 최종 유사도를 측정하게 된다. 이때, T는 사용자에 의해 입력된 값이다.Function I of Equation 4 is an indicator function for comparing whether two given nodes and edges are the same type. The function I reflects the characteristics that the nodes and edges of the ontology graph have types such as concept and property. The function I returns 1 if the types are the same and 0 if the types are different. Kernel Function

Denotes an ontology kernel in which the node and edge comparison functions of Equation 4 are applied to the graph kernel. In particular, when comparing nodes and edges, applying structural information makes it impossible to calculate the similarity between all concepts or properties at once. Therefore, to solve this problem, it is calculated repeatedly for t times. First, at the beginning of t = 0, the similarity is calculated using the graph kernel using the existing string comparison function, and the calculated similarity

. Based on this, the final similarity is measured by calculating t = 1, t = 2, ..., t = T. In this case, T is a value input by the user.

As described above, when determining the similarity between the two ontology objects, the ontology kernel unit 220 arranges the determined similarity. At this time, there are two ways to align the two ontologies.

The first method is when two ontologies have a 1: 1 relationship. In this case, the concept and the property having the highest similarity value with the given concept or property are matched and sorted. This method can be used when the ontology of the same domain and the same content is assumed because it is assumed that the relationship between the entities shown in the two ontology is 1: 1.

In the second method, there is a case where two ontologies are matched in a 1: n relationship, that is, a given ontology object and several objects of a relative ontology can be matched. For example, there are two individuals named "people" in ontology A and "people" and "women" in ontology B. These entities are matched 1: 2. If the objects of the two ontologies are 1: n, the ontology kernel unit 220 may match and sort all objects having similarities greater than a value input by the user. This method is generally available when the A ontology and the B ontology share content for similar areas. For example, if the A ontology describes the content of a company and the B ontology contains the content of tourism, it is difficult for the two ontologies to have a 1: 1 relationship, but it is related to the two ontologies. This is useful for determining similarity to.

The ontology kernel unit 220 generates the similarity between the objects calculated above as a matrix. For example, when n and m objects are aligned in two ontologies, a similarity matrix of size n x m is generated. In this case, the similarity is normalized from 0 to 1 to determine the similarity of the individuals in the same criterion.

The result generator 230 finds the entity pairs to be aligned using the similarity matrix sheared from the ontology kernel unit 220 and transmits them to the service UI unit 100 in the form of RDF (Resource Description Framework). As described above, when the objects of the two ontology are 1: 1, the final sorting object is the object having the highest similarity, and when the objects of the two ontology are 1: n, the final sorting object is the similarity value set by the user. All these objects become.

As described above, by directly comparing the graph structure of the objects included in the ontology, it is possible to reduce the loss of information, thereby helping to improve performance.

Hereinafter, an ontology alignment method will be described with reference to FIG. 4. 4 is a flowchart illustrating an ontology alignment method according to an embodiment of the present invention.

First, the ontology alignment system 10 receives two ontology to be aligned (S410).

When two ontologies are input, the ontology alignment system 10 extracts a graph of two ontologies (S420). In particular, the ontology alignment system 10 extracts a subgraph that includes structural information of the objects of the two ontology. At this time, the entity of the ontology may be a concept or a property.

The ontology alignment system 10 determines the similarity between individuals of the two ontologies using the subgraph of the extracted individual, and arranges the two ontology based on the determined similarity (S430). Specifically, the ontology alignment system 10 uses an ontology kernel to determine the similarity of the ontology. At this time, the ontology kernel is a kind of convolution kernel proposed to measure the similarity between the graphs of the ontology. When the ontology graph is compared, the ontology kernel measures the similarity by the number of sub-structures shared by the two ontology graphs. At this time, the partial structure in the present invention can be obtained by using a random walk.

  Upon determining the similarity between the two ontology individuals, the ontology alignment system 10 sorts the determined similarities. At this time, there are two ways to align the two ontologies. The first method is when two ontologies have a 1: 1 relationship. In this case, the concept and the property having the highest similarity value with the given concept or property are matched and sorted. The second method is when two ontologies have a 1: n relationship. In this case, all objects having similarities greater than the value input by the user may be matched and sorted. The ontology sorting system 10 generates the sorted result as a similarity matrix.

The ontology alignment system 10 outputs the generated similarity matrix to the user through the service UI unit 100 (S440).

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is clearly understood that the same is by way of illustration and example only and is not to be construed as limiting the scope of the invention as defined by the appended claims. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.

100: service UI unit 200: ontology alignment module
210: graph extraction unit 220: ontology kernel unit
230: result generator

Claims (14)

In ontology sorting method,
Receiving two ontologies;
Extracting an ontology graph for the two ontologies;
Determining similarity of the two ontologies by using the extracted ontology graph;
And aligning the two ontologies based on the determined similarity. The method of claim 1,
The extracting step,
Ontology sorting method characterized in that for extracting a subgraph containing the structural information of the objects of the two ontologies. The method of claim 2,
The determining step,
And determining similarity by measuring similarity with the number of substructures shared by the subgraphs of the two ontologies. The method of claim 3,
The substructure,
Ontology sorting method characterized in that the node-edge-node sequence of a particular length shown in the graph. The method of claim 4, wherein
The determining step,
Comparing whether two nodes or edges of the two ontologies are the same type; Ontology alignment method further comprising. The method of claim 2,
The sorting step,
If the two ontologies have a 1: 1 relationship, sort the entities having the highest similarity value,
If the objects of the two ontologies are 1: n relationship, where n is a natural number other than 1, the ontology sorting method characterized in that to sort all the objects having a value higher than the similarity value input by the user. The method of claim 1,
And outputting an alignment result of the two ontologies in the form of a resource description framework (RDF). In ontology alignment system,
A service UI unit for receiving two ontologies;
A graph extraction unit for extracting an ontology graph for the two ontologies;
Ontology alignment system comprising; an ontology kernel unit for determining the similarity of the two ontologies using the extracted ontology graph, and aligning the two ontologies based on the determined similarity. The method of claim 8,
The graph extraction unit,
Ontology sorting system, characterized in that for extracting a subgraph containing the structural information of the objects of the two ontologies. The method of claim 9,
The ontology kernel unit,
And measuring similarity by the number of substructures shared by the subgraphs of the two ontology. The method of claim 10,
The substructure,
Ontology alignment system, characterized in that the node-edge-node sequence of a particular length shown in the graph. The method of claim 11,
The ontology kernel unit,
And comparing the two nodes or edges of the two ontologies with the same type to determine the similarity between the two ontologies. 10. The method of claim 9,
The ontology kernel unit,
If the two ontologies have a 1: 1 relationship, sort the entities having the highest similarity value,
The ontology alignment system of claim 2, wherein when the objects of the two ontology are in a relationship of 1: n (where n is a natural number other than 1), all objects having a value higher than a similarity value input by the user are sorted. The method of claim 8,
And a result generator for outputting the sorting results of the two ontologies in the form of RDF (Resource Description Framework).

Images