KR101553397B1 - Apparatus and method for matching large-scale biomedical ontology - Google Patents

Abstract

A large scale ontology matching apparatus according to the present invention includes a preprocessing unit for generating a subset of an ontology by classifying received candidate ontologies into at least one ontology subset, a subset of the generated ontology subset using a distribution algorithm, And transmits the generated matching threads to the individual cores of the participating nodes, and a matching processing unit for collecting and summing the matching results generated by performing matching operations based on the matching threads in the individual cores, And a collecting section for generating the collecting section.

Description

[0001] APPARATUS AND METHOD FOR MATCHING LARGE-SCALE BIOMEDICAL ONTOLOGY [0002]

The present invention relates to ontology matching in a large scale biomedical field, and more particularly to parallel processing and distributed processing of ontology matching.

Recently, Semantic Web technology has been applied to information processing and information management in the biomedical field, which has brought a lot of benefits to the biomedical system. In particular, ontology technology in biomedical systems is widely used for standardization of biomedical information, knowledge sharing, and reusability. As a result, it is possible to use the Gene Ontology (GO), the National Cancer Institute Thesaurus (NCI), the Foundation Model of Anatomy (FMA), and the Systemic Nomenclature of Medicine Clinical Terms Ontology is applied to biomedical field in the form. In this way, various studies are being conducted to apply ontology more effectively to the biomedical field, and researches are being conducted to provide continuity to the already developed ontology.

Ontology in the biomedical field is very complex because it has a large scale, which is an obstacle to integration and interoperability. OBO (Open Biomedical Ontologies) consortium is trying to solve these obstacles through the Introducing Strategy for ontology evolution. These biomedical ontologies include overlapping information. The overlapping information is the information necessary for the integration of biomedical systems and the interoperability of information processing. In ontology, the relationship between different candidate ontologies is called mapping or alignment. The candidate ontology is an ontology for the mapping process that establishes the relationship between the pipes. The process of mapping discovery, which establishes the relationship between the different ontologists, is called ontology matching.

The ontology matching process for finding the ontology mapping on a large scale biomedical ontology requires computational complexity and computational overhead. Ontology matching is computed as a Cartesian product of two candidate ontologies, as shown in "On Matching Large Life Science Ontologies in Parallel" in 2010 Data Integration in the Life Sciences. This task requires a resource-based matching algorithm. In this mapping process, an excessive computation operation due to the second-order computational complexity may cause a delay, and the delay of the mapping result ineffectively processing the ontology mapping for the biomedical system within the processing request time .

Anika Groß, Michael Hartung, Toralf Kirsten, Erhard Rahm. "On Matching Large Life Science Ontologies in Parallel". Data Integration in the Life Sciences, Lecture Notes in Computer Science Volume 6254, 2010, pp 35-49.

An object of the present invention is to provide an ontology matching apparatus and method for performing parallel matching for improving performance on multiple cores in an ontology matching process for ontology mapping in an ontology in a large scale biomedical field.

A large scale ontology matching apparatus according to the present invention includes a preprocessing unit for generating a subset of an ontology by classifying received candidate ontologies into at least one ontology subset, a subset of the generated ontology subset using a distribution algorithm, And transmits the generated matching threads to the individual cores of the participating nodes, and a matching processing unit for collecting and summing the matching results generated by performing matching operations based on the matching threads in the individual cores, And a collecting section for generating the collecting section. The ontology storage unit may further include an ontology storage unit that stores the serialized ontology subset and provides the pre-stored serialized ontology subset to the preprocessor when the same candidate ontology as the received candidate ontology is received.

The preprocessor can generate a serialized ontology subset by serializing the created ontology subset to a binary form and reconstruct the serialized ontology subset received from the ontology storage through de-serialization You can create an ontology subset. The matching thread generated in the distributed processing unit may include one or more matching requests, one or more matching jobs, and one or more matching tasks. The matching request corresponds to each of the participating nodes on a one-to-one basis, and the matching operation corresponds one-to-one to the individual cores provided in the corresponding participating node on a one-to-one basis in the matching request. The maximum number of one or more matching requests does not exceed the number of participating nodes and the maximum number of one or more matching operations does not exceed the number of individual cores provided throughout the participating nodes. The matching operation then operates independently of other matching operations and matching requests that operate on other participating nodes.

When the candidate ontology is received, the distributed processing unit confirms the number of participating nodes and the number of individual cores included in the participating node, sets the number of distributed nodes by applying the number of participating nodes and individual cores to the distributed algorithm, , And generates a matching thread including a matching request, a matching operation, and a matching task by dividing an ontology subset and applying a matching algorithm.

In the large-scale ontology matching method according to the present invention, the received candidate ontology is classified into one or more ontology subsets. Then, the generated ontology subset is divided by applying a distributed algorithm, and a matching algorithm is applied to the divided ontology subset to generate a matching thread. The matching thread is created and the generated matching thread is delivered to the individual cores of the participating nodes. Next, an ontology mapping is generated by collecting and summing the matching results generated by performing matching operations based on the matching threads in the individual cores.

The large scale ontology matching method according to the present invention confirms the number of participating nodes and the number of individual cores included in participating nodes and sets the number of dispersions by applying the number of participating nodes and individual cores to the decentralization algorithm. Then, the ontology subset is divided in consideration of the set number of distributions, and a matching algorithm is applied to generate a matching thread including a matching request, a matching operation, and a matching task.

The large scale ontology matching apparatus and the matching method according to the present invention can distribute the ontology matching operation to the individual cores of the participant nodes and perform parallel processing to effectively reduce the operation resources and computation time required for the matching operation.

1 is a block diagram showing an embodiment of a biomedical ontology according to the present invention.
FIG. 2 is a block diagram showing an embodiment of a large scale ontology matching apparatus according to the present invention.
3 is a detailed view showing a preprocessing unit of a large scale ontology matching apparatus according to an embodiment of the present invention.
4 is a detailed view showing a distributed processing unit of a large scale ontology matching apparatus according to an embodiment of the present invention.
5 is a diagram illustrating an embodiment of a matching thread that is transmitted to an individual core in a large scale ontology matching apparatus according to an embodiment of the present invention.
6 is a diagram illustrating a data flow of a large scale ontology matching apparatus according to an embodiment of the present invention.
7 is a flowchart illustrating a method of matching a large scale ontology according to an embodiment of the present invention.
8 is a flowchart illustrating a method of omitting a pre-processing of a large-scale ontology matching method according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The terms and words used in the present specification are selected in consideration of the functions in the embodiments, and the meaning of the terms may vary depending on the intention or custom of the invention. Therefore, the terms used in the following embodiments are defined according to their definitions when they are specifically defined in this specification, and unless otherwise specified, they should be construed in a sense generally recognized by those skilled in the art.

1 is a block diagram showing an embodiment of a biomedical ontology according to the present invention.

Referring to FIG. 1, in the present invention, a biomedical ontology is formed of one or more constituent elements 10 in which mutual relations are established. One or more components 10 establish the relationship between the individual component 10 and the corresponding component 10 through an ontology matching process and the result of the matching between the individual components 10 is collected to form an ontology mapping . The large scale ontology matching device utilizes the hardware of participating nodes to efficiently perform operations in a large scale ontology matching process. A node participating in a distributed environment may include one or more nodes that depend on a possible distributed environment. And the participating nodes may include both single-core or multi-core hardware. The large scale ontology matching apparatus allocates an ontology subset to a single-core processor or a multicore processor held by each node in a distributed environment including one or more nodes and processes the data by using a data parallelization technique.

The biomedical ontology belongs to the semantic-web ontology group as the general ontology. However, biomedical ontologies and generic ontologies differ in size and evolution. Generic ontologies are smaller in size than biomedical ontologies, and evolution is also slower. On the other hand, biomedical ontologies evolve more rapidly because of the rapid evolution in biomedical domains and biomedical data, and are therefore much larger.

In the distributed environment, ontology matching is a process of establishing or linking a matching relationship based on the concept between a source ontology and a target ontology. This process extends the ontology classification system . The source ontology is an ontology for establishing a new relationship, and the target ontology is an ontology that is an object to be matched with the source ontology through an ontology matching process.

The total number of matching operations between the source ontology and the target ontology is evenly distributed among available computational resources (cores) as shown in Equation (1).

MT represents a matching task of an individual, Os represents a source ontology, Ot represents a target ontology, MTcore represents a matching task of an individual core, m is a concept of a source ontology, n is a target It is the concept of ontology. The need for large-scale biomedical ontology matching is being addressed by several experts in biomedicine, including researchers, biomedical / bioinformatics systems, and even third-party healthcare information services running on cloud platforms. Lt; / RTI > source.

FIG. 2 is a block diagram showing an embodiment of a large scale ontology matching apparatus according to the present invention.

2, an apparatus 100 for matching a large scale ontology according to the present invention includes a preprocessing unit 110, an ontology storage unit 120, a distributed processing unit 130, and an aggregation unit 140.

The preprocessing unit 110 receives the candidate ontology including the source ontology and the target ontology. The candidate ontologies including the source ontology and the target ontology may match one to one with each other, one source ontology may match two or more target ontologies, or two or more source ontologies may match one target ontology.

The preprocessing unit 110 classifies the candidate ontology including the received source ontology and the target ontology into one or more subsets. An ontology can generally consist of five subsets. The subset that constitutes the ontology can be broadly classified into Name, Lable, Relationship, Axiom, and Property. However, the ontology can not be divided into only the five subset described above, and the preprocessing unit 110 can classify the ontology into one or more subsets according to the type and characteristics of the received candidate ontology. The pre-processing unit 110 transfers the classified ontology subset to the ontology storage unit 120 and the distributed processing unit 130. The preprocessing unit 110 may directly transmit the sorted ontology subset to the distributed processing unit 130. The preprocessing unit 110 also receives the ontology subset stored in the ontology storage unit 120 and transmits the received ontology subset to the distributed processing unit 130. [ . In particular, when a candidate ontology identical to an ontology subset previously classified and stored in the ontology storage unit 120 is received, the preprocessing unit 110 notifies the ontology storage unit 120 of the pre- The serialized and stored ontology subset is transferred to the distributed processing unit 130 without the preprocessing process, thereby reducing the computational resources consumed in the preprocessing process. The preprocessing unit 110 serializes the ontology subset to the ontology storage unit 120 and de-derializes the serialized ontology subset received from the ontology storage unit 120 To the distributed processing unit 130.

The ontology storage unit 120 serializes the ontology subset received from the preprocessing unit 110 and stores the serialized data. The ontology storage unit 120 can efficiently use the storage space by serializing the received ontology subset in a binary form and improve data processing efficiency. The pre-processing unit 110 previously stores the ontology storage unit 120, It is possible to reduce or avoid unnecessary preprocessing re-processing in the preprocessing unit 110 by transmitting the ontology subset stored in serial form in binary format according to the request of the preprocessing unit 110. [

The distributed processing unit 130 divides the ontology subset received from the preprocessing unit 110 into two or more matching threads in order to perform data parallel processing. Data parallelization requires each processing core to perform a matching operation on each piece of the candidate ontology segmented into two or more distributed ontologies. In order to enable parallel processing, the distributed processing unit 130 divides the ontology subset in consideration of the distribution algorithm. First, the distributed processing unit 130 determines the number of cores of a multi-core processor of an individual participating node. The distributed processing unit 130 divides the received ontology subset into predetermined distributed ontologies by applying the number of cores of the participating nodes and the individual participating nodes to the distributed algorithm.

The distributed processing unit 130 applies a matching library to one or more divided subset of ontologies based on the distribution algorithm to generate a matching request (MR), a matching job (MJ) And a matching task (hereinafter referred to as MT). Equation (2) represents a matching request (MR), a matching job (MJ), and a matching task (MT) of a distributed algorithm applied to the distributed processing unit Relationship.

In Equation (2), MR _i denotes a matching request received at each node, i denotes the number of participating nodes for parallel matching in the distributed environment, MR denotes the number of participating nodes received by the large scale ontology matching apparatus 100 _J represents a matching operation assigned to an individual core 20 constituting each participating node, j represents the number of individual cores 20 provided in the participating node, MT _k represents a matching operation for each participating node And k represents the number of matching tasks included in the matching task assigned to one individual core 20. In this case, And, m is the concept of the source ontology, and n is the concept of the target ontology. The single matching task MT _k is a Cartesian product of the concepts of the source ontology and the target ontology, respectively.

The matching request is a mathing thread that is transmitted in correspondence with each of the participating nodes (or the CPUs of the participating nodes) of the distributed environment. The distributed processing unit 130 firstly divides the ontology subset into one or more matching requests corresponding to the respective participating nodes (or the CPUs of the participating nodes) considering the number of the participating nodes (or the CPUs of the participating nodes). The maximum number of matching requests may be the number of participating nodes (or participating node's CPUs). That is, the matching request can be a parallel processing task that is assigned to an individual node for ontology parallel processing. The distributed processing unit 130 divides the divided matching request into matching jobs according to the number of cores of the participating nodes to which the corresponding matching requests are assigned, and includes the divided matching requests in the matching request. That is, if the matching operation is a parallel processing operation assigned to the individual core 20, and the number of cores of the participating node to which the matching request is assigned has four cores, the matching request may include four matching operations . Next, the distributed processing unit 130 divides the matching task allocated to the individual core into a matching task, which is a unit for performing actual matching processing, and incorporates the matching task into the matching task.

Matching requests, matching tasks, and matching tasks are three layers of abstraction for the entire matching process. The distribution algorithm of the distributed processing unit 130 must preserve the tracks of all the jobs being executed since the classification algorithm for the parallel processing processing must provide classification at different levels.

The matching task can be classified as the smallest overall matching process as a unit of the matching process. For example, if the source ontology is {A, B, C, D} and the target ontology is {a, b, c, d} in the matching task assigned to the core of the participating node, then compare the source ontology with the target ontology When performing the matching process, A = a is the first MT, B = b is the second MT, C = c is the third MT, D = d is the fourth MT, If possible, this could also be the other MT. In other words, the matching task is the smallest unit that processes the individual matching processing within the individual core of the participating node, and processes the matching processing of individual terms (Term) constituting the ontology within the core.

The matching operation may be a unit of one or more matching tasks assigned to the individual cores 20 of the participating nodes to perform the matching process. And, the matching request is a collection of all the matching tasks that are executed in one participating node (or CPU of the participating node).

For example, if four participating nodes are present in the matching process, all four participating nodes have four cores, and the received ontology subset requires a total of 2000 matches, then the distributed processing unit 130 will have a total of 2000 The matching is divided into 4 equal parts according to the number of participating nodes, and is divided into a total of 4 matching requests. Each of the four matching requests divided comprises 500 matches. The distributed processing unit 130 classifies each of the four matching requests into four matching operations in consideration of the number of cores. Then, each matching operation includes 125 matching, which is 500 divided by 4. As a result, one matching task includes 125 matches, and the number of matching tasks included in the individual matching task is 125. That is, each of the four matching requests includes four matching tasks, each of the four matching tasks including 125 matching tasks, so that one or more of the ontology subsets comprising a total of 2000 matches are divided into four matching requests , Each matching request includes four matching tasks, each matching task including 125 matching tasks.

The number of matching requests, matching tasks, and matching tasks can be set considering not only the number of participating nodes and individual participating nodes, but also scenarios that can perform the most optimized parallel processing. And each matching operation is independent of other matching operations and other matching requirements that operate remotely. Also, if the number of matching tasks included in one or more matching jobs is kept the same among one or more matching jobs in the process of subdividing the matching requests into matching jobs, most of the individual cores finish the matching operation at a similar time, Lt; RTI ID = 0.0 > (Idel). ≪ / RTI > The matching request, the matching operation and the matching task will be described again in FIG. 5 to be described later.

The distributed processing unit 130 transmits a matching request having a predetermined number of at least one ontology subset to the participating node through the above-described process. One matching request is forwarded to the corresponding one participating node. One matching request delivered to one participating node includes a matching operation corresponding to each of the one or more cores included in the participating node. One or more matching operations included in one matching request are assigned one-to-one to the corresponding cores. One matching task assigned to one core performs matching processing in the corresponding core through the included matching task. Accordingly, the large scale ontology matching apparatus 100 according to the present invention can parallel process the matching processing in each of the cores of the participating nodes.

The collecting unit 140 receives the matching result of performing the matching operation through the matching thread from the individual core 20 of the participating node that has received the matching thread from the distributed processing unit 130. [ The distributed processing unit 130 delivers a matching thread including a matching request, a matching operation, and a matching task to each core of the participating node. The individual core 20 of the participant node that receives the matching thread including the matching request, the matching operation, and the matching task from the distributed processing unit 130 performs the matching operation based on the received matching thread. The participating node receiving the matching request included in the matching thread assigns a matching operation to the individual core 20 provided according to the matching operation included in the matching request, and the individual core 20 performs matching By performing the task, the matching processing of the received candidate ontology is processed in parallel. The collection unit 140 collects matching results (operation results) generated by the ontology matching operation performed on the individual cores.

The concentrator 140 may receive the matching results generated by the individual cores 20 directly from the individual cores 20 or may output the matching results generated by the individual cores 20 to the individual cores 20, It can be received from the participating node at once.

The summation unit 140 sums up the matching results received from the individual cores 20 to generate an ontology mapping. The ontology mapping is a set of individual matching results between the components of two or more candidate ontologies. The accumulating unit 140 accumulates the matching result calculated by the calculation of the matching operation in the individual core 20 and applies a bridge pattern to the accumulated matching result to obtain a formal representation of the ontology mapping ). The bridge pattern is a type of mapping file, which is a custom defined format for mapping files according to individual users. For example, if you want a mapping file in XML format with a specific schema, it may be different from the mapping format defined by another user. All these individual formats are called bridge patterns. That is, the collection unit 140 can provide the ontology mapping according to the individual format requested by the individual user through the bridge pattern. Then, the collection unit 140 stores the used bridge pattern for another user.

The summation unit 140 may vary the summation process according to the number of participating nodes.

는 매칭 태스크의 매칭 결과(매칭 태스크의 브릿지 온톨로지)이고,

은 매칭 작업의 중앙 브릿지 온톨로지(Intermediate Bridge Ontology)이다. 두 번째 단계는 하나의 참여 노드 상에서 실행되는 모든 매칭 작업의 결과를 합산한다. 두 번째 단계에서,

는 단일 참여 노드에서 실행되는 모든 매칭 작업에서 모든 중앙 브릿지 온톨로지

의 합이다. 현재의 모든 노드에서 생성된 중앙 브릿지 온톨로지

는 노드로부터의 중앙 브릿지 온톨로지가 된다. 세 번째 단계는 분산 환경 상의 모든 참여 노드에서 실행되는 모든 매칭 요구의 모든 매칭 결과를 합산한다. 세 번째에서,

는 최종 생성되는 브릿지 온톨로지로서

는 출력 파일인 온톨로지 매핑 결과로 전환된다.Equation (3) represents three steps for generating a bridge ontology in a distributed environment in which two or more participating nodes are connected. The first step collects all the results of all single matching tasks in all matching jobs. In the first step,

(The bridge ontology of the matching task) of the matching task,

Is the intermediate bridge ontology of the matching operation. The second step sums the results of all matching operations running on one participating node. In the second step,

In all matching operations running on a single participating node, all central bridge ontologies

. The central bridge ontology generated from all current nodes

Is the central bridge ontology from the node. The third step is to sum all the matching results of all matching requests running on all participating nodes in the distributed environment. In the third,

Is the final generated bridge ontology

Is converted to an ontology mapping result which is an output file.

를 통합하며, 매칭 태스크의 결과

를 통합 결과는 매칭 작업의 중앙 브릿지 온톨로지

이다. 두 번째 단계는 하나의 참여 노드에서 실행되는 모든 매칭 작업의 결과를 합산한다. 단일 참여 노드에서 실행되는 매칭 작업에서 모든 중앙 브릿지 온톨로지

의 합은 노드로부터의 중앙 브릿지 온톨로지

가 된다.Equation (4) represents two steps for generating a bridge ontology in a distributed environment in which one participant node is connected. If there is one participating node, the result of a matching operation performed on the individual core 20 of a single participating node (CPU) must be summed. The first step is the result of all the matching tasks running in a single matching task

, And the result of the matching task

The result of the integration is the central bridge ontology of the matching operation

to be. The second step sums the results of all matching operations performed on one participating node. In a matching operation running on a single participating node, all the central bridge ontologies

Lt; RTI ID = 0.0 > ontology < / RTI &

.

The matching results computed by all participating nodes participating in the distributed environment are summed together with one ontology object called the bridge ontology. Then, the collection unit 140 converts the collected bridge ontology into an ontology mapping result, which is a physical mapping file, and outputs the result.

3 is a detailed view showing a preprocessing unit of a large scale ontology matching apparatus according to an embodiment of the present invention.

3, the preprocessing unit 110 of the large scale ontology matching apparatus 100 according to an embodiment of the present invention includes an ontology model unit 111, a serialization processing unit 112, and a deserialization processing unit 113 do.

The ontology model unit 111 receives a candidate ontology including a source ontology and a target ontology. The candidate ontologies including the source ontology and the target ontology may match one to one with each other, one source ontology may match two or more target ontologies, or two or more source ontologies may match one target ontology.

The ontology modeling unit 111 classifies the candidate ontology including the received source ontology and the target ontology into one or more subsets. An ontology can generally consist of five subsets. The subset that constitutes the ontology can be broadly classified into Name, Lable, Relationship, Axiom, and Property. However, the ontology can not be divided into only the five subset described above, and the ontology model unit 111 can classify it into one or more subsets regardless of the number according to the type and characteristics of the received candidate ontology. Also, the ontology modeling unit 111 may classify the candidate ontology into one or more subsets in consideration of the matching library including the matching algorithm. The ontology model unit 111 delivers the classified ontology subset to the distributed processing unit 130. [

The serialization processing unit 112 serializes the ontology subset generated by the ontology model unit 111. The serialization processing unit 112 can efficiently use the storage space by serializing the ontology subset classified into two or more subsets into a binary form, thereby improving the efficiency of data transmission and processing. The serialization processing unit 112 may transmit the serialized ontology subset to the ontology storage unit 120 and store the subset. The ontology storage unit 120 can efficiently use the storage space by storing the serialized ontology subset, thereby improving the data processing efficiency.

The deserialization processor 113 reconstructs the serialized ontology subset stored in the ontology storage unit 120 in the serialized form by the serialization processing unit 112 into the original ontology subset state again. Through this process, the deserialization processing unit 113 transfers the ontology subset, which is serialized and stored in the binary format, to the ontology storage unit 120 according to the request of the ontology model unit 111, Re-processing of the preprocessing process can be reduced or avoided. That is, if a candidate ontology that is the same as the ontology subset previously stored in the ontology storage unit 120 is received, the ontology subsets stored in the ontology storage unit 120 are stored in the ontology storage unit 120 without repeating the pre- (Reconstructed) through the deserialization processing unit 113 and can transfer the reconstructed data to the distributed processing unit 130. By skipping the preprocessing process through such a process, the computational resources consumed in the preprocessing process can be reduced.

The preprocessing process of the preprocessing unit 110, which subdivides the received candidate ontologies by subset to generate a subset ontology, enables less memory footprint during execution time.

4 is a detailed view showing a distributed processing unit of a large scale ontology matching apparatus according to an embodiment of the present invention.

4, the distributed processing unit 130 of the apparatus for matching a large scale ontology according to an embodiment of the present invention includes a parallel strategy unit 131, a distribution unit 132, and a parallel interface 133.

The parallel strategy unit 131 confirms the number of participating nodes and the number of individual cores (multicore) included in each participating node. In the distributed environment, a large number of participating nodes participate. The participating node may include a terminal or an apparatus capable of performing arithmetic processing such as a workstation and a server computer, as well as a personal computer such as a commonly used desktop and a notebook. The large scale ontology matching apparatus according to the present invention distributes the ontology matching for the mapping of the candidate ontology to the individual cores of the participating nodes and processes them in parallel. Accordingly, the parallel scheduling unit 131 first ascertains the number of participating nodes and the number of individual cores included in each attached node, and grasps available computing resources. Then, the parallel strategy unit 131 sets the optimal number of dispersions considering the number of identified participating nodes and individual cores, and the distribution algorithm.

The distributing unit 132 divides the ontology subset in consideration of the number of participating nodes and the individual cores and the optimal number of dispersions considering the dispersion algorithms in the parallel strategy unit 131, To generate a matching thread. The generated matching thread includes a matching request, a matching operation, and a matching task. The matching algorithm stored in the matching library is an algorithm for computing the matching between the individual components constituting the candidate ontology to generate the ontology mapping between the candidate ontologies. The matching algorithm includes a synonym-based matching, a label- based matching, broader term-based matching, and child-based matching, which can be applied to ontology matching.

The matching request generated by the distributing unit 132 is a mathing thread transmitted corresponding to each of the participating nodes (or the CPUs of the participating nodes). The parallel strategy unit 131 first divides the ontology subset into one or more matching requests corresponding to each participating node (or the CPU of the participating node) considering the number of participating nodes (or CPUs of the participating nodes). The maximum number of matching requests may be the number of participating nodes (or participating node's CPUs). That is, the matching request can be a parallel processing task that is assigned to an individual node for ontology parallel processing. The distributing unit 132 divides the divided matching request into a matching operation according to the number of cores of the participating node to which the matching request is assigned, and includes the divided matching request in the matching request. That is, if the matching operation is a parallel processing task assigned to an individual core, and the number of cores of the participating node to which the matching request is assigned has four cores, the matching request may include four matching operations. Next, the distributing unit 132 divides the matching task allocated to the individual core into a matching task, which is a unit for performing actual matching processing, and includes the matching task in the matching task.

Matching requests, matching tasks, and matching tasks are three layers of abstraction for the entire matching process. Since the classification algorithms for parallel processing must provide classification at different levels, the distribution algorithm of the distributor 132 must preserve the tracks of all running tasks.

The matching task can be classified as the smallest overall matching process as a unit of the matching process. For example, if the source ontology is {A, B, C, D} and the target ontology is {a, b, c, d} in the matching task assigned to the core of the participating node, then compare the source ontology with the target ontology When performing the matching process, A = a is the first MT, B = b is the second MT, C = c is the third MT, D = d is the fourth MT, If possible, this could also be the other MT. In other words, the matching task is the smallest unit that processes the individual matching processing within the individual core of the participating node, and processes the matching processing of individual terms (Term) constituting the ontology within the core.

A matching task is a unit that is assigned to an individual core of a participating node and performs a matching process, and may be a collection of one or more matching tasks. And, the matching request is a collection of all the matching tasks that are executed in one participating node (or CPU of the participating node).

For example, if four participating nodes are in a distributed environment, all four participating nodes have four cores, and the received ontology subset requires a total of 2000 matches, The matching is divided into 4 equal parts according to the number of participating nodes, and is divided into a total of 4 matching requests. Each of the four matching requests divided comprises 500 matches. Then, the distributing unit 132 classifies each of the four matching requests into four matching jobs in consideration of the number of cores. Then, each matching operation includes 125 matching, which is 500 divided by 4. As a result, one matching task includes 125 matches, and the number of matching tasks included in the individual matching task is 125. That is, each of the four matching requests includes four matching tasks, each of the four matching tasks including 125 matching tasks, so that one or more of the ontology subsets comprising a total of 2000 matches are divided into four matching requests , Each matching request includes four matching tasks, each matching task including 125 matching tasks. The number of matching requests, matching tasks, and matching tasks can be set considering not only the number of participating nodes and individual participating nodes, but also scenarios that can perform the most optimized parallel processing. And each matching operation is independent of other matching operations and other matching requirements that operate remotely.

The parallel interface 133 conveys a matching thread corresponding to each participating node and each individual core of each participating node. One matching request is forwarded to the corresponding one participating node. One matching request delivered to one participating node includes a matching operation corresponding to each of the one or more cores included in the participating node. One or more matching operations included in one matching request are assigned one-to-one to the corresponding cores. One matching task assigned to one core performs matching processing in the corresponding core through the included matching task. Accordingly, the large scale ontology matching apparatus according to the present invention can parallel process the matching processing in each of the cores of the participating nodes constituting the distributed environment.

5 is a diagram illustrating an embodiment of a matching thread that is transmitted to an individual core in a large scale ontology matching apparatus according to an embodiment of the present invention.

Referring to FIG. 5, the matching thread generated in the distributed processing unit of the large scale ontology matching apparatus according to the present invention may include a matching request, a matching operation, and a matching task corresponding to each node and each individual node. For example, assume that three participating nodes including a first node 510, a second node 520, and a third node 530 are connected to a distributed environment. The first node 510 includes two cores 511 and 512 and the second node 520 includes four cores 521,522,523 and 524 and the third node 530 includes two cores 531 and 532 . The large scale ontology matching apparatus generates two or more matching threads from a candidate ontology including a source ontology and a target ontology through a preprocessing process and a distribution process, and delivers the generated two or more matching threads to the individual cores of the participating nodes.

The matching request is a matching thread that is delivered corresponding to each of the participating nodes (or the CPUs of the participating nodes). The distributed processing unit 130 firstly divides the ontology subset into one or more matching requests corresponding to the respective participating nodes (or the CPUs of the participating nodes) considering the number of the participating nodes (or the CPUs of the participating nodes). The maximum number of matching requests may be the number of participating nodes (or participating node's CPUs). That is, the matching request can be a parallel processing task that is assigned to an individual node for ontology parallel processing. The large scale ontology matching apparatus divides the divided matching request into matching operations according to the number of cores of the participating nodes to which the corresponding matching request is assigned, and includes the divided matching requests in the matching request. That is, if the matching operation is a parallel processing task assigned to an individual core, and the number of cores of the participating node to which the matching request is assigned has four cores, the matching request may include four matching operations. Next, the distributed processing unit 130 divides the matching task allocated to the individual core into a matching task, which is a unit for performing actual matching processing, and incorporates the matching task into the matching task.

5, the large scale ontology matching apparatus classifies the candidate ontology into a subset and applies three matching requests MR1, MR2, and MR3 by applying a distribution algorithm and a matching algorithm. To create a matching thread. The first matching request MR1 corresponds to the first node 510 and the second matching request MR2 corresponds to the second node 520 and the third matching request MR3 corresponds to the third node 530, Respectively. The first matching request MR1 includes two matching operations 511-1 and 512-1 corresponding to the two cores 511 and 512 of the first node 510 and the second matching request MR2 includes 522-1, 523-1, and 524-1 corresponding to the four cores 521, 522, 523 and 524 of the second node 520 and the third matching request MR3 includes the third node 530, And two matching operations 531-1 and 532-1 corresponding to the two cores 531 and 532 provided in the first and second cores 531 and 532, respectively.

The matching operations 511-1, 512-1, 521-1, 522-1, 523-1, 524-1, 531-1, and 532-1 included in each of the matching requests MR1, MR2, MR, 510, 520, 530) and individual cores 511, 512, 521, 522, 523, 524, 531, 532 of each participant node. The generated first to third matching requests MR1, MR2 and MR3 are respectively transmitted to the corresponding first node 510, the second node 520 and the third node 530, respectively. The process of transmitting the matching request to each node in the large scale ontology matching apparatus can be directly transferred to each node from the large scale ontology matching apparatus and is transferred from the large scale ontology matching apparatus to one node according to the structure of the distributed environment , And may be sequentially transmitted along the connected nodes.

6 is a diagram illustrating a data flow of a large scale ontology matching apparatus according to an embodiment of the present invention.

Referring to FIG. 6, in the data flow of the large scale ontology matching apparatus according to an embodiment of the present invention, the preprocessing unit 110 receives a candidate ontology including a source ontology and a target ontology (S601). The candidate ontologies including the source ontology and the target ontology may match one to one with each other, one source ontology may match two or more target ontologies, or two or more source ontologies may match one target ontology.

Then, the preprocessing unit 110 classifies the candidate ontology including the received source ontology and the target ontology into one or more subsets (S602). An ontology can generally consist of five subsets. The subset that makes up the ontology can be broadly divided into name, label, relation, axiom and property. However, the ontology can not be divided into only the five subset described above, and the preprocessing unit 110 can classify the ontology into one or more subsets according to the type and characteristics of the received candidate ontology. Then, the preprocessing unit 110 transfers the classified ontology subset to the distributed processing unit 130 (S603).

Next, the preprocessing unit 110 serializes the classified ontology subset (S604). The preprocessing unit 110 serializes the ontology subset in a binary form to increase data processing efficiency and storage space efficiency. The preprocessing unit 110 transfers the serialized ontology subset to the ontology storage unit 120 and stores it (S605). Thereafter, when the same candidate ontology is received, the corresponding ontology subset stored in the ontology storage unit 120 is called and used. This saves time and computational resources by omitting the preprocessing process of subsetting.

When the ontology subset is transferred from the pre-processing unit 110 to the distributed processing unit 130 in step S603, the distribution processing unit 130 checks the number of the participating nodes 30 and the number of individual cores included in each participating node 30 (S606). The distributed processing unit 130 first determines the number of participating nodes 30 and the number of individual cores included in each attached node, and grasps available computing resources. Then, the dispersion processing unit 130 sets the optimal number of dispersions considering the number of the participating nodes 30 and the individual cores and the dispersion algorithm that have been confirmed.

The distributed processing unit 130 divides the ontology subset in consideration of the number of distributed nodes and the number of individual cores and the optimal distribution number set in consideration of the distribution algorithm (S607), and compares the divided ontology subset with the matching library matching algorithm To generate a matching thread (S608). The generated matching thread includes a matching request, a matching operation, and a matching task. The matching algorithm stored in the matching library is an algorithm for computing the matching between the individual components constituting the candidate ontology to generate the ontology mapping between the candidate ontologies. The matching algorithm includes a synonym-based matching, a label- based matching, broader term-based matching, and child-based matching, which can be applied to ontology matching.

The distributed processing unit 130 first divides the ontology subset into one or more matching requests corresponding to the respective participating nodes 30 in consideration of the number of participating nodes 30. The maximum number of matching requests may be the number of participating nodes 30. The distributed processing unit 130 divides the divided matching request into a matching operation according to the number of cores of the participating node 30 to which the corresponding matching request is assigned, and includes the divided matching request in the matching request. Matching requests, matching tasks, and matching tasks are three layers of abstraction for the entire matching process. Since the classification algorithms for parallel processing must provide classification at different levels, the distribution algorithm of the distributor 132 must preserve the tracks of all running tasks. The number of matching requests, matching tasks, and matching tasks may be set considering the number of participating nodes 30 and individual participating nodes 30, as well as scenarios that can perform the most optimized parallel processing. And each matching operation is independent of other matching operations and other matching requirements that operate remotely.

The distributed processing unit 130 that generated the matching thread delivers each generated matching thread to the corresponding participating node 30 (S609). Each node that receives the corresponding matching thread from the distributed processing unit 130 assigns the matching operation included in the matching request of the received matching thread to the individual node. One matching request is forwarded to a corresponding one participating node 30. One matching request transmitted to one participating node 30 includes a matching operation corresponding to each of one or more cores included in the participating node 30. [ One or more matching operations included in one matching request are assigned one-to-one to the corresponding cores.

The individual core of the participant node 30 that has been assigned the matching task included in the matching thread by the participating node 30 performs the matching operation based on the matching task included in the matching task (S610). A matching task assigned to an individual core includes one or more matching tasks. The matching task includes a matching algorithm for performing a matching operation of the constituent elements of the candidate ontology and a divided and subdivided ontology subset. Individual cores perform ontology matching operations based on matching tasks. Then, the participant node 30 transmits the matching result generated by the matching operation performed in the individual core to the aggregator 140 (S611). The collecting unit 140 receives a matching result of performing a matching operation through a matching thread from a participating node that has received the matching thread from the distribution processing unit 130. [ Then, the collection unit 140 adds the matching result received from the participating node 30 to generate an ontology mapping (S612). The ontology mapping is a set of individual matching results between the components of two or more candidate ontologies. The accumulator 140 accumulates the matching result calculated by the calculation of the matching operation in the individual core of the participating node 30 and applies a bridge pattern to the accumulated matching result to obtain a formal expression of the ontology mapping (Formal Representation). The bridge pattern is a type of mapping file, which is a custom defined format for mapping files according to individual users. That is, the collection unit 140 can provide the ontology mapping according to the individual format requested by the individual user through the bridge pattern. Then, the collection unit 140 stores the used bridge pattern for another user.

7 is a flowchart illustrating a method of matching a large scale ontology according to an embodiment of the present invention.

Referring to FIG. 7, in step S701, a large-scale ontology matching method according to an exemplary embodiment of the present invention classifies candidate ontologies including a received source ontology and a target ontology into one or more subsets. An ontology can generally consist of five subsets. The subset that makes up the ontology can be broadly divided into name, label, relation, axiom and property. However, the ontology can not be divided into the five subsets described above, but can be classified into one or more subsets regardless of the number depending on the type and characteristics of the received candidate ontology.

If the candidate ontology is classified into one or more subsets, the number of participating nodes and the number of individual cores included in each participating node are checked (S702). The large scale ontology matching apparatus first identifies the number of participating nodes and the number of individual cores included in each attached node, and grasps available computing resources. Then, the large scale ontology matching apparatus sets the optimal number of dispersions considering the number of participating nodes and individual cores determined and the distributed algorithm (S703).

When the number of participating nodes and individual cores is applied to the decentralization algorithm to set the optimal number of dispersions, the ontology subset is divided considering the set optimal number of dispersions, and the matching algorithm of the matching library is applied to the divided ontology subset, And creates a thread (S704). The generated matching thread includes a matching request, a matching operation, and a matching task. The matching algorithm stored in the matching library is an algorithm for computing the matching between the individual components constituting the candidate ontology to generate the ontology mapping between the candidate ontologies. The large scale ontology matching device first divides the ontology subset into one or more matching requests corresponding to each participating node, considering the number of participating nodes. Then, the divided matching request is divided into a matching operation according to the number of cores of the participating node to which the corresponding matching request is assigned, and is included in the matching request.

When the matching thread is generated, each generated matching thread is transmitted to the corresponding participating node (S705). The individual cores of the participating nodes that have been assigned the matching tasks included in the matching thread perform matching operations based on the matching tasks included in the matching operation.

When the matching operation based on the matching thread is performed in the individual core of the participating node, the matching result by the matching operation is collected and summed up (S706). The large scale ontology matching apparatus collects matching results generated by the matching operation performed on the individual cores of the participating nodes, and adds the collected matching results to generate an ontology mapping (S707). The ontology mapping is a set of individual matching results between the components of two or more candidate ontologies. The large scale ontology matching apparatus accumulates the matching result calculated by the operation of the matching operation in the individual cores of the participating nodes and generates a formal expression of the ontology mapping by applying the bridge pattern to the accumulated matching result.

8 is a flowchart illustrating a method of omitting a pre-processing of a large-scale ontology matching method according to an embodiment of the present invention.

Referring to FIG. 8, in order to omit the preprocessing process in the large-scale ontology matching method according to an embodiment of the present invention, the received candidate ontology is classified into a subset (S801). An ontology can generally consist of five subsets. The subset that makes up the ontology can be broadly divided into name, label, relation, axiom and property.

If the ontology subset is classified as an ontology subset, the sorted ontology subset is serialized (S802). An ontology subset divided into two or more subsets can be serialized in a binary form to efficiently use the storage space and increase the efficiency of data transmission and processing. Then, the serialized ontology subset is stored (S803). The storage space can be efficiently used by serializing and storing the ontology subset, and the data processing efficiency can be improved.

Thereafter, if a candidate ontology identical to the previously received candidate ontology is received, the serialized ontology subset stored in the serialized form is de-serialized into the original ontology subset state again (S804). Through this process, unnecessary re-processing of the preprocessing process can be reduced or avoided by deserializing the serialized stored ontology subset. That is, when a candidate ontology identical to the previously stored ontology subset is received, the serialized ontology subset is directly deserialized and restored (reconstructed) without the preprocessing process, The computational resources consumed in the process can be reduced.

The reconstructed ontology subset is deserialized through deserialization and is distributed to individual cores of the participating nodes (S805). First, the number of participating nodes and the number of individual cores in each participating node are checked. The large scale ontology matching apparatus sets the optimal number of dispersions considering the number of identified participating nodes and individual cores and the distribution algorithm. When the number of participating nodes and individual cores is applied to the decentralization algorithm to set the optimal number of dispersions, the ontology subset is divided considering the set optimal number of dispersions, and the matching algorithm of the matching library is applied to the divided ontology subset, Create a thread. The generated matching thread includes a matching request, a matching operation, and a matching task. Once the matching thread is created, it forwards each generated matching thread to the corresponding participating node. The individual cores of the participating nodes that have been assigned the matching tasks included in the matching thread perform matching operations based on the matching tasks included in the matching operation. When the matching operation based on the matching thread is performed in the individual cores of the participating nodes, the matching result by the matching operation is collected and added. The large scale ontology matching device collects the matching results generated by the matching operation performed on the individual cores of the participating nodes, and adds the collected matching results to generate an ontology mapping. Step S805 may be performed in the same manner as steps S703 to S707 of FIG.

The present invention including the above-described contents can be written in a computer program. And the code and the code segment constituting the program can be easily deduced by a computer programmer of the field. In addition, the created program can be stored in a computer-readable recording medium or an information storage medium, and can be read and executed by a computer to implement the method of the present invention. And the recording medium includes all types of recording media readable by a computer.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, It is possible.

100: Ontology matching device
110:
111: ontology model part
112: serialization processing unit
113: Deserialization processor
120: an ontology storage unit
130:
140:

Claims (19)

A preprocessor for generating an ontology subset by classifying the received candidate ontologies into one or more ontology subset;
A distribution processor for dividing the generated ontology subset by applying a distribution algorithm, generating a matching thread by applying a matching algorithm to the divided ontology subset, and delivering the generated matching thread to individual cores of participating nodes; And
A collecting unit for collecting and summing the matching results generated by performing a matching operation based on the matching thread in the individual core to generate an ontology mapping;
Scale ontology matching apparatus. The method according to claim 1,
The pre-
And generates a serialized ontology subset by serializing the generated ontology subset in a binary form. 3. The method of claim 2,
Further comprising an ontology storage unit that stores the serialized ontology subset and provides a pre-stored serialized ontology subset to the preprocessor upon receipt of a candidate ontology identical to the received candidate ontology. The method of claim 3,
The pre-
Wherein the ontology subset is generated by reconstructing the serialized ontology subset received from the ontology storage unit through de-serialization. The method according to claim 1,
The matching thread includes at least one matching request corresponding to each of the participating nodes on a one-to-one basis, one or more matching jobs corresponding one-to-one to individual cores included in the corresponding participating node on a one- And at least one matching task for performing matching operations in the individual cores. 6. The method of claim 5,
The relationship of the matching request, the matching operation and the matching task is

Lt; / RTI >
Wherein MR _i denotes a matching request assigned to each of the participating nodes, MR denotes a set matching request of matching requests MR _i assigned to the participating node, MJ _j denotes a matching request assigned to individual cores MT _k is assigned to an individual core to perform a matching operation. 6. The method of claim 5,
Wherein the matching operation is independent of other matching operations and matching requests operating at other participating nodes. The method according to claim 1,
The dispersion processing unit may include:
When the candidate ontology is received, the number of participating nodes that can be used for matching and the number of individual cores included in the participant node are checked, and the number of distributed nodes is set by applying the number of participating nodes and the individual cores to the decentralization algorithm. And a matching thread including a matching request, a matching operation, and a matching task is generated by dividing the ontology subset in consideration of the set number of dispersions and applying a matching algorithm. The method according to claim 1,
In a distributed environment including two or more participating nodes, a process of collecting and aggregating the matching result of the aggregator

,

And

Lt; / RTI >

(The bridge ontology of the matching task) of the matching task,

Is an intermediate bridge ontology of the matching task that collects all the results of all single matching tasks in all matching tasks,

In all matching operations running on a single participating node, all central bridge ontologies

Lt; / RTI >

Is a central bridge ontology from nodes created at all nodes,

Is converted into an ontology mapping result that is an output file as a bridge ontology that is finally generated. 10. The method of claim 9,
The computed matching result is summed up with one ontology object called a bridge ontology. The aggregating unit sums the bridge ontologies, which are the matching results calculated by all participating nodes participating in the distributed environment, to obtain a physical mapping file And outputs the converted result as an ontology mapping result. In an ontology matching method using a large scale ontology matching apparatus,
Generating an ontology subset by classifying the received candidate ontologies into one or more ontology subset;
Dividing the generated ontology subset using a distribution algorithm, and applying a matching algorithm to the divided ontology subset to generate a matching thread;
Delivering the generated matching thread to an individual core of a participating node;
Generating an ontology mapping by collecting and summing the matching results generated by performing matching operations on the individual cores based on the matching threads;
Dimensional ontology matching method. 12. The method of claim 11,
The step of classifying the received candidate ontology into one or more ontology subset,
Wherein the serialized ontology subset is generated by serializing the generated ontology subset in a binary form. 13. The method of claim 12,
Storing the serialized ontology subset and receiving a candidate ontology identical to the received candidate ontology, reconstructs the pre-stored serialized ontology subset through de-serialization to generate an ontology subset. Scale ontology matching method. 12. The method of claim 11,
The matching thread includes at least one matching request corresponding to each of the participating nodes on a one-to-one basis, one or more matching jobs corresponding one-to-one to individual cores included in the corresponding participating node on a one- And one or more matching tasks for performing matching operations in the individual cores. 15. The method of claim 14,
The relationship of the matching request, the matching operation and the matching task is

Lt; / RTI >
Wherein MR _i denotes a matching request assigned to each of the participating nodes, MR denotes a set matching request of matching requests MR _i assigned to the participating node, MJ _j denotes a matching request assigned to individual cores Wherein MT _k represents a matching task assigned to an individual core to perform a matching operation. 15. The method of claim 14,
Wherein the matching operation is independent of other matching operations and matching requests operating at other participating nodes. ≪ Desc / Clms Page number 19 > 12. The method of claim 11,
Wherein the generating the matching thread comprises:
Confirming the number of participating nodes in the distributed environment and the number of individual cores included in the participating node;
Applying the number of participating nodes and the individual cores to a variance algorithm to set a number of variances; And
Dividing the ontology subset in consideration of the set number of dispersions and applying a matching algorithm to generate a matching thread including a matching request, a matching operation, and a matching task;
Dimensional ontology matching method. 12. The method of claim 11,
The step of collecting and summing the generated matching results to generate an ontology mapping,

,

And

Lt; / RTI >

(The bridge ontology of the matching task) of the matching task,

Is an intermediate bridge ontology of the matching task that collects all the results of all single matching tasks in all matching tasks,

In all matching operations running on a single participating node, all central bridge ontologies

Lt; / RTI >

Is a central bridge ontology from nodes created at all nodes,

Is converted into an ontology mapping result that is an output file as a bridge ontology that is finally generated. 19. The method of claim 18,
The step of generating an ontology mapping by collecting and summing the matching results generated by performing the matching operation based on the matching thread in the individual core may include summing up the bridge ontologies that are the matching results calculated by all the participating nodes, And outputs the converted ontology mapping result to an ontology mapping result which is a physical mapping file.

Images