WO2014092209A1

WO2014092209A1 - Semantic cloud-based semantic annotation method and apparatus

Info

Publication number: WO2014092209A1
Application number: PCT/KR2012/010670
Authority: WO
Inventors: 고인영; 고한규
Original assignee: 한국과학기술원
Priority date: 2012-12-10
Filing date: 2012-12-10
Publication date: 2014-06-19

Abstract

A semantic cloud-based semantic annotation method and apparatus are provided. The semantic annotation apparatus generates one or more RDF nodes on the basis of linked data. The semantic annotation apparatus generates one or more recommended semantic clouds on the basis of the RDF nodes. One semantic cloud is selected from among the semantic clouds recommended by a user. An annotation is added to content on the basis of the selected semantic cloud, according to the semantic cloud that has been selected by the user.

Description

Method and apparatus for semantic annotation based on semantic cloud

The present invention relates to a semantic annotation method and apparatus.

A method and apparatus for performing semantic annotation on content based on a semantic cloud generated from linked data is disclosed.

As users become at the center of content creation and dissemination in the web environment, users play a more important role in metadata generation.

There are many ways to add metadata to web content. The most popular of these methods is annotation. Annotation is a way for users to add tags to web content according to their ideas. Here, the tag is metadata generated by the user.

Because tags can improve content search results, many Web sites encourage users to add tags. However, since tags are only considered plain text, attempts to improve content search yield only limited results.

In particular, the approach of adding tags to web content encounters the problem of semantic ambiguity of tags. For example, it is unclear whether "Apple" means fruit "Apple" or company "Apple".

Semantic annotations exploit ontology to resolve anomalies in searches such as the company "Apple" and fruit "Apple". Thus, semantic web research organizations have used semantic annotations of content as a means to overcome the above limitations.

However, previous efforts to semantic annotation of web content have failed to satisfy the requirements of scalability and usability. Most existing semantic annotation tools use terms in the ontology generated by domain experts. However, these ontologies do not provide enough options to cover various kinds of semantics. In other words, only domain specific terms are available. In addition, domain-specific terms do not necessarily reflect newly generated knowledge in an up-to-date manner. Finding the optimal semantic term from the ontology requires several interactions between the user and the annotation tool.

That is, a major drawback of the conventional semantic annotation method is that the conventional semantic annotation method is not intuitive enough to easily resolve semantic ambiguities while the user associates semantic meaning to a given keyword.

Korean Laid-Open Patent Publication No. 10-2009-0078986 (published May 18, 2011) discloses a semantic annotation apparatus and method for annotating a document using a semantic information knowledge base structured with ontology's semantic information. .

One embodiment may provide an apparatus and method for providing semantic annotation for metadata generated by a user.

One embodiment may provide an apparatus and method for solving the problem of semantic ambiguity of tags in plain text.

In one aspect, generating a query based on a keyword, obtaining one or more Resource Description Framework (RDF) nodes associated with the query from linked data, the one or more RDF Generating one or more recommended semantic clouds based on nodes, determining a semantic cloud corresponding to an annotation to be added among the one or more recommended semantic clouds, and annotating the content based on the determined semantic cloud. Adding to perform semantic tagging of the content.

The semantic cloud corresponding to the annotation to be added among the one or more recommended semantic clouds may be determined by selecting an optimal term by the user.

Generating the one or more recommended semantic clouds includes retrieving the location of one or more spotting points within the linked data, selectively visiting neighboring nodes connected to the spotting point for each of the one or more spotting points. Thereby traversing a graph and generating the one or more recommended semantic clouds by clustering the spotting point and the selectively visited connected neighboring nodes for each of the one or more spotting points.

The one or more spotting points may be representative RDF nodes covering concepts related to the keyword.

The connected neighboring nodes may be selectively visited via user context and interlinked relationship terms.

The query may be a query to the linked data, the query may be generated based on a sparkle.

According to another aspect, a query manager for generating a query based on a keyword and obtaining one or more Resource Description Framework (RDF) nodes associated with the query from linked data, the one or more A semantic cloud generator that generates one or more recommended semantic clouds based on RDF nodes and a semantic cloud corresponding to an annotation to be added among the one or more recommended semantic clouds are determined, and the content is based on the determined semantic cloud. A semantic annotation device is provided that includes an annotation adder that performs semantic tagging of the content by adding the annotation.

According to another aspect, providing a first content to a user, receiving a keyword associated with the first content from the user, generating a query based on the keyword, Linked Data Obtaining one or more Resource Description Framework (RDF) nodes associated with the query from, generating one or more recommended semantic clouds based on the one or more RDF nodes, the one or more recommended semantics Determining a semantic cloud corresponding to the annotation to be added among the clouds and performing semantic tagging on the second content by adding the annotation to the second content based on the determined semantic cloud; Semantic Annotation Method It is provided.

According to another aspect, a content providing unit for providing a first content to the user, a transceiver for receiving a keyword related to the first content from the user, generates a query based on the keyword, linked data (Linked A query manager that obtains one or more Resource Description Framework (RDF) nodes associated with the query from Data, a semantic cloud generator that generates one or more recommended semantic clouds based on the one or more RDF nodes; Add an annotation to perform semantic tagging on the second content by determining a semantic cloud corresponding to the annotation to be added among the one or more recommended semantic clouds and adding the annotation to the second content based on the determined semantic cloud Containing wealth, Smart devices are provided.

An apparatus and method are provided for providing semantic annotation for metadata generated by a user.

An apparatus and method are provided for solving the problem of semantic ambiguity of tags in plain text.

1 illustrates a semantic annotation device according to one embodiment.

2 illustrates a semantic annotation method according to one embodiment.

3 illustrates semantic cloud generation from linked data according to an example.

4 illustrates a method of finding spotting points in linked data according to an example.

5 shows demonstration results for each set of selected datasets according to an example.

6 illustrates a result of generating a semantic cloud for the keyword 'apple' according to an example.

7 illustrates a semantic annotation interface in a web-based IPTV environment according to an example.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, the present invention is not limited or limited by the embodiments. Like reference numerals in the drawings denote like elements.

In the following description, linked data may refer to data or a service provided by linking open data (LOD). The LOD may be the World Wide Web Consortium (W3C) Semantic Web Education and Outreach (SWEO) Community Project.

In the following, a semantic-cloud-based annotation method is proposed which can easily add semantic annotations to web content. The proposed method may provide semantic clouds to the user as a primary interface for semantic annotation. Users can select the most suitable concept among candidate semantic clouds. One of the most important elements of the semantic-cloud based annotation method is the creation of efficient semantic clouds that enable the authoritative recognition of candidate concepts to be annotated without semantic ambiguity.

The semantic cloud generation method described below may first determine the location of essential points to begin retrieving relevant concepts within Linked Data to generate semantic clouds. . Next, the semantic cloud generation method can iteratively analyze potential merges of different semantic data. The user can easily resolve the semantic ambiguity and select the most appropriate semantic cloud from the set of candidates.

A method for creating a semantic cloud, comprising: 1) selective querying of multiple datasets in linked data, and 2) providing a context sensitive traversal to such data to handle large amounts of linked data. It can be focused on reducing the complexity of. Also described below is the quality of the semantic clouds generated by the proposed method through the case of comparing the experimental results.

도 1은 일 실시예에 따른 시맨틱 어노테이션 장치를 설명한다.1 illustrates a semantic annotation device according to one embodiment.

The semantic annotation device 100 may include a transceiver 110, a query manager 120, a semantic cloud generator 130, a semantic cloud searcher 140, and an annotation adder 150.

The query manager 120 may be an LOD query manager.

The semantic annotation device 100 may be a smart device. When the semantic annotation device 100 is a smart device, the semantic annotation device 100 may further include a content providing unit 160. Here, a smart device may refer to a series of devices that provide interaction with a user in addition to playing content, such as a smart phone and a smart television.

도 2는 일 실시예에 따른 시맨틱 어노테이션 방법을 설명한다.2 illustrates a semantic annotation method according to one embodiment.

In FIG. 1, semantic clouds can be generated from linked data. Semantic clouds can be used as the primary interface for semantic annotation.

In operation 200, the content provider 160 provides multimedia content to the user.

In operation 210, a user may input keywords related to the multimedia content while watching the multimedia content. The transceiver 110 may receive a keyword input by a user. Here, the keyword may be text.

In operation 220, the query manager 120 may generate a query based on the keyword.

In step 230, query manager 120 may obtain one or more RDF nodes associated with the query from the linked data based on the query.

For example, the transceiver 110 may transmit the generated query as linked data. The transceiver 110 may receive RDF and XML data from the linked data. The query manager 120 may generate RDF nodes based on the RDF and XML data.

In operation 240, the semantic cloud generator 130 may generate one or more recommended semantic clouds based on one or more RDF nodes.

At 250, a semantic cloud of one of the one or more recommended semantic clouds may be determined by the user. The user can determine the semantic cloud corresponding to the annotation to be added among the one or more clouds by selecting the optimal term. The semantic cloud searcher 140 may search for and determine an optimal semantic cloud corresponding to an annotation to be added among one or more recommended semantic clouds based on a user's manipulation, for example, left and right slides or a semantic cloud regeneration request. The left and right slides enable horizontal navigation of the multiple semantic clouds created in step 240 in addition to the semantic clouds currently shown on the screen and the regeneration request creates a more detailed semantic cloud for that semantic cloud. Is requested in step 240 to enable vertical search for each semantic cloud.

In operation 260, the annotation adding unit 150 may perform semantic tagging on the multimedia content or the web content by selecting and adding an optimal term to the multimedia content or the web content based on the determined semantic cloud. have.

In the semantic annotation method described above, three technical issues to be addressed can be pointed out.

The first issue concerns access to large amounts of semantic web data and processing of the large amounts of semantic web data.

The second issue relates to the creation of related semantic clouds.

A third issue relates to the provision of an efficient user interface that enables intuitive interactions.

Among these issues, the focus can be on creating semantic clouds from large scale web data. Requirements for well-organized semantic clouds can be proposed as follows to achieve the goal of generating a semantic cloud from large scale web data.

The first requirement is that the number of clouds is small. For example, the number of semantic clouds recommended at one time may be limited to a maximum of four.

The second requirement is the balance of content within the semantic cloud. For example, semantically related terms are preferably included within the same semantic cloud.

The third requirement is that there is no ambiguity between the clouds. For example, it is desirable that the semantic ambiguity between generated semantic clouds be minimized.

A method for creating a semantic cloud that can meet these requirements is described in detail below.

도 3은 일 예에 따른 링크된 데이터로부터의 시맨틱 클라우드 생성을 설명한다.3 illustrates semantic cloud generation from linked data according to an example.

According to statistics on the linked data, the linked data includes more than 31 billion RDF triples from 295 different datasets that are domain-independent. Thus, the linked data can be a large, heterogeneous semantic web data store. In order to generate semantic clouds from the linked data, the semantic cloud generation method can be incremental and repeated.

Step 240 described above may include steps 310 to 330.

In operation 310, the semantic cloud generator 130 may search for locations of one or more spotting points for clustering within the linked data.

One or more spotting points may be representative RDF nodes that cover concepts related to the keyword. That is, the semantic cloud generator 130 may search for representative RDF nodes covering concepts related to the input keyword.

After retrieving the location of the spotting points, in step 320, the semantic cloud generator 130 selectively visits neighboring nodes connected to the spotting point for each of the one or more spotting points, thereby Can be traversed.

The semantic cloud generator 130 may selectively visit neighboring nodes connected through user context and interlinked relation terms.

Selective visits to neighboring nodes can reduce the complexity of handling large amounts of linked data. In addition, selectively visiting neighboring nodes is the quality of the semantic coherence of the semantic coherence generated by filtering out less relevant relationships and corresponding nodes. ) Can be guaranteed.

In order to determine whether to include visiting semantic nodes in the same tag cloud, it is necessary to measure the semantic similarity between the spotting point and the visiting node. Basically, the number of concepts that overlap the spotting pointing and landing nodes may be the basis for measuring the semantic similarity between the spotting pointing and landing nodes. Additionally, determining the maximum number of hops to traverse and the threshold value for semantic similarity can be discussed.

In step 330, the semantic cloud generator 130 may generate the one or more recommended semantic clouds by clustering the spotting point and the selectively visited connected neighboring nodes for each of one or more spotting points. .

도 4는 일 예에 따른 링크된 데이터 내에서 스포팅 포인트들을 찾는 방법을 설명한다.4 illustrates a method of finding spotting points in linked data according to an example.

Retrieving 310 the location of one or more spotting points in the linked data may be the most important step in creating high quality semantic clouds that can meet the requirements described above. This is because in generating high quality semantic clouds, it may be more efficient to traverse and retrieve RDF nodes associated with each of the RDF nodes starting from more important and densely connected RDF nodes.

Locating the spotting points can be viewed as starting from querying the linked data 230 to obtain relevant RDF nodes.

In step 220, there are two ways for the query manager 120 to create a query with the linked data. The first approach is query generation via Sparkle, and the second approach is Semantic Web, such as the Semantic Web Search Engine (SWSE), Falcons, and Syndic. ) Query generation through search engines. Here, Sparkle is one of the query languages for RDF.

Semantic web search engines can rank RDF nodes with their own algorithms. As a result of the ranking above, RDF nodes retrieved by the second scheme can be biased without covering all datasets. Thus, at step 220, the query manager 120 may generate a query to the linked data based on the SPARQL, according to the first approach.

Although the SPARQL endpoints for each linked dataset such as DBpedia, Freebase and Geonames must be handled, the latest version of relevant RDF nodes are obtained. Can be.

Querying all datasets in the linked data can take a long time. Thus, at step 230, query manager 120 may issue SPARQL queries for the dataset selected for performance.

ckan.net is a web site that collects and maintains metadata of datasets such as the number of triples, the number of in-links, the number of out-links, topics, SPARQL endpoints and license information.

The transceiver 110 may receive metadata of each data set. In addition, the query manager 120 may construct a graph having data sets of linked data by using the metadata. In addition, the semantic cloud generation unit 130 may analyze the graph with respect to the degree (or hub) and thebetweenness. The above analysis can explain how many different datasets are connected to the dataset and how often the dataset appears on the shortest paths between the datasets.

The semantic cloud generator 130 should be able to efficiently traverse related RDF nodes including all existing semantics through the spotting points. Accordingly, the semantic cloud generator 130 may return improved results by selecting spotting points with different topics, as well as order and interimportance, from the datasets.

The query manager 120 may determine the ranking of each dataset by considering the topic, order, and importance of each dataset. More important datasets may be determined by ranking of the dataset. The query manager 120 may selectively query more important data sets based on the ranking of each data set.

For example, if a query for the keyword 'apple' is generated and passed to multiple of the datasets from the linked data (410), more than 140,000 related RDF nodes may be returned (420). The returned RDF nodes can be bundled into concepts with one independent meaning. For example, dbpedia.org/resource/Apple and rdf.freebase.com/ns/m/0k8z are the same concept as 'Apple Inc.' RDF nodes that represent. To do this, you can use links that represent relationships between RDF nodes that have the same meaning, such as owl: sameAs or skos: exactMatch.

Also, rather than having all the RDF nodes, selecting some representative nodes can reduce the complexity in forming the semantic cloud. The more abstract the nodes, the higher the order of connections can be. Accordingly, the semantic cloud generator 130 may select the most common concepts of RDF nodes as spotting points by comparing relative concept hierarchies of RDF nodes. This choice ensures the semantic unambiguity between the spotting points above. Thus, this choice may support the requirement that there should be no ambiguity between the generated semantic clouds.

Simple Knowledge Organization System 3 (SKOS3) is a common data model for sharing and linking knowledge organization systems. SKOS3 provides useful relationship terms such as skos: broader and skos: narrower. Relationship terms may be exploited to retrieve relative concept layers for comparing RDF nodes and to select the most common RDF node as a spotting point.

In FIG. 4, several RDF nodes such as 'Apple I', 'Apple IIGS' and 'Apple Lisa' have been extracted by the keyword 'apple'. The semantic cloud generation unit 130 parses the concept graph of the RDF nodes by parsing identical semantic links such as owl: sameAs or skos: exactMatch and semantic layer links such as skos: broader and skos: narrower links. It may generate and recognize 430 the most common and important concepts of the above RDF nodes. In the case illustrated in FIG. 4, the most common concept is 'Apple Inc.' and 'Apple Inc.' was selected as the spotting point.

In the following, step 320 is further described.

In step 320, the semantic cloud generator 130 may select relations that link related RDF nodes. Since all the RDF nodes in the linked data are connected as a large graph structure, the semantic cloud generator 130 can retrieve important and related RDF nodes by traversing the graph from spotting points. In the case of Apple Inc., if the semantic node's labels contain 'iPod', 'iPhone', 'MacBook' or 'Steve Jobs', the semantic node above is important even if it does not contain the keyword 'apple' Can be considered a semantic node.

Accordingly, the semantic cloud generator 130 may reduce the complexity of the semantic cloud generation by setting the traversal boundaries. The semantic cloud generation unit 130 may reduce the computational complexity for traversing the large RDF graph by considering only related relationships. To ideally select semantically related relationships, semantic cloud generator 130 may automatically associate relationship terms with user contexts, such as interests and preferences. Here, since automatic association between contextual information and relationship terms may not be currently technically possible, the interests of the user may be interlinked with the relationship terms, and the interests of the user may already be present for each user before the semantic cloud is created. For example, in a case where the user interest is 'movie', relationship terms such as 'actor', 'director', 'rating', 'background music' and 'story' will be the relations to be traversed. Can be.

In addition, the W3C requires linked data publishers, such as Friend Of A Friend 4 (FOAF4), Dublin Core 5, to ensure interoperability between linked data datasets. 5) It is recommended to use well-defined, popular terms such as DC5) and Semantically-Interlinked Online Communities 6 (SIOC6) and SKOS. In step 320, the semantic cloud generator 130 may first traverse the relationships and may take into account the relationships selected by the user in consideration of contexts. This traversal and reference facilitates visiting related nodes while reducing the complexity for clustering these nodes.

In the following, step 330 is further described.

In operation 330, the semantic cloud generator 130 may measure semantic similarity between RDF nodes to determine whether to include visited PDF nodes in the same cloud in the semantic cloud.

Similar to 'term frequency' in the field of information retrieval, the semantic cloud generator 130 calculates the number of query responses from the semantic web search engine to measure the similarity between nodes. Can be used. In Equations 1 and 2 below, l ₁ and l ₂ are labels of RDF nodes. n ( l ) represents the number of query responses for RDF node l . h represents the number of hops to traverse. w is a weight.

Equation 1

Equation 2

TermFreq ( l ₁ , l ₂ ) may indicate the term frequency between l ₁ and l ₂ .

As the number of hops from the spotting point increases, the value of semantic similarity may decrease exponentially. For this reason, SemSim ( l ₁ , l ₂ ) representing the semantic similarity between two RDF nodes l ₁ and l ₂ may be proportional to the inverse of w ^h .

The semantic cloud generator 130 may determine a weight w and a threshold h for clustering to include semantically related concepts towards keywords.

도 5는 일 예에 따른 각 선택된 데이터셋들의 집합에 대한 실혐 결과들을 도시한다.5 shows demonstration results for each set of selected datasets according to an example.

The first four

graphs

510, 520, 530 and 540 show that for each experimental term, the sets of datasets generated based on domain and intercriticality show high coverage rates, respectively. , Triple and type respectively. In the triple coverage of the keyword 'tiger', the domain and interimportance are much higher than in other cases. This is because the domain and inter importance include a dataset with multiple RDF nodes for the above keywords.

In the case of type coverage, it is difficult to recognize significant differences between the generated sets of datasets. This is because each RDF node has multiple types, many of which are redundant.

The final graph 550 is the parsing of the same semantic links (eg owl: sameAs, skos: exactMatch) and semantic layer links (eg skos: broader, skos: narrower, rdf: subClassOf) used to find the spotting points. It shows the percentage of RDF nodes that are reduced. The tag terms used in the experiment were selected from some of the terms used for tagging on the Internet, and the maximum decrease rate was 75.44% (term 'Nature'), and the average was 19.05%. The higher the reduction rate, the higher the probability that RDF nodes that are not duplicated and have a significant meaning for the term are selected as spotting points.

도 6은 일 예에 따른 키워드 'apple'에 대한 시맨틱 클라우드 생성의 결과를 도시한다.6 illustrates a result of generating a semantic cloud for the keyword 'apple' according to an example.

In creating a semantic cloud, there are three ways. In FIG. 6, the

results

610, 620, and 630 of generating the semantic cloud by rdf: type, SKOS parsing, and the proposed method (see FIGS. 1-4), respectively, are shown.

The first method is to cluster RDF nodes according to the rdf; type of RDF nodes. However, the first method may not guarantee high quality semantic cloud generation. For example, 'Apple Inc' with rdf: type 'company' has been separated from groups such as 'Apple I' and 'Apple II GS' with rdf: type 'Personal Computer'. It is clear that there is a semantic relationship between Apple Inc, Apple I and Apple II GS.

The second method is to use SKOS relationships. The second method is useful for understanding the relative concept layer between RDF nodes. However, since the clouds generated by the second method only contain RDF nodes that contain only the keyword 'apple', the result obtained by the second method failed to meet the balance of content in each cloud. In addition, some of the generated clouds have only a single node.

Semantic clouds generated by the third method provide better results than semantic clouds generated by other methods. The semantic clouds generated by the third method also include related RDF nodes such as 'ITunes' and 'Macintosh' via relationship traversal. Related RDF nodes do not contain the keyword 'apple'.

도 7은 일 예에 따른 웹-기반 IPTV 환경에서의 시맨틱 어노테이션 인터페이스를 설명한다.7 illustrates a semantic annotation interface in a web-based IPTV environment according to an example.

The user can add annotations to the multimedia content by selecting semantic options from semantic clouds generated from the linked data. Annotation results can be used to provide semantic search. Semantic search can improve search results for multimedia content.

In FIG. 7, annotation timing 710 to which annotations are added, keyword 720 used to generate semantic clouds, generated semantic cloud 730, selected linked data 740, cloud generation through search 750, cloud generation through input 760, start button 770, and the like are shown as examples of interfaces.

Method according to an embodiment is implemented in the form of program instructions that can be executed by various computer means may be recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the media may be those specially designed and constructed for the purposes of the embodiments, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the claims below but also by the equivalents of the claims.

Applied to the field of performing semantic annotation on multimedia content or web content based on the semantic cloud.

Claims

Generating a query based on the keyword;

Obtaining one or more Resource Description Framework (RDF) nodes associated with the query from Linked Data:

Generating one or more recommended semantic clouds based on the one or more RDF nodes;

Searching for and determining a semantic cloud corresponding to an annotation to be added among the one or more recommended semantic clouds; And

Performing semantic tagging on the content by adding the annotation to the content based on the determined semantic cloud

Including, semantic annotation method.
The method of claim 1,

The semantic annotation method of the one or more recommended semantic clouds corresponding to the annotation to be added is determined by selecting an optimal term by the user.
The method of claim 1,

Generating the one or more recommended semantic clouds,

Retrieving the location of one or more spotting points within the linked data;

Traversing a graph by selectively visiting neighboring nodes connected to the spotting point for each of the one or more spotting points; And

Generating the one or more recommended semantic clouds by clustering the spotting point and the optionally visited connected neighboring nodes for each of the one or more spotting points

Including, semantic annotation method.
The method of claim 2,

The one or more spotting points are representative RDF nodes that cover concepts related to the keyword.
The method of claim 3,

The connected neighboring nodes are selectively visited via user context and interlinked relationship terms.
The method of claim 1,

The query is a query to the linked data, the semantic annotation method being generated based on a sparkle.
A query manager that generates a query based on a keyword and obtains one or more Resource Description Framework (RDF) nodes associated with the query from Linked Data;

A semantic cloud generator for constructing a temporary graph of the one or more RDF nodes, searching for one or more spotting points, and generating one or more recommended semantic clouds based on the one or more spotting points;

A semantic cloud search unit for searching a cloud corresponding to an annotation to be added among the one or more recommended semantic clouds; And

An annotation adder that performs semantic tagging on the content by selecting an optimal term based on the searched semantic cloud and adding the annotation to the content.

It includes, a semantic annotation device.
The method of claim 7, wherein

The semantic annotation device is determined by searching for a semantic cloud corresponding to the annotation to be added among the one or more recommended semantic clouds and selecting an optimal term included in the semantic cloud by the user.
The method of claim 7, wherein

The semantic cloud generator retrieves the location of one or more spotting points in the linked data, traverses the graph by selectively visiting neighboring nodes connected to the spotting point for each of the one or more spotting points, and the one Generating the one or more recommended semantic clouds by clustering the spotting point and the optionally visited connected neighboring nodes for each of the spotting points.
The method of claim 9,

By grouping nodes having the same meaning among the one or more RDF nodes obtained by the one or more query managers (eg, owl: sameAs, skos: exactMatch) and identifying semantic hierarchical relationships between RDF nodes (eg, skos: broader , skos: narrower, rdf: subClassOf) A semantic annotation device that constructs one or more temporary concept graphs and analyzes the importance of nodes in each graph to determine spotting points.
The method of claim 9,

And the semantic cloud generator selectively visits the connected neighboring nodes through a user context and interlinked relationship terms.
The method of claim 7, wherein

The query is a query to the linked data, and the query manager generates the query based on a sparkle.
Providing first content to a user;

Receiving a keyword associated with the first content from the user;

Generating a query based on the keyword;

Obtaining one or more Resource Description Framework (RDF) nodes associated with the query from Linked Data:

Generating one or more recommended semantic clouds based on the one or more RDF nodes;

Searching for a cloud corresponding to an annotation of the one or more recommended semantic clouds;

Determining a term corresponding to an annotation to be added among terms in the one or more discovered semantic clouds; And

Performing semantic tagging on the second content by adding the annotation to the second content based on the determined semantic cloud

Comprising, a semantic annotation method of a smart device.
A content providing unit providing first content to a user;

A transceiver for receiving a keyword related to the first content from the user;

A query manager that generates a query based on the keyword and obtains one or more Resource Description Framework (RDF) nodes associated with the query from Linked Data;

A semantic cloud generator for generating one or more recommended semantic clouds based on the one or more RDF nodes;

A semantic cloud search unit for searching a cloud corresponding to an annotation to be added among the one or more recommended semantic clouds; And

An annotation adder that performs semantic tagging on the second content by selecting an optimal term among the terms included in the searched semantic cloud and adding the annotation to the second content.

Including, a smart device.