KR20180129001A - Method and System for Entity summarization based on multilingual projected entity space - Google Patents

Abstract

The present invention provides a method and a system for generating an entity summary based on a multilingual property projected entity space. According to the present invention, the method for generating an entity summary based on a multilingual property projected entity space comprises: a step of extracting a triple indicating a category in a multilingual knowledge base to integrate information of identical entity units; a step of extracting a three-term relation indicating a category in the multilingual knowledge base to perform entity grouping; a step of finding a main description relation and a main entity-object correlation for each group based on entity groups constructed by a group construction unit, and calculating a weight of a triple of the multilingual knowledge base; a step of repeating analysis for all three-term relations, and aligning a summary in accordance with importance order for all three-term relations based on calculated weights; and a step of minimizing redundancy in accordance with a request of a user for the aligned summary, and retrieving the aligned summary from the highest of the importance order.

Description

Field of the Invention < RTI ID = 0.0 > [0006] < / RTI >

In the present invention, the features of various languages are projected in a single space from a large-capacity knowledge base created in the form of a Resource Description Framework (RDF) triple, followed by grouping of objects based on integrated category tags, And more particularly, to a method and system for generating an entity summary that calculates and priorities of triples.

Object summarization technology is a technology that selects major information of each entity in a large-capacity knowledge base described by an entity and reconstructs it according to the length of a proper summary. It is a technique of searching large data, extracting information, (Question and Answer) and other natural language processing applications.

As the number of information that is organically connected to one entity on the web becomes wider as the study of the linked data environment in which the openness and connectivity of the data existing in the semantic web and the web becomes active becomes active, It is now widely studied as a necessary technique to solve the problem of searching only important information quickly and accurately in a large scale knowledge base. Conventionally, an entity summary generation system has been developed after executing the division of the information based knowledge base acquired in a single language environment. However, there are limitations in modeling the boundary of the knowledge base by expressing the inherent characteristics of the entity, (WordNet). However, since it is impossible to expand the objects that are not registered in the external dictionary resource, the existing object summary generation system is limited in its method and application.

Entity Summarization was first defined in the paper "RELIN: Relatedness and Informativeness-Based Centrality for Entity Summarization" presented by Gong Cheng, Thanh Tran and Yuzhong Qu in 2011 at the International Semantic Web Conference (ISWC) It is a technology that extracts small-scale data of individual units in order to quickly and conveniently access information of a specific entity among RDF triple data existing in a large-scale data space which is continuously growing through the use of open data.

The entity summarization technique can be applied as an additional service to the search system to provide quick information about the entity by providing the essential information necessary to describe the entity among the information gathered from various data sources about the entity used in the search query term .

Currently, Google provides a service similar to Knowledge Graph, which is a non-automated technology.

From the integrated information of various information sources being disclosed by big data related companies and government departments, it is possible to provide basic basic information about various objects and to provide information search of objects.

In addition, when the smartphone is applied as a service for providing visualization of knowledge based on a smart phone and some of the large amount of data needs to be exposed to a small screen, essential information can be provided through exposure and application of essential information.

In the future prospect of enterpriseization, it is possible to commercialize the area that interprets data and knowledge base which can be accessed by open data. Related researches have been actively conducted by corporations related to knowledge provision and search systems such as the above-mentioned knowledge graph of Google. In addition, it is possible to utilize e-Learning curriculum for various thematic concepts by extracting and constructing content elements that are essential for the individual.

However, in the case of the prior art, by generating a summary based on the relative importance between the entity appearing in the entity description and the attribute value (another entity), information that is not essential for describing the given entity is included in the summary There are limitations that can be included. In addition, in the case of the prior art, there is a weak point that the performance of the summary may be degraded when a plurality of important information describing the characteristics of the entity are included in one page, in addition to the limitation of utilizing external resources. We also extended the qualities of long word sequences that can be inferred from objects using WordNet, an external dictionary resource. However, this is not applicable to object names that are not registered in advance, or to language data in which dictionaries are not defined. There are limitations to this.

The object of the present invention is to integrate a knowledge base and a multilingual knowledge base describing an object by taking the lengths of objects and a summary to be generated as an input, The present invention is to provide a method and system for generating a given entity summary by sorting triples of entity description queries by identifying important information of a cluster unit, and then outputting the sorted results by a user in a priority order.

According to an aspect of the present invention, there is provided a method for generating a multi-lingual feature-based object space based entity summary by a multi-language knowledge base, the method comprising the steps of: extracting triples indicating classification schemes in a multi- In this paper, we propose a method of extracting ternary relations from a base and constructing an object clustering based on the clustering based on the clusters. , Repeating the analysis of all ternary relations, sorting the summary according to the order of importance of all ternary relations based on the calculated weights, and arranging the summary according to the user's request on the sorted summary Minimizing and taking precedence of the sorted summaries in order of importance. The.

Extracting a triple marking a classification system from a multilingual knowledge base, and integrating information of the same entity unit, deriving qualities commonly used in a plurality of languages by linking triples created in a plurality of languages to the same entity, Extracts the qualities independently used in each of the plurality of languages, and integrates the classification scheme characteristics of the object for the corresponding object generated in the plurality of language communities.

The step of extracting the ternary relations from the multilingual knowledge base and constructing the entity clustering is to derive the qualities for clustering the entities from the triple describing the classification system existing in the multilingual knowledge base, And clusters the corresponding objects of the group.

The step of finding the main narrative relation and the major object-object correlation and calculating the weight of the triple of the multilingual knowledge base based on the individual cluster constituted in the clustering component, Based on the main attribute type in the entity cluster, which is a combination of points indicating the number of points representing the entity-attribute value co-occurrence.

The step of repeating the analysis of all ternary relations and sorting the summaries according to the order of importance of all ternary relations based on the calculated weights derives the main attribute types of the triples according to the individual clusters, And summarizes the summary according to the order of importance by using the combination of the main attribute type and the entity-attribute value correlation.

The step of minimizing the redundancy according to the request of the user for the sorted summary and fetching from the order of priority among the sorted summaries minimizes the duplication of the attribute type and attribute value used in the object description statement, Generate a summary as long as the length.

According to another aspect of the present invention, there is provided a multilingual feature-based object space based entity summary generation system proposed by the present invention, which extracts triples indicating classification schemes in a multilingual knowledge base, Project module, and multilingual knowledge base, extracts the ternary relations that mark the classification system, finds the main narrative relation of each community based on the object community that is composed in the object clustering module and the clustering component that constitute the object clustering, An entity descriptor ranking module and a descriptive relation analysis unit for calculating the weights of the triples and calculating the weight of the triples of the multilingual knowledge base based on the individual entity group constituted by the grouping unit, The analysis of all ternary relations is repeated through the Object-Object Analysis Division, Based on the weights, we summarize the summary according to the order of importance of all ternary relations, minimize the duplication according to the user 's request for the sorted summary, and generate the summary summary of the sorted summary from the order of priority. .

The object clustering module derives the qualities commonly used in a plurality of languages by associating triples created in a plurality of languages with respect to the same object, derives the qualities independently used in each of the plurality of languages, And integrates the classification scheme characteristics of the object for the corresponding object.

The object clustering module derives the qualities for clustering objects from the triple describing the classification scheme existing in the multilingual knowledge base, and clusters the objects of similar derived qualities.

The Entity Knowledge Ranking module is based on the main attribute types in the Entity Clusters, which consist of a combination of the frequency of attribution types used in the Entity Clusters and the scores indicating the backward house frequencies, and the Entity-Attribute Value Co- .

The entity summary generation module derives the main attribute types of the triple according to the object group, derives the entity-attribute value correlation, and sorts the summary according to the order of importance using the combination of the main attribute type and the entity-attribute value correlation .

The object summary generation module minimizes the duplication of attribute types and attribute values used in the object description statement, and generates a summary of the length required by the user.

According to the embodiments of the present invention, it is possible to reproduce as efficiently as possible an object-by-object information on a large-capacity knowledge base by clustering the objects through multi-lingual characteristic projection, and reproducing them as close as possible to a summary method of experts, It can be useful for providing search and quick query processing.

FIG. 1 is a flowchart illustrating a method for generating a multi-language feature-based object space based entity summary according to an exemplary embodiment of the present invention.
Figure 2 illustrates a single spatialization of category tags found in three different languages for an entity in accordance with an embodiment of the present invention.
FIG. 3 is a block diagram of a multi-language projected object space based entity summary generation system according to an exemplary embodiment of the present invention.
4 is a diagram illustrating a category tag for an entity existing in a Korean community according to an embodiment of the present invention.
5 is a diagram illustrating a category tag for an entity existing in an English community according to an embodiment of the present invention.
6 is a diagram illustrating vectorization of a root found from a classification system word according to an embodiment of the present invention.
FIG. 7 is a diagram illustrating the addition of weights to the vectorization of a root found from a classification system word according to an embodiment of the present invention.
8 is a diagram illustrating a triple set comparison of two entities of a cluster according to an embodiment of the present invention.
9 is a diagram illustrating a duplicate attribute type for one entity according to an embodiment of the present invention.
FIG. 10 is a diagram illustrating a final summary comparison according to whether a duplicate attribute type is permitted according to an embodiment of the present invention.
11 is a diagram illustrating a final summary comparison according to whether or not a duplicate attribute value is permitted according to an embodiment of the present invention.

The multi-language feature projection system based on the projected classification system according to the present invention utilizes the characteristics of the knowledge base, which is publicly provided in more than 120 languages, And an object summary is generated by applying a triple importance calculation method according to a knowledge base boundary of the object cluster unit calculated in the step. According to the method of generating entity summaries proposed in the present invention, the clustering in a space modeled by only one language is performed, the clustering of the projected multilingual features is performed, and the triple It is possible to generate a summary of excellent performance that can select required triples to be included in the entity summary.

In the detailed description of the present invention, the term " resource " refers to all objects identifiable by a URI, regardless of its form in the RDF data model, and one information resource may have multiple attribute types and attribute values have.

In the detailed description of the present invention, the term 'entity' refers to a continuous string of text that can have a name among information resources, for example, a name, an institution name, a place name, and the like.

In the detailed description of the present invention, the term 'property type' means that an attribute of a resource is expressed by a proper name such as 'author', 'signature' and the like.

In the detailed description of the present invention, the term " value " is a value corresponding to an attribute type and may be described in detail in a natural language such as a character string or a number. Attribute.

In the detailed description of the present invention, the term " triple " means including information resources, attribute types, and attribute values.

In the detailed description of the present invention, the term 'entity description' refers to a set of triples that refer to the same entity as an information resource. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a flowchart illustrating a method for generating a multi-language feature-based object space based entity summary according to an exemplary embodiment of the present invention.

The method of generating a projected object space based object summary according to the present invention includes a step 110 of extracting a triple marking a classification system from a multilingual knowledge base and integrating information of the same entity unit 110, (120) constructing an object clustering by extracting a tagging ternary relation, (120) searching for the main description relation and major object-object correlation according to the clusters based on the individual clusters constructed in the clustering component, (130), repeating the analysis of all ternary relationships, and sorting the summary according to the order of importance of all ternary relations based on the calculated weights (140) (Step 150) of minimizing redundancy according to the priorities of the order of importance among ordered summaries.

In step 110, triples marking the classification scheme are extracted from the multilingual knowledge base, and information of the same entity unit is integrated. First, we construct a unit space from a plurality of knowledge bases written in various languages into one space. The triples formed in a plurality of languages are linked to the same entity to derive the qualities commonly used in a plurality of languages and the qualities used independently in each of the plurality of languages are derived. And integrates the classification scheme characteristics of the object for the corresponding object generated in a plurality of language communities.

In step 120, a ternary relation marking the classification system is extracted from the multilingual knowledge base to form an object clustering. Clustering refers to finding commonalities among individuals and dividing them into one set. From the triple describing the classification system existing in the multilingual knowledge base, the qualities for clustering individuals are derived and the corresponding entities of similar qualities are grouped together. In the process of clustering individuals, it is possible to acquire the common points of individuals from triples using specific property types that represent the classification systematic characteristics of individuals from the knowledge base.

For example, in the qualification selection process of object clustering, the boundaries of words consisting of noun phrases excluding common prefixes can be distinguished, and the stemming process for finding the root of each word can be performed.

Using a specific algorithm among the division methods widely used in the field of computation, the given object is divided into several clusters based on the previously selected qualities. The process of dividing the cluster is done by minimizing the sum of squares of the distance between the center of each cluster and the individual in the cluster as a cost function. In this process, the degree of similarity between individuals in the same cluster increases, and the similarity with individuals in other clusters decreases. This process can be replaced by using the existing ontology classification system. In the present invention, a k-means algorithm is used among the division methods in the field of computation, but the present invention is not limited thereto.

In step 130, the major descriptive relationships and major object-object correlations of the clusters are searched based on the individual clusters constructed in the clustering component, and the weights of the triples of the multilingual knowledge base are calculated. In other words, the most important attribute type is derived for each of the generated community through the step 120. This process is defined as a statistical number that indicates how important an attribute type is in a particular cluster when there are several communities. The weight calculation is based on the main attribute type in the object cluster, which is a combination of the attribute type frequency used in the object cluster and the score indicating the reverse population frequency, and is based on the object-attribute value co-occurrence by the object cluster unit .

The weight of an attribute type is defined by a combination of the following two qualities: the attribute frequency and the inverse group frequency of the attribute type word in the cluster.

The frequency of attribute type words in a cluster uses the total frequency of attribute type words appearing in the cluster. The inverse group frequency indicates how common an attribute type word appears in the whole population. The value obtained by dividing the total number of populations by the number of populations including the attribute type word is used.

The higher the attribution type word frequency in a particular cluster, and the smaller the population including the attribute type word among all the clusters, the higher the weight value of the narrative relation. Using this, it is possible to obtain the effect of filtering the attribute type words that are common in all the clusters, so that it is possible to grasp the meaningful property types in the community.

In step 140, the analysis of all ternary relationships is repeated and the summary is sorted in order of importance for all ternary relationships based on the calculated weights. We derive the main attribute types of the triple according to the entity clusters and derive the entity - property value correlations and sort the summary according to the order of importance by using the combination of the main attribute type and entity - attribute value correlation.

At step 150, duplication is minimized according to the user's request for the sorted summary, and the priority is taken from the order of priority among the sorted summaries. Minimize the duplication of attribute types and attribute values used in the entity description statement, and generate a summary of the length required by the user.

In other words, some of the summaries are taken and returned according to the user's request. If the user requests n in length of the summary, n triples are taken from the final result and returned to the user. In particular, if the n lengths required by the user are extremely small (n = 5) in order to minimize redundancy, which is the essential function that the final result should have, the redundancy between the triples included in the final result is limited as follows:

Final summary = Duplicate attribute type Not allowed ∧ Attribute value Duplicate not allowed

Also, if the length of n requested by the user is increased (n = 10), the redundancy between the triples included in the final summary is adjusted as follows:

Final summary = Allow duplicate attribute type ∧ Allow duplicate attribute value

Figure 2 illustrates a single spatialization of category tags found in three different languages for an entity in accordance with an embodiment of the present invention.

In the present invention, in order to generate a summary of individual units in a knowledge base composed of triples (entities-attributes-object words) describing entities, the entities in the knowledge base are clustered by using the characteristics of the classification system of the knowledge base as qualities, This is used in the object summarization system. In order to better model the systematic characteristics of individuals, a knowledge extraction method and a knowledge extension device are added to the system by integrating various knowledge bases.

All of the object summary techniques so far focused on the collected object information based on a single language (for example, English). However, when published data collects information about an entity on a massive web, one of the most important things is that the information about the entity may differ depending on the source of the information gathered. FIG. 2 shows semantic classification schemes 210, 220 and 230 existing in three different languages with respect to Jeju Island. There is a classification system that is found only in certain language databases such as "Volcanic Island", "Geography", and "Geopark", while there are classification systems that are found redundantly in all languages, such as "Biosphere Reserve". This indicates that the world wide known semantic classification system can be found redundantly in multiple languages and it is difficult to understand the problem of inconsistency due to differences in cultural viewpoints that occur in unknown or unknown language data This is an example. Therefore, if we can automatically integrate (240) a semantic tag for a single entity in multiple languages and perform a single-space significance analysis, the advantage of collecting non-biased information about an entity's unique attributes Lt; / RTI > It is also intended to provide a global and common summary of the entity.

The present invention provides an object summarizing system and an object summarizing method that reveal an object group optimized based on a single space on which a multilingual feature is projected and include essential information on an object basis while minimizing redundancy, Try a summary based on the concept of identifying unique properties. An entity-specific property represents information that is indispensable to be included in the description of an entity but is distinguishable from other objects. Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, this is merely an example and the present invention is not limited thereto.

FIG. 3 is a block diagram of a multi-language projected object space based entity summary generation system according to an exemplary embodiment of the present invention.

The proposed entity summary creation system 300 includes a multilingual feature projection module 311, an entity clustering module 312, an entity description ranking module 313, and an entity summary generation module 314.

The entity summary generation system 300 according to the present embodiment may include a processor 310, a bus 320, a network interface 330, a memory 340, and a database 350. The memory 340 may include an operating system 341 and an entity summary generation routine 342. The processor 310 may include a multilingual feature projection module 311, an entity clustering module 312, an entity description ranking module 313, and an entity summary generation module 314. In other embodiments, the entity summary generation system 300 may include more components than the components of FIG. However, there is no need to clearly illustrate most prior art components. For example, the entity summary generation system 300 may include other components such as a display or a transceiver.

The memory 340 may be a computer-readable recording medium and may include a permanent mass storage device such as a random access memory (RAM), a read only memory (ROM), and a disk drive. In addition, the memory 340 may store program codes for the operating system 341 and the entity summary creation routine 342. These software components may be loaded from a computer readable recording medium separate from the memory 340 using a drive mechanism (not shown). Such a computer-readable recording medium may include a computer-readable recording medium (not shown) such as a floppy drive, a disk, a tape, a DVD / CD-ROM drive, or a memory card. In other embodiments, the software components may be loaded into the memory 340 via the network interface 330 rather than from a computer readable recording medium.

The bus 320 may enable communication and data transfer between components of the entity summary creation system 300. The bus 320 may be configured using a high-speed serial bus, a parallel bus, a Storage Area Network (SAN), and / or other suitable communication technology.

The network interface 330 may be a computer hardware component for connecting the entity summary generation system 300 to a computer network. The network interface 330 may connect the entity summary creation system 300 to a computer network via a wireless or wired connection.

The database 350 may store and maintain all the information necessary for generating the entity summary. In FIG. 3, the database 350 is constructed to be included in the entity summary creation system 300, but it is not limited thereto and may be omitted according to the system implementation method or environment, It is also possible to exist as an external database built on a separate, separate system.

The processor 310 may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input / output operations of the entity summary generation system 300. The instructions may be provided by the memory 340 or the network interface 330 and to the processor 310 via the bus 320. The processor 310 may be configured to execute program code for the multilingual feature projection module 311, the entity clustering module 312, the entity description language ranking module 313, and the entity summary generation module 314. Such a program code may be stored in a recording device such as the memory 340. [

The multilingual feature projection module 311, the entity clustering module 312, the entity description ranking module 313 and the entity summary generation module 314 may be configured to perform the steps 110-150 of FIG. have.

The entity summary generation system 300 may include a multilingual feature projection module 311, an entity clustering module 312, an entity description ranking module 313, and an entity summary generation module 314.

The multilingual feature projection module 311 extracts triples that indicate classification schemes in a multilingual knowledge base and integrates information of the same entity unit. First, we construct a unit space from a plurality of knowledge bases written in various languages into one space. The triples formed in a plurality of languages are linked to the same entity to derive the qualities commonly used in a plurality of languages and the qualities used independently in each of the plurality of languages are derived. And integrates the classification scheme characteristics of the object for the corresponding object generated in a plurality of language communities.

The object clustering module 312 extracts a ternary relation marking the classification system from a multilingual knowledge base to construct an object clustering. Clustering refers to finding commonalities among individuals and dividing them into one set. From the triple describing the classification system existing in the multilingual knowledge base, the qualities for clustering individuals are derived and the corresponding entities of similar qualities are grouped together. In the process of clustering individuals, it is possible to acquire the common points of individuals from triples using specific property types that represent the classification systematic characteristics of individuals from the knowledge base.

For example, in the qualification selection process of object clustering, the boundaries of words consisting of noun phrases excluding common prefixes can be distinguished, and the stemming process for finding the root of each word can be performed.

Using a specific algorithm among the division methods widely used in the field of computation, the given object is divided into several clusters based on the previously selected qualities. The process of dividing the cluster is done by minimizing the sum of squares of the distance between the center of each cluster and the individual in the cluster as a cost function. In this process, the degree of similarity between individuals in the same cluster increases, and the similarity with individuals in other clusters decreases. This process can be replaced by using the existing ontology classification system. In the present invention, a k-means algorithm is used among the division methods in the field of computation, but the present invention is not limited thereto.

The Entity Description Ranking module 313 finds the key narrative relationships and major entity-object correlations for each cluster based on the entity clusters constructed in the clustering component, and calculates the weights of the triples in the multilingual knowledge base. In other words, we derive the most important attribute types for each of the generated entity clusters as described above. This process is defined as a statistical number that indicates how important an attribute type is in a particular cluster when there are several communities. The weight calculation is based on the main attribute type in the object cluster, which is a combination of the attribute type frequency used in the object cluster and the score indicating the reverse population frequency, and is based on the object-attribute value co-occurrence by the object cluster unit .

The weight of an attribute type is defined by a combination of the following two qualities: the attribute frequency and the inverse group frequency of the attribute type word in the cluster.

The frequency of attribute type words in a cluster uses the total frequency of attribute type words appearing in the cluster. The inverse group frequency indicates how common an attribute type word appears in the whole population. The value obtained by dividing the total number of populations by the number of populations including the attribute type word is used.

The higher the attribution type word frequency in a particular cluster, and the smaller the population including the attribute type word among all the clusters, the higher the weight value of the narrative relation. Using this, it is possible to obtain the effect of filtering the attribute type words that are common in all the clusters, so that it is possible to grasp the meaningful property types in the community.

The entity summary generation module 314 repeats the analysis of all the ternary relationships and sorts the summary according to the order of importance of all ternary relations based on the calculated weight. We derive the main attribute types of the triple according to the entity clusters and derive the entity - property value correlations and sort the summary according to the order of importance by using the combination of the main attribute type and entity - attribute value correlation.

Thereafter, the sorted summary is minimized in accordance with the user's request, and the sorted summaries are taken from the order of priority. Minimize the duplication of attribute types and attribute values used in the entity description statement, and generate a summary of the length required by the user.

In other words, some of the summaries are taken and returned according to the user's request. If the user requests n in length of the summary, n triples are taken from the final result and returned to the user. In particular, if the n lengths required by the user are extremely small (n = 5) in order to minimize redundancy, which is the essential function that the final result should have, the redundancy between the triples included in the final result is limited as follows:

Final summary = Duplicate attribute type Not allowed ∧ Attribute value Duplicate not allowed

Also, if the length of n requested by the user is increased (n = 10), the redundancy between the triples included in the final summary is adjusted as follows:

Final summary = Allow duplicate attribute type ∧ Allow duplicate attribute value

Hereinafter, a method and system for generating a multi-language feature-based object space based entity summary will be described in more detail with reference to FIG. 4 to FIG.

4 is a diagram illustrating a category tag for an entity existing in a Korean community according to an embodiment of the present invention.

The multilingual feature projection module described in FIG. 3 collects all the entities present in the knowledge base and extracts words used in a plurality of classification schemes from a set of Wikipedia documents created in one or more languages of the collected entities.

A Wikipedia document set is a collection of documents written in a plurality of languages, each document containing information about one or more classification schemes for representing the author's background and editor's background on a particular entity and the author's cultural background. For example, the classification system word "People from Trelawny Parish" that exists in English Wikipedia on an object of "Usain Bolt" does not exist in Korean Wikipedia, and this information is used only in English Wikipedia. It is information that can not be done. The present invention can integrate specific classification schemes of various languages into a single vector space to calculate the overall statistical-based score for the words used.

The classification system is an important resource for processes such as information extraction, retrieval, etc., and is extracted from Wikipedia category tags. Fig. 4 shows the category tags that exist for the object " Usain ' bolt " existing in Korean Wikipedia. This is a collaborative tagging between authors who constitute a kind of Wikipedia document set for an entity. Data quality is maintained by using collective intelligence.

5 is a diagram illustrating a category tag for an entity existing in an English community according to an embodiment of the present invention.

Algorithms for extracting a specific classification scheme from a Wikipedia document are well known in the art and will not be described here. The classification scheme data extracted between different languages does not match the amount and range of the objects included. Therefore, the pivot language is set to English and all the words used in the English classification system are generated as the classification pivot vectors. Figure 5 shows the category tag used in the object " Usain Bolt " found in the English Wikipedia document set, which is a pivot language. This is different from FIG. 4 in which the category tag of the corresponding object existing in the Korean Wikipedia is shown, and the word sequence constituting the classification system is also different.

6 is a diagram illustrating vectorization of a root found from a classification system word according to an embodiment of the present invention.

The classification system pivoted vector consists of dividing the boundaries of words consisting of noun phrases constituting the classification system, stemming the root of each word, and then generating a vector. The pivot classification scheme vector has the word sequence of the root unit in its length and the pivot vector calculates the score (610) for the word-based classification word in each word sequence of the classification system found in the English Wikipedia document set. Next, the process of integrating the classification system extracted from the languages other than English, which is a pivotal language, into the Pivot classification system vector is performed.

FIG. 7 is a diagram illustrating the addition of weights to the vectorization of a root found from a classification system word according to an embodiment of the present invention.

The process of entity correspondence between different languages can reveal interlanguage links existing in Wikipedia using SPARQL queries, which has the same effect as using two different interlanguage translations. The processing of the SPARQL query for the knowledge base is well known in the technical field of the present invention, and a description thereof will be omitted here. The weight of the translated English words corresponding to the words extracted from the set of different language Wikipedia documents is added to the existing Pivot classification system vector. At this time, the weight for the word root existing in the existing pivot vector increases by the number of times of finding. 7 is an example showing that the weights 710 for the specific words " Jamaican 721 " and " sprinters 722 " are increased by +2 and +1, respectively, from the weights in FIG. Additional found words not present in the pivot vector are not considered in the present invention.

Using a word vector extracted from the previously selected classification system, the given object is divided into several clusters by using a specific algorithm among the division methods widely used in the field of computer. The present invention means that the < property-value > pair shared by the cluster members (neighbors) of an entity is more important than the function shared by the individuals not in the cluster. For example, if there are two communities of A = {"Usain Bolt", "Carl Lewis", "Michael Johnson"}, B = {"Babe Ruth", "Hyun-jin Ryu"}, Usain Bolt "has the required attributes such as" Sport Event "or" Medal Information ", but" Babe Ruth "can generate summaries with more emphasis on his" Position "or" Team ".

In this case, the process of dividing a plurality of clusters from a plurality of objects is performed by a method of minimizing the sum of squares of the distances between the centers of the clusters and the objects in the clusters as a cost function. In this process, And the degree of similarity with individuals belonging to other clusters is decreased. In the present invention, a k -means algorithm is used in the division method, but the present invention is not limited thereto.

8 is a diagram illustrating a triple set comparison of two entities of a cluster according to an embodiment of the present invention.

This process is defined as a statistical number that indicates how important an attribute type is in a particular cluster when there are several communities. For example, as shown in FIG. 8 below, an attribute type 810 of two objects "Usain_Bolt" belonging to the same community and an attribute type 820 of "Michael_Johnson_ (sprinter) birthPlace, dbo: sport, and dbo: event, which is an important descriptive relationship for describing two entities, but the commonly used attribute types dbo: honorificSuffix and dbo: collegeteam It can not be seen as a necessary element to describe.

Thus, the weight of an attribute type is defined by a combination of two qualities: the attribute frequency and the inverse group frequency of attribute type words in the cluster.

Property Frequency in Cluster: Property Frequency: The total frequency of the attribute type words appearing in the cluster is used as Equation (1).

수학식(1)

Equation (1)

Inverse Group Frequency: Indicates how common an attribute type word appears in the entire set of clusters. The value obtained by dividing the total number of clusters by the number of clusters containing the corresponding attribute type word, Equation (2) is obtained.

수학식(2)

Equation (2)

9 is a diagram illustrating a duplicate attribute type for one entity according to an embodiment of the present invention.

The higher the attribution type word frequency in a particular cluster, and the smaller the population including the attribute type word among all the clusters, the higher the weight value of the narrative relation. Using this, it is possible to obtain the effect of filtering the attribute type words that are common in all the clusters, so that it is possible to grasp the meaningful property types in the community.

수학식(3)

Equation (3)

In Equation (3), e represents a subject of a triple, i.e., a given entity, and v represents an attribute value of a triple. (s, p, o) represents a triple present in the knowledge base, and E (e) represents a population to which a given object e belongs. | x | represents the number of circle introductions belonging to the corresponding set x. At this stage, we derive the most important subject-object (ie, object-attribute value) relationship for each of the population units. This process is a process of determining a major relative entity based on an entity, and an example is shown in FIG.

Here, the correlation between two objects defined by the same attribute type, dbr: Spanish_Town (910) and dbr: Jamaica (920), for a given object, Usain_Bolt, is scored so that the more important property value The weighting factors are calculated to determine the following equation.

수학식(4)

Equation (4)

In Equation (4), v represents an object in a triple. That is, the more the two entities wanting to know the interconnection weights are found together in triplets within each cluster, the higher the score. That is, the weight v-score is calculated based on the co-occurrence of the two individuals constituting the triple. Specifically, the first two elements connected by addition operations are correlation-based scores for two entities of subject and object, and are normalized to the range [0-1].

In the next step, the entire triple of the object unit (that is, the object description statement) is sorted through the product of the two weights calculated in the above step. The score for the final sort between triples is the product of the "weight between subject-attribute values" value independent of the value of the independent "attribute type weight" and the "weight between attribute-type weight" and "subject- And the equation is as follows.

수학식(5)

Equation (5)

FIG. 10 is a diagram illustrating a final summary comparison according to whether a duplicate attribute type is permitted according to an embodiment of the present invention.

In the next step, some of the summaries are taken and returned according to the length requested by the user. That is, if the user requests n in length of the summary, n triples are taken from the final result and returned to the user. In particular, if the n lengths required by the user are extremely small (n = 5) in order to minimize redundancy, which is the essential function that the final result should have, the overlap between the triples included in the final result is limited as follows.

- Final summary = Duplicate attribute type not allowed ∧ Duplicate attribute value not allowed

That is, the attribute type used in the ternary relationship included in the final summary is used solely between the subject-attribute values, and the attribute value can not be found more than once in the plurality of attribute types, and the final summary state 1010 and the final summary impossible An example of state 1020 is shown in FIG.

11 is a diagram illustrating a final summary comparison according to whether or not a duplicate attribute value is permitted according to an embodiment of the present invention.

10, if the length n required by the user is increased (n = 10), the overlap between the triples included in the final summary is adjusted as follows.

- Final summary = Duplicate attribute type ∧ Duplicate attribute value allowed

That is, the attribute types used in the triples included in the final summary can be found multiple times with different attribute values, and the object can also be used multiple times with multiple attribute types, and the final summary possible state (1110) An example of state 1120 is shown in FIG.

In the following, a description will be given of a method for generating a multivariate projected object space based object summary and an experimental result of the system.

In order to verify the performance of the proposed technology, performance comparison evaluation was conducted through the same experimental data used in the existing technology. Experimental data used are given by triple sets of fifty given DBpedia objects of fifteen independent users, and the important triple of each object is selected as Top5 and Top10. do. The performance of the system can be measured from the correct answer data with the same quality as Equation (6).

수학식(6)

Equation (6)

In Equation (6), Summ (e) represents the summary result generated by the system for the object e and SummiI (e) represents the summary selected by the i-th user from the data used as the correct answer data for the given object e. The performance of the system is calculated as the average value of all users included in the correct answer data. Since the information of the triple to be included in the summary to be generated is different for each user who generates the correct answer data, a target value of a quality that the system can ideally reach is given, and the target value is determined according to equation (7). For reference, the target values indicating the correspondence between the correct data generated by a plurality of users are 1.9596 (n = 5) and 4.6770 (n = 10).

수학식(7)

Equation (7)

Table 2 shows the performance of the proposed state-of-the-art (FACES) and the proposed technique when the summary generation length is 5 and 10, respectively. In this experiment, two comparative group systems were added to analyze the efficiency of the entity cluster based method. Specifically, we compared the cases where the proposed technology was applied to one language environment for comparison between a single language environment and various language projected environments (comparative group 1). We added hazard comparison group 2.

- FACES : Prior art

- Multi-EGS : Proposal technology, Multilingual feature projection based classification system

- EGS : proposed technology (comparative group 1), based on single language classification system

- Typed : based on the proposed technology (comparative group 2), single language ontology type

Table 1 shows that the multi-EGS, EGS, and Typed systems are superior to FACES using the existing knowledge base partitioning technique. In addition, we can confirm that executing the clustering by using the classification tag is superior to using the predefined type in the similar technology ontology, and it is confirmed that the performance improvement according to the multi-EGS of the multilingual classification system there was.

<Table 1>

Table 2 shows the result of the detailed comparison between the conventional technology and the proposed technology, and the comparative analysis of the result of the summary of the evaluation data for the object having the highest quality and the lowest quality in the proposed method. Compared to the proposed technique, the quality score of the existing method is remarkably high. However, it is found that the average quality score is high as it is calculated similar to the correct answer of many users due to the increase of object redundancy, but it is not suitable as a summary because it contains redundant contents . On the other hand, the summary according to the proposed method not only includes the main characteristics for a given entity but also can represent the representativeness of the entity by minimizing the redundancy in the summary.

<Table 2>

In the prior art, in an environment where object-oriented data is rapidly increasing, a set of triples composed of <object-attribute-object> becomes too large to quickly identify important information. It is possible to classify individual objects by the clustering of the objects proposed in the present invention. By integrating the relative individual characteristics generated from various language resources, it is possible to improve the performance of the object clustering Improvement. The present invention can expect a more effective summary result system by reproducing the summary method as close as possible to the summary method of the experts including the essential items for describing the entity. It is also possible to efficiently search information and prompt inquiry on a per unit basis on a large- And may be usefully used to provide processing.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA) A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device As shown in FIG. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media 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 machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like.

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. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI &gt; or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (12)

Extracting triples marking a classification system from a multilingual knowledge base, and integrating information of the same entity unit;
Extracting a ternary relation from the multilingual knowledge base to mark the classification system, and constructing the object clustering;
Searching for major descriptive relationships and major object-object correlations for each of the clusters based on the individual clusters constructed in the clustering component, and calculating the weights of the triples in the multilingual knowledge base;
Repeating the analysis of all ternary relations, and sorting the summary according to the order of importance of all ternary relations based on the calculated weight; And
Minimizing the redundancy according to the user's request for the sorted summary and fetching from the priority order of the sorted summary
/ RTI &gt; The method according to claim 1,
Extracting a triple marking a classification scheme from a multilingual knowledge base, and integrating information of the same entity unit,
The present invention relates to a method for deriving qualities commonly used in a plurality of languages by linking triples created in a plurality of languages with respect to the same entity and deriving qualities independently used in each of the plurality of languages, Integrating the classification system characteristics of the object
How to create an entity summary. The method according to claim 1,
In the multi-linguistic knowledge base, the step of extracting the ternary relations that mark the classification system and constructing the object clustering,
We derive qualities for clustering entities from triples that describe classification schemes existing in a multilingual knowledge base, and cluster the corresponding entities of similar qualities
How to create an entity summary. The method according to claim 1,
The step of finding the main narrative relation and major object-object correlation according to the clusters based on the individual cluster constituted in the clustering component and calculating the weight of the triple of the multilingual knowledge base,
Based on the main attribute type in the object group consisting of a combination of the attribute type frequency used in the object group and the score indicating the reverse group frequency, and based on the object-attribute value co-
How to create an entity summary. The method according to claim 1,
The steps of repeating the analysis of all ternary relations and sorting the summary according to the order of importance of all ternary relations based on the calculated weight,
We derive the main attribute types of the triple according to the object clusters and derive the object-attribute value correlations and sort the summaries according to the order of importance by using the combination of the main attribute type and object-attribute value correlation
How to create an entity summary. The method according to claim 1,
The step of minimizing redundancy according to a request of a user for an ordered summary and the order of priority among the ordered summary,
Minimize the duplication of attribute types and attribute values used in object description statements, and generate a summary as long as the user requires
How to create an entity summary. A multilingual feature projection module for extracting triples marking a classification system in a multilingual knowledge base and integrating information of the same entity unit;
An object clustering module for extracting ternary relations from a multilingual knowledge base and constructing object clustering;
Based on the individual clusters constructed in the clustering component, the major descriptive relations of the clusters are searched, the weights of the triples of the multilingual knowledge base are calculated, and the main object-object correlations of the clusters are found based on the individual clusters constructed in the clustering component An entity descriptor ranking module for calculating a weight of a triple of a multilingual knowledge base; And
The analysis of all ternary relations is repeated through the descriptive relationship analysis unit and the object-object analysis unit, and the summary is sorted according to the order of importance of all the ternary relations based on the calculated weights. An object summary generation module that minimizes the redundancy and obtains the sorted summary from the order of priority,
The object summary generating system comprising: 8. The method of claim 7,
The object clustering module,
The present invention relates to a method for deriving qualities commonly used in a plurality of languages by linking triples created in a plurality of languages with respect to the same entity and deriving qualities independently used in each of the plurality of languages, Integrating the classification system characteristics of the object
Object summary creation system. 8. The method of claim 7,
The object clustering module,
We derive qualities for clustering entities from triples that describe classification schemes existing in a multilingual knowledge base, and cluster the corresponding entities of similar qualities
Object summary creation system. 8. The method of claim 7,
The Entity Knowledge Ranking module,
Based on the main attribute type in the object group consisting of a combination of the attribute type frequency used in the object group and the score indicating the reverse group frequency, and based on the object-attribute value co-
Object summary creation system. 8. The method of claim 7,
The object summary generation module,
We derive the main attribute types of the triple according to the object clusters and derive the object-attribute value correlations and sort the summaries according to the order of importance by using the combination of the main attribute type and object-attribute value correlation
Object summary creation system. 8. The method of claim 7,
The object summary generation module,
Minimize the duplication of attribute types and attribute values used in object description statements, and generate a summary as long as the user requires
Object summary creation system.

Images