KR20210064862A - Knowledge Graph Generation Method and Renewable Energy Item Recommendation System and Method - Google Patents

Abstract

The present invention relates to a method for recommending a new/regenerable energy generation item to provide valid relevant information. According to the present invention, the method comprises the following steps: receiving inquiries or asks about items related to new/renewable power generation from power generation companies; forming initial knowledge graph facts in which a head entity and a tail entity consist of items related to new/renewable power generation and relationship between the head entity and the tail entity is defined; collecting information on relationship paths for the head and tail entities of each fact according to accumulated user's inquiries or ask activities; adding reliability functions for the relationship paths to each fact to form a reliability fact; updating a reliability score of each relationship path of each fact according to the accumulated user's inquiries or ask activities; when the updated reliability score is lower than a cutoff reference value, deleting a corresponding relationship path; and notifying the user of items in which the updated reliability score is higher than the reference value among items matching the inquiry or ask of the power generation companies.

Description

Knowledge Graph Generation Method and Renewable Energy Item Recommendation System and Method using the same

The present invention relates to a method for generating a knowledge graph based on type constraint-based relationship path learning, and more particularly, to a method for generating a knowledge graph for improving the quality of an energy knowledge graph for a new and renewable power microgrid, and a new and renewable power recommendation system using the same .

As one of the effective data expression methods to reflect the meaning of data, the knowledge graph is being used and attracting attention in many applications such as knowledge extraction, Q&A, and information retrieval.

In general, the knowledge graph is composed of a set of triples <h,r,t> called facts (facts), and the head entity (h) and the tail entity (t) are connected by a semantic relationship (r). Various knowledge graphs such as DBpedia, WikiData, Freebase, WordNet, and Google Knowledge Graph have been developed for massive heterogeneous knowledge integration and interoperability.

However, knowledge graphs composed of a plurality of triples are incomplete because a lot of information is missing or inaccurate, so it is difficult to actively utilize them in the real world.

Looking at this individually, first, the translation-based knowledge graph completion technique embeds the entities and relationships of the knowledge graph into a vector space, and performs expression learning so that the structural meaning of the knowledge graph can be contained in the vector expression. However, most of the translation-based knowledge graph completion techniques, including TransE, are only useful for inferring and selecting only a direct relationship with a path length of 1 between two entities. Therefore, when learning a relationship path having a length of 2 or more between two entities, there is a problem in that the accuracy of relationship inference and the accuracy of relationship validity determination decrease.

In order to improve the accuracy of relational path learning during relational inference, various relational path learning techniques such as PTransE, RTransE, COMP, RPE, PaSKOGE, and CKRL have been proposed. However, techniques based on conventional learning techniques reflect both unnecessary paths and semantically insignificant relational paths in learning, and as a result, there is a problem of inaccurate relational inference.

As a method to find a semantically major relationship path, the PRA (Path Ranking Algorithm) and PCRA (Path-specific Resource Allocation) algorithms have been proposed, but there is no reliability measurement method that reflects the structural information of the knowledge graph. It is difficult to find a reliable relationship path that guarantees

Korean Public Publication No. 10-2018-0093158

An object of the present invention is to provide a method for generating a knowledge graph that can search a knowledge graph relational path having high reliability and a semantically effective knowledge graph, and a new regeneration development recommendation system using the same.

Renewable power recommendation method according to an aspect of the present invention, the step of receiving an inquiry or inquiry about a new and renewable power generation related items of a power plant; Forming initial knowledge graph facts in which the head entity and the tail entity are composed of items related to new and renewable development, and the relationship between the head entity and the tail entity is defined; collecting information on relationship paths for the head entity and the tail entity of each fact according to the accumulated inquiry or inquiry activity of the power generation companies; adding reliability functions for relationship paths to each fact to form a reliability fact; updating the reliability score of each relationship path of each fact according to the accumulated user's inquiry or inquiry activity; if the updated confidence score is lower than a cutoff reference value, deleting a corresponding relationship path; and notifying the user of items in which the updated reliability score is higher than a reference value among items matching the inquiry or inquiry of the power generation business operator.

A knowledge graph generation method according to another aspect of the present invention comprises the steps of: forming initial knowledge graph facts in which a relationship between a head entity and a tail entity, and a head entity and a tail entity is defined; collecting information on relational paths for the head entity and the tail entity of each fact according to the user's inquiry or inquiry activity; adding reliability functions for relationship paths to each fact to form a reliability fact; updating the reliability score of each relationship path of each reliability fact according to the accumulated user's inquiry or inquiry activity; and if the updated confidence score is lower than the cutoff reference value, deleting the corresponding relationship path.

Here, the initial knowledge graph facts may be templates for graph facts generated based on future real data.

Here, in the step of collecting information on the relationship paths, the types of the head entity, tail entity, relationship, and relationship path of each fact are determined, and items with low similarity in types (head entity, tail entity, relation, relationship) are determined. path) can be deleted.

Renewable power generation recommendation system according to another aspect of the present invention receives a request for inquiry or inquiry about an item related to renewable power from a terminal of a power generation company, and identifies an item that is optimal for the received inquiry or inquiry recommendation service server; And to confirm the optimal item, reliability functions for relational paths for the head and tail of the fact are added to the fact generated from the initial knowledge graph facts in which the relationship between the head and tail is defined. It may include a knowledge graph server that prepares and provides a knowledge graph of one reliability fact and performs learning on the knowledge graph.

Here, the knowledge graph server comprises the steps of: forming initial knowledge graph facts in which a relationship between a head entity and a tail entity, and a head entity and a tail entity is defined; collecting information on relational paths for the head entity and the tail entity of each fact according to the user's inquiry or inquiry activity; adding reliability functions for relationship paths to each fact to form a reliability fact; updating the reliability score of each relationship path of each reliability fact according to the accumulated user's inquiry or inquiry activity; and if the updated confidence score is lower than the cutoff reference value, deleting the corresponding relationship path may be performed.

When the method for generating a knowledge graph of the present invention configured as described above and the new regeneration generation recommendation system using the same are implemented, there is an advantage in that it is possible to provide effective related information by searching a knowledge graph relational path with high reliability and semantically effective.

The knowledge graph generation method of the present invention and the new regeneration generation recommendation system using the same have the advantage of automatically detecting erroneous data among data constituting the knowledge graph and automatically inferring lost data.

The method for generating a knowledge graph of the present invention and a new regeneration generation recommendation system using the same have advantages of high accuracy in inferring new relationships and detecting incorrect relationships in the knowledge graph and excellent scalability.

1 is a flowchart illustrating a method for generating a knowledge graph according to the spirit of the present invention.
FIG. 2 is a flowchart in which the knowledge graph generation method shown in FIG. 1 is applied to a method of recommending items related to renewable power generation.
3 is a block diagram illustrating an embodiment of a renewable power generation recommendation system for performing a method of recommending items related to renewable power generation according to the flowchart of FIG. 2 .

In describing the present invention, terms such as first, second, etc. may be used to describe various components, but the components may not be limited by the terms. The terms are only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component.

When a component is referred to as being connected or connected to another component, it may be directly connected or connected to the other component, but it can be understood that other components may exist in between. .

The terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention. The singular expression may include the plural expression unless the context clearly dictates otherwise.

In this specification, the terms include or include are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, and includes one or more other features or numbers, It may be understood that the existence or addition of steps, operations, components, parts or combinations thereof is not precluded in advance.

In addition, shapes and sizes of elements in the drawings may be exaggerated for clearer description.

If a score function is simply introduced to measure the accuracy of relationship inference in the conventional relationship path learning technique, only intermediate relationships on the relationship path are considered, but intermediate entities included on the relationship path are not considered. In other words, even if the intermediate relationship on the relationship path is the same, it does not take into account that the meaning of the relationship path changes depending on which intermediate entity is included, and as a result, the accuracy of the relationship inference and the accuracy of the relationship validity judgment decreases. .

For example, two triples (Leonardo DiCaprio, Awarded, Best Actor), (Leonardo DiCaprio, nominated, Best Actor) and two relationship paths (Leonardo DiCaprio, Appears, The Wolf of Wall Street, Entered, Oscar) , award, Best Actor) (Leonardo DiCaprio, Appears, Revenant, Entered, Oscar, Award, Best Actor), if no intermediate entity is considered (Leonardo DiCaprio, Appears, The Wolf of Wall Street) , entered, Oscar, award, Best Actor) through the relationship path (Leonardo DiCaprio, Suspicious , Best Actor) may occur.

The present invention proposes an expression learning technique that reflects the meaning of intermediate entities and intermediate relationships included in the knowledge graph relationship path and reflects the reliability score of the relationship path. The proposed score function is structured so that the characteristics of the relationship path can be well reflected in learning when learning by embedding the entities and relationships constituting the knowledge graph into a vector space.

In addition, we propose a knowledge graph relationship path evaluation algorithm KGPRA (Knowledge Graph Path Ranking Algorithm) to explore a knowledge graph relationship path that is highly reliable and semantically valid.

1 is a flowchart illustrating a method for generating a knowledge graph to which KGPRA (Knowledge Graph Path Ranking Algorithm) is applied according to the spirit of the present invention.

The illustrated knowledge graph generation method includes the steps of forming initial knowledge graph facts in which the relationship between a head entity and a tail entity, and a head entity and a tail entity is defined (S10); According to the inquiry and/or inquiry activity of the users, collecting information on the relationship paths for the head entity and the tail entity of each fact (S210); adding reliability functions for relationship paths to each fact to form a reliability fact (S240); updating the reliability score of each relationship path of each reliability fact according to the accumulated user's inquiry and/or inquiry activity (S270); and if the updated reliability score is lower than the cutoff reference value, deleting the corresponding relationship path (S280).

Depending on the implementation, the initial knowledge graph facts generated in step S10 may be formed before learning using real data, and in this case, the initial knowledge graph facts may be generated by learning sample data for reference in the relevant field.

According to implementation, the initial knowledge graph facts generated in step S10 may be templates for graph facts generated based on future real data. In this case, graph facts may be formed as instances for the template by reflecting real data.

In step S210, the user's inquiry and/or inquiry information may be reflected in the relationship of each fact, a new fact may be formed, and information on the relationship paths may be collected.

In performing the above-described operation in step S210, the type of each head entity, tail entity, relationship, and relationship path may be determined. Depending on the implementation, it is possible to record the determined type or delete items with low similarity of types (head object, tail object, relationship, relationship path).

The reliability fact is a fact of the form of limitation in the present invention, and is obtained by adding reliability functions for relationship paths that may exist between the head entity and the tail entity to the existing triple factor. The step S240 may be performed by analyzing a plurality of facts obtained from the actual data accumulated and collected in the step S210 according to the inquiry and/or inquiry activity of the users.

The step S270 is a process of learning the reliability of the relationship path constituting the reliability fact proposed by the present invention based on a kind of real data.

The step S280 can be viewed as a process of reflecting the learning result of the step S270 to the reliability fact, and depending on implementation, the cutoff reference value is fixed, adjusted according to the application field, or given as a ratio of the average to the overall reliability. can

As shown, the learning process consisting of steps S270 and S280 may be repeatedly performed based on actual data obtained during actual system operation.

Hereinafter, a specific algorithm applied to the knowledge graph generation method will be described.

)을 함께 고려하기 위한 것이다. 상기 제안하는 스코어 함수는 상기 신뢰성 팩트를 규정한다.The proposed representation learning technique will be named RKRL (Reliable Knowledge Graph Path Representation Learning). RKRL is based on a new score function, and the score function proposed according to Equation 1 below is an index for evaluating the connection possibility between the head entity (h) and the tail entity (t), and the relationship between h and t (r) At the same time taking into account the semantically valid relational paths between h and t (

) to be considered together. The proposed score function defines the reliability factor.

)의 집합을 의미하며,

는 KGPRA를 통해 계산되는 관계 경로의 신뢰도 점수이다. In Equation 1, the two functions f(h, r, t) and f(h, Π, t) are, respectively, a linkability score with a relationship r directly connected between h and t, and a relationship path between h and t their connectability scores were calculated. The score functions f(h, r, t) and f(h, Π, t) are defined as in Equation 2 below. v() means a vector value. For example, v(h) means a vector expression of the entity h. Π is the possible relationship path between the head entity (h) and the tail entity (t) (

) means a set of

is the confidence score of the relationship path calculated through KGPRA.

는 하기 수학식 3과 같이 정의된다. 이에 따라, 관계 경로의 연결 가능성을 평가하기 위해 관계 경로상에 포함된 중간 관계들을 활용할 뿐 아니라, 중간 개체들의 가중치를 활용함으로써 경로상에 포함된 개체들의 의미가 학습에 반영될 수 있도록 할 수 있다. In Equation 2, a function for calculating the connectability score of each relationship path

is defined as in Equation 3 below. Accordingly, in order to evaluate the linkability of the relationship path, not only the intermediate relationships included in the relationship path are used, but also the meaning of the entities included in the path can be reflected in learning by using the weights of the intermediate entities. .

는 중간 개체의 가중치 벡터를 나타내며,

는 관계 경로의 벡터 표현을 나타낸다. In Equation 3 above,

is the weight vector of the intermediate object,

represents the vector representation of the relationship path.

According to the above methods, two triple shown earlier (Leonardo DiCaprio, water is associative best actor), (Leonardo DiCaprio, rise to the candidate, Actor) invalid in the wrong relationship path (Leonardo DiCaprio, it is awarded, Actor) relationship Inferences can be prevented from occurring.

Here, the objective function (loss function) L for learning RKRL is defined as in Equation 4 below. Assume that the dataset T used for training consists of triples in which h, r, and t are actually connected, and that dataset T' consists of triples that are incorrectly connected to one element of h, r, t constituting each triple. do.

The knowledge graph relationship path evaluation algorithm KGPRA (Knowledge Graph Path Ranking Algorithm) proposed in the present invention searches for a knowledge graph relationship path that has guaranteed reliability and is semantically effective. The RDF (Resource Description Framework) schema that describes the structure of the knowledge graph has constraints to ensure semantic validity between objects and relationships. KGPRA excludes and removes the path that violates the type constraint from the search when searching for a relationship path. yes no?

로 이어져 있을 때 관계

와 연결된 타 개체의 수가 많을 시 신뢰도가 낮다고 평가하는 PCRA의 가정을 따른다. 뿐만 아니라 제안하는 기술은 현재 개체와 연결 가능한 모든 관계

중 관계

로 연결된 경로의 비중에 따라 신뢰도를 달리 평가한다. 즉, 현 개체에서 관계 r의 비중이 낮을수록 주요한 의미를 담고 있으며 신뢰도가 높다고 평가한다. 하기 수학식 5는 관계 경로의 신뢰도

과 관계

의 비중

을 표현한 것이다.Also, KGRPA indicates that objects are related

relationship when connected with

It follows the assumption of PCRA, which evaluates to low reliability when the number of other individuals connected to Not only that, but the proposed technology is all about all possible relationships with the current object.

relationship

Reliability is evaluated differently according to the weight of the paths connected to the In other words, the lower the weight of the relationship r in the current entity, the more important it is and the higher the reliability is evaluated. Equation 5 below is the reliability of the relationship path

relationship with

share of

will express

는 관계 경로상 포함된 임의의 개체를 나타내며,

는 관계

와 관계

로 이어진 머리 개체를 나타내며,

는 개체

와 관계

로 연결된 꼬리 개체들을 의미한다.

는 개체

와 관계

로 이어진 연결 수

가, 개체

와 연결된 모든 연결 수

중 얼마만큼의 비중을 차지하는지 계산한다.In Equation 5 above

represents any entity included in the relationship path,

is the relationship

relationship with

Represents a head object connected by

is the object

relationship with

means tail entities connected by

is the object

relationship with

number of connections leading to

autumn object

number of all connections associated with

Calculate how much of the

는 RDF 스키마의 domain, range 두 타입 제약 조건을 만족하는 개체들로만 구성된 것이다. KGPRA는 지식 그래프에 내재된 스키마 정보를 관계 경로 탐색 시에 활용하여, 타입 제약 조건을 위배한 경로를 탐색하지 않을 수 있다.Each set in Equation 5 above

is composed only of objects that satisfy the two type constraints of the RDF schema, domain and range. KGPRA may not search for a path that violates the type constraint by using the schema information contained in the knowledge graph when searching for a relationship path.

2 is a flowchart in which the knowledge graph generation method shown in FIG. 1 is applied to a method of recommending items related to renewable power generation.

FIG. 3 shows an embodiment of a renewable power generation recommendation system that performs a method of recommending items related to renewable power generation according to the flowchart of FIG. 2 .

For example, the new and renewable power generation recommendation system may provide recommendation information helpful in selecting a business area to a person (collectively referred to as a power generation company) with an intention of a distributed power generation business.

For example, the renewable power generation recommendation system uses the regional power facility status (substation and distribution line) and map to display the location of the substation and distribution line, and the distance from the area where the power generation customer wants to power generation to the substation and distribution line , provides information on the status of related power facilities (whether or not circuit breakers can be expanded for the installation of distribution lines in the substation, the completion schedule for new/expanded facilities, etc.) By collecting such data, it is possible to suggest a recommended power plant type, a recommended size, and a recommended system linkage plan.

For example, the renewable power generation recommendation system utilizes meteorological data (insolation, temperature, precipitation, etc.) and land information (approved land price, location, slope, direction) to provide information on weather conditions and land types preferred by power generation customers by region. After confirming, areas with potential for increased renewable energy in the future can be recommended as power generation business areas.

The illustrated renewable generation recommendation system receives an inquiry or inquiry request for an item related to renewable power (ie, distributed power) from the terminal 100 of a power generation operator, and an item optimal for the received inquiry or inquiry Recommendation service server 300 to check and recommend; and a knowledge graph server 200 that creates and provides a knowledge graph according to the spirit of the present invention so as to confirm the optimal item, and performs learning on the knowledge graph.

The power generation operator may access the recommendation service server 300 in a wired/wireless communication method using the terminal 100 of the asset, and the recommendation service server 300 is a specific information transmitted from the terminal of the connected power generation operator. You can receive information related to new and renewable power generation, such as address, capacity, and desired date. To this end, the recommendation service server 300 may have general configurations that perform member management, wired/wireless connection management, communication packet processing, etc., perform customer guidance visualization of analysis result information, and configure a guidance UI screen It may be provided with an output interface means for In addition, the recommendation service server 300 stores customer information (as member information) necessary for performing the recommendation service, and the customer DB 360, and information on the power plant that is currently operating, is scheduled to be operated in the future, or is desired to be operated. It may be provided with a power plant DB (370) to store.

The knowledge graph server 200 performs the method for generating the knowledge graph shown in FIG. 1 , and may include an arithmetic device for performing an operation for generating the knowledge graph, a storage means for storing the generated knowledge graph facts, and the like. .

The method of recommending an item related to renewable power generation shown in FIG. 2 includes the steps of receiving an inquiry or inquiry for a renewable power related item from a power company (S100); Forming a knowledge graph for items related to renewable power generation (S200); and providing item information with high reliability on the formed knowledge graph (S400).

Here, the step S200 comprises the steps of forming initial knowledge graph facts in which the head entity and the tail entity are composed of items related to new and renewable development, and the relationship between the head entity and the tail entity is defined (S10); According to the accumulated inquiry and inquiry activities of the power generation companies, collecting information on the relationship paths for the head and tail entities of each fact (S210); adding reliability functions for relationship paths to each fact to form a reliability fact (S240); updating the reliability score of each relationship path of each fact according to the accumulated user's inquiry and/or inquiry activity (S270); and if the updated reliability score is lower than the cutoff reference value, deleting the corresponding relationship path (S280). Steps S10 to S280 are similar to the processes shown in FIG. 1 , and overlapping descriptions will be omitted.

In the step S400, only the relationships in which the reliability score is higher than the reference value are notified to the user according to the inquiry request/inquiry of the power generation company.

For example, the inquiry/inquiry item of the power generation company in step S100 is retrieved from the head object or tail object of the knowledge graph facts accumulated by performing step S200, and among the knowledge graph reliability facts of the searched entity, the fact that exceeds the reliability standard value of the facts classified in step S300, the head object or tail object corresponding to the inquiry/inquiry item, and information on the corresponding relationship may be provided to the power generation company as recommended information.

Those skilled in the art to which the present invention pertains should understand that the present invention can be embodied in other specific forms without changing the technical spirit or essential characteristics thereof, so the embodiments described above are illustrative in all respects and not restrictive. only do The scope of the present invention is indicated by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention. .

100: power generation operator terminal
200: knowledge graph server
300: service server

Claims (6)

Receiving inquiries or inquiries about items related to new and renewable power generation from power generation companies;
Forming initial knowledge graph facts in which the head entity and the tail entity are composed of items related to new and renewable development, and the relationship between the head entity and the tail entity is defined;
collecting information on relationship paths for the head entity and the tail entity of each fact according to the accumulated inquiry or inquiry activity of the power generation companies;
adding reliability functions for relationship paths to each fact to form a reliability fact;
updating the reliability score of each relationship path of each fact according to the accumulated user's inquiry or inquiry activity;
if the updated confidence score is lower than a cutoff reference value, deleting a corresponding relationship path; and
Notifying the user of items in which the updated reliability score is higher than a reference value among items matching the inquiry or inquiry of the power generation company
Renewable power generation recommendation method, including.
Forming initial knowledge graph facts in which the relationship between the head entity and the tail entity, and the head entity and the tail entity is defined;
collecting information on relational paths for the head entity and the tail entity of each fact according to the user's inquiry or inquiry activity;
adding reliability functions for relationship paths to each fact to form a reliability fact;
updating the reliability score of each relationship path of each reliability fact according to the accumulated user's inquiry or inquiry activity; and
If the updated confidence score is lower than the cutoff reference value, deleting the corresponding relationship path;
A method of generating a knowledge graph comprising
3. The method of claim 2,
The initial knowledge graph facts are a knowledge graph generating method that is a template for graph facts generated based on future real data.
3. The method of claim 2,
In the step of collecting information on the relationship paths,
Determining the type of head entity, tail entity, relationship, and relationship path of each fact,
A knowledge graph creation method that deletes items with similar types (head object, tail object, relationship, relationship path).
a recommendation service server that receives an inquiry or inquiry request for an item related to renewable power generation from a terminal of a power generation company, and recommends an item that is optimal for the received inquiry or inquiry; and
In order to confirm the optimal item, reliability functions for the relationship paths for the head and tail are added to the fact generated from the initial knowledge graph facts in which the relationship between the head and tail is defined. A knowledge graph server that creates and provides a knowledge graph of reliability facts and performs learning on the knowledge graph
Renewable power generation recommendation system that includes.
6. The method of claim 5,
The knowledge graph server,
Forming initial knowledge graph facts in which the relationship between the head entity and the tail entity, and the head entity and the tail entity is defined;
collecting information on relational paths for the head entity and the tail entity of each fact according to the user's inquiry or inquiry activity;
adding reliability functions for relationship paths to each fact to form a reliability fact;
updating the reliability score of each relationship path of each reliability fact according to the accumulated user's inquiry or inquiry activity; and
If the updated reliability score is lower than the cutoff reference value, a new and renewable generation recommendation system for performing a knowledge graph generating method comprising the step of deleting a corresponding relationship path.

Images