Construction method based on electricity marketing inspection theme knowledge graph

Abstract

The invention relates to the technical field of data processing, in particular to a construction method based on a power marketing inspection theme knowledge graph. The construction method of the knowledge graph comprises the following steps: constructing a marketing inspection mode layer; performing entity and relation joint extraction by using a deep learning model; and storing marketing inspection knowledge. Compared with a traditional marketing inspection method, on one hand, structured, unstructured and semi-structured data are subjected to business carding and entity extraction; on the other hand, entity fusion is carried out on the conditions of entity ambiguity, inaccurate semantic run-through and the like generated in the application process by means of a semantic identification and run-through method, so that the consistency and uniqueness of service entities are achieved, and finally a semantic structure conforming to the marketing inspection service application knowledge graph is formed.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for constructing a knowledge graph based on a power marketing audit theme. Background Art

[0002] Marketing audit is an important tool for quality control and management level improvement of power marketing business, an important means to ensure the compliance and effectiveness of various professional work in power marketing, and an objective guarantee for providing better services to electricity users. Power companies attach great importance to the construction of marketing audit monitoring system, and the relevant organization, business and system systems are constantly improved. The level of marketing lean management and high-quality service continues to improve, and the results of improving quality and efficiency are obvious.

[0003] In recent years, due to the diverse sources, complex formats and rich content of information in various business areas of power marketing, traditional auditing techniques are single and cannot fully identify marketing business risks, nor can they effectively utilize the large amount of business knowledge contained in unstructured marketing data. In the marketing auditing business, it is necessary to formulate corresponding auditing topics based on national policies, local documents and marketing business specifications, such as "multiple households at one address" and "temporary electricity use expires but the account is not cancelled". Auditing business personnel interpret various policy documents such as business expansion registration, electricity prices, metering collection, customer service, etc., and combine the current auditing management methods to compile texts that can assist professionals in auditing and controlling. These texts have been repeatedly verified, optimized and improved, and finally formed a set of rules that can identify specific professional problems. Due to the complete reliance on manual compilation, these rules have certain limitations in terms of coverage and direction, resulting in low utilization of auditing information and inaccurate problem clues, which cannot meet the needs of power companies and regulatory authorities. Therefore, it is urgent to introduce new technologies and methods to improve the auditing technology system and improve the quality and efficiency of auditing work.

[0004] Knowledge graph is a knowledge representation and management method based on semantic network. It can effectively organize, store and query large-scale, multi-source and heterogeneous knowledge data, providing strong support for deep knowledge mining and intelligent analysis.

[0005] Explanation of relevant terms:

[0006] Business object: a semantic object used in the enterprise domain to describe business concepts. These objects are independent of computing facilities and technical implementation details. Business activities can be described by a series of mutually cooperating business objects, and business objects are not necessarily computing objects. For example, customers, products, orders, bank accounts, etc. are all concepts in the business domain and can all be represented by business objects. Business objects express business concepts from a high-level point of view, thereby more completely reflecting the essence of the business. Compared with the traditional "object", a business object is a special object that focuses on expressing business concepts and is mainly applied to enterprise computing environments. Objects in user interfaces, database access objects, etc. are objects, but they do not belong to business objects. Business process objects are a type of business object. Business process objects control the execution of business processes by sending control information to other business objects. Complex business processes can be completed by multiple business process objects working together.

[0007] Utility objects: SAP utility objects define business master data and technical master data, covering various business processes such as marketing business expansion, metering, billing and accounting. By abstracting the relevant participants in marketing business activities, a unified construction and multiple reuse method for marketing business objects is established. The advantage of this construction is more efficient and lightweight management objects.

[0008] Entity extraction: The BERT model is used to perform entity recognition on labeled structured, unstructured, and semi-structured data and form corresponding triple data.

[0009] Semantic identification and integration method: refers to the process of unifying, standardizing and disambiguating entities and relationships from different sources, forms and contexts during the construction of the knowledge graph. Summary of the invention

[0010] This invention proposes a new knowledge graph construction method for marketing audit topics, which is used for business scenario identification, audit rule verification, and work order reply review in marketing audit topics. The specific technical solution is as follows:

[0011] A method for constructing a knowledge graph based on the subject of power marketing audit includes the following processes:

[0012] S100: Based on the marketing audit business, with SAP business master data and technical master data as the model guidance, the data collected from various sources are classified and sorted, and the data is cleaned to form a standardized data format, and finally the marketing audit model layer is constructed, including entities and relationships;

[0013] S200: Based on the ontology recognition policy defined in the pattern layer, data with specific meanings in the document, such as customers, stations, and regions, recognizes the semantic relationship between entities from the text; uses deep learning models such as BERT, Bi-LSTM combined with CRF to jointly extract entities and relationships;

[0014] S400: Store the corresponding graph data for the entities that have been integrated in the knowledge graph, and organize the knowledge in the form of triples (entity-relationship-entity) to facilitate the management and application of knowledge.

[0015] Preferably, S400 is followed by the following process:

[0016] S500: Further optimize the relationships between entities in the formed knowledge graph, and through knowledge reasoning, consistency checking and error detection, perform secondary fusion of potential relationships in entity relationships and reduce data conflicts or inconsistencies.

[0017] Preferably, the following process is also included between S200 and S400:

[0018] S300: In response to the overlapping entity relationships and inaccurate entity identification in S200, entity alignment and entity disambiguation operations are performed.

[0019] Preferably, in S100, the entities are as shown in Table 1, and the relationships are as shown in Table 2.

[0020] Table 1

[0021]

[0022] Table 2

[0023]

[0024] The biggest difference between the marketing audit construction method and the traditional knowledge graph construction is the definition of the model layer. By referring to the SAP utility model and combining the characteristics of the marketing business itself, the marketing business entities and relationship objects are formed by source attachment, separation, purification, and simplification. This process not only removes duplication and disambiguation of multi-source data of the marketing business, but also reduces the difficulty of identifying disordered content in semantic recognition. The present invention has high practical value in dealing with problems such as business relationship query and marketing audit problem clue identification in the marketing audit theme.

[0025] Compared with traditional marketing audit methods, the present invention, on the one hand, conducts business combing and entity extraction on structured, unstructured and semi-structured data; on the other hand, it uses semantic recognition and penetration methods to fuse entities for entity ambiguity and inaccurate semantic coherence generated during the application process, thereby achieving consistency and uniqueness of business entities, and finally forming a semantic structure that conforms to the knowledge graph of marketing audit business applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of the method for constructing a knowledge graph of the present invention. DETAILED DESCRIPTION

[0027] A method for constructing a knowledge graph based on the subject of power marketing audit includes the following processes:

[0028] S100: Based on the marketing audit business, with SAP business master data and technical master data as the model guidance, data is collected from multiple sources, such as structured data, semi-structured data, and unstructured data. The data needs to be classified and sorted, and the data needs to be cleaned to form a standardized data format, and then finally a marketing audit model layer is constructed, including entities and relationships; the entities are shown in Table 1, and the relationships are shown in Table 2;

[0029] S200: Based on the ontology recognition policy defined in the pattern layer, data with specific meanings in the document, such as customers, stations, and regions, identifies semantic relationships between entities from the text; uses deep learning models such as BERT, Bi-LSTM and CRF to jointly extract entities and relationships;

[0030] S300: performing entity alignment and entity disambiguation operations for the overlapping entity relationships and inaccurate entity recognition in S200;

[0031] S400: Store the entities integrated in the knowledge graph in corresponding graph data, such as Neo4j; use semantic web technologies such as RDF to store knowledge in triples (entity-relationship-entity) to facilitate knowledge management and application;

[0032] S500: After the marketing audit knowledge graph is constructed, it still needs to be iteratively updated. Knowledge reasoning is an important step, in which new relationships are derived from existing entities and relationships. For example, in the knowledge graph, if "the owner of electricity meter 1 is A" and "the owner of electricity meter 2 is B, but A and B have the same ID number, then A and B are the same user. On the other hand, data sometimes conflict or are inconsistent. Knowledge reasoning can be used to check these inconsistencies. If it is found that the same property is both newly installed and transferred, the reasoning engine will detect this error and mark or correct it, thereby completing the update of the marketing audit knowledge graph.

[0033] Table 1

[0034]

[0035] Table 2

[0036]

Claims

1. A method for constructing a knowledge graph based on the subject of power marketing audit, characterized in that: The process includes the following: S100: Based on the marketing audit business, with SAP business master data and technical master data as the model guidance, the data collected from various sources are classified and sorted, and the data is cleaned to form a standardized data format, and finally the marketing audit model layer is constructed, including entities and relationships; S200: Based on the ontology recognition policy defined in the pattern layer, data with specific meanings in the document, identify the semantic relationship between entities from the text; use deep learning models to jointly extract entities and relationships; S400: Store the corresponding graph data for the entities that have been integrated in the knowledge graph, and organize the knowledge in the form of triples (entity-relationship-entity) to facilitate the management and application of knowledge.

2. According to claim 1, a method for constructing a knowledge graph based on the subject of power marketing audit, characterized in that: The following process follows after S400: S500: Further optimize the relationships between entities in the formed knowledge graph, and through knowledge reasoning, consistency checking and error detection, perform secondary fusion of potential relationships in entity relationships and reduce data conflicts or inconsistencies.

3. According to claim 1, a method for constructing a knowledge graph based on the subject of power marketing audit, characterized in that: The following process is also included between S200 and S400: S300: In response to the overlapping entity relationships and inaccurate entity identification in S200, entity alignment and entity disambiguation operations are performed.

4. According to claim 1, a method for constructing a knowledge graph based on the subject of power marketing audit, characterized in that: In S100, the entities are shown in Table 1, and the relationships are shown in Table 2. Table 1 Table 2

Images