An industrial park knowledge graph construction method and system fusing multi-source heterogeneous data

Abstract

The application provides a kind of industrial park knowledge graph construction method and system fusing multi-source heterogeneous data, pass;Different sources about digital economy knowledge and digital economy industrial park data are integrated into the structured knowledge with digital economy industrial park as the center by data cleaning;According to the labeling strategy, the entity relationship in the structured knowledge is labeled;Based on convolutional neural network, a digital economy industrial park knowledge fusion model is constructed and trained, which is used for preliminary extraction of entity relationship in structured knowledge;Repeated entity relationship triples are retrieved and entity relationship extraction is performed, so as to construct entity-relation-entity triples of digital economy industrial park knowledge graph, and knowledge fusion is carried out.The application realizes the rapid construction of large-scale digital economy industrial park knowledge graph, which helps to improve the information analysis and processing efficiency, facilitates to find the hidden influence and transmission effect between digital economy industrial park elements, and forms the scale industrial analysis results.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital economic industry knowledge analysis, and particularly relates to a method and system for constructing an industry park knowledge graph by fusing multi-source heterogeneous data. BACKGROUND

[0002] A digital economic industry park refers to a smart park that takes cultivating a digital economic industry as the main guide, mainly uses cloud computing, big data, block chain, artificial intelligence, the Internet of Things, 5G and other technologies, and realizes industrial ecosystem, infrastructure networking, function service precision and operation development intelligence. Through effective processing and deep mining of multi-source heterogeneous data of the digital economic industry park, a new and intelligent digital economic industry park knowledge graph is constructed, which has great significance for future industry investment, development, regulation, standardization, scheduling, management and decision-making. For example, in the past, most of the research work on industrial parks was based on the understanding and cognition of experts on industrial parks, and an analysis model was established through limited enterprise samples and manual research and carding of park-related businesses. Due to the limitations of expert cognition and the difficulty of research sampling, it is difficult to form a large-scale industrial analysis result, and most of the research results exist in the form of fragmented knowledge.

[0003] Therefore, there is currently no universally applicable method to solve the problem of finding the hidden influence and transmission effect between the elements of the digital economic industry park. SUMMARY

[0004] Therefore, the present application proposes a method for constructing an industry park knowledge graph by fusing multi-source heterogeneous data, which is used to solve the problem of finding the hidden influence and transmission effect between the elements of the digital economic industry park.

[0005] The technical solution of the present application is as follows:

[0006] In a first aspect, the present application discloses a method for constructing an industry park knowledge graph by fusing multi-source heterogeneous data, which comprises the following steps:

[0007] S1, collecting digital economic knowledge and digital economic industry park data from different sources;

[0008] S2, integrating the knowledge and data into structured knowledge centered on the digital economic industry park through data cleaning;

[0009] S3, labeling the entity relationship in the structured knowledge according to a labeling strategy;

[0010] S4, constructing a digital economic industry park knowledge fusion model based on a convolutional neural network and training the model;

[0011] S5, preliminary extraction of entity relationships in the structured knowledge through the digital economic industrial park knowledge fusion model;

[0012] S6, repeated entity relationship triples are searched and entity relationship extraction is performed, so as to construct entity-relation-entity triples of the digital economic industrial park knowledge graph and perform knowledge fusion.

[0013] The present application establishes the mutual connection between the related economic factors of the digital economic industrial park by the above method, then quantifies the relationship by big data, finally finds the implicit influence and transmission effect between the factors of the digital economic industrial park by machine learning, and finally sorts out a complete logical relationship chain of the digital economic industrial park knowledge graph.

[0014] On the basis of the above technical scheme, preferably, step S1 specifically comprises:

[0015] The digital economic knowledge includes digital industrialization knowledge and industrial digitization knowledge, the digital industrialization includes digital product manufacturing industry, digital product service industry, digital technology application industry and digital factor driven industry, and the industrial digitization includes digital efficiency improvement industry.

[0016] The digital economic industrial park data includes: digital economic industrial park policy, enterprise upstream and downstream relationship, industry classification and directory, park development and present situation, main enterprises in the park, main products, main equipment, investment and financing, intellectual property, talent distribution, technical transformation, output value and profit.

[0017] The collection method includes: using a crawler framework to crawl semi-structured open data of an industrial park website; summarizing and sorting data of the digital economic industrial park including related authoritative policies, industrial reports, industry analysis, related materials and books; correcting and deleting the data of the digital economic industrial park by experts to improve the professional level of data knowledge.

[0018] The present application collects a large amount of information in multiple ways by considering the professional nature of the digital economic industrial park database.

[0019] On the basis of the above technical scheme, preferably, step S2 specifically comprises:

[0020] The knowledge and data are preprocessed by information desensitization and text segmentation, and at the same time, the knowledge and data are modified, supplemented and deleted by experts to integrate the knowledge and data into structured knowledge centered on the digital economic industrial park to obtain entity information in the knowledge and data.

[0021] The present application cleanses the data to obtain accurate, clear and concise data by the above method.

[0022] On the basis of the above technical solutions, preferably, step S3 specifically comprises:

[0023] determining the entity relationship in the structured knowledge, for the records x and y of two entities, the values of x and y on the i-th attribute are x i ,y i , and the record connection of x and y is performed through the following steps:

[0024] ③Integrating the single attribute similarity to obtain an attribute similarity vector

[0025] [sim(x1,y1),sim(x2,y2),…sim(x n ,y n ),];

[0026] ④Obtaining the similarity between entities according to the attribute similarity vector;

[0027] adding a label to the entity relationship according to the similarity between entities, the label being composed of the position in the entity, the relationship type, and the order of the entities;

[0028] The calculation method of the similarity between entities according to the attribute similarity vector includes but is not limited to the edit distance, the set similarity calculation, the vector-based similarity calculation, and the synonym-based word similarity calculation.

[0029] On the basis of the above technical solutions, preferably, step S4 specifically comprises:

[0030] building a digital economic industrial park knowledge fusion model comprising a convolution layer, a pooling layer, a hidden layer, and a conversion layer, and training the same; the digital economic industrial park knowledge fusion model uses a similarity calculation method based on a convolution neural network, generates a word vector as input after dividing the search sentence, convolves the word vector of the sentence through a convolution window with different sizes and different convolution kernels in the convolution layer to obtain convolution values respectively, filters out the maximum value through a pooling layer and retains it, calculates the output coefficient of the hidden layer, and finally obtains the output of the conversion layer by calculating the output coefficient of the hidden layer and the clustering features of the sentence, i.e., the similarity value of the search sentence and the searched sentence; the convolution features and clustering features of the search sentence are mapped through the model to calculate the similar feature description sentence, so as to complete the knowledge fusion.

[0031] On the basis of the above technical solutions, preferably, step S5 specifically comprises:

[0032] preliminarily extracting the entity relationship in the structured knowledge through the trained digital economic industrial park knowledge fusion model; and simultaneously collecting corresponding information through online crawling to supplement the data:

[0033] First, the url rule of the target page is analyzed, and second, the content directly crawled is processed, all punctuation and conjunctions are replaced with "*", and then the regular expression is cut and added to the data.

[0034] The present application constructs a digital economic industrial park knowledge fusion model for knowledge fusion by the above method, extracts entity relationships, and simultaneously performs data completion to avoid data loss and improve information.

[0035] On the basis of the above technical scheme, preferably, step S6 specifically comprises:

[0036] S6-1, generating a word vector after dividing all the feature description sentences of the industrial park into words;

[0037] S6-2, clustering the word vector by using a DBSCAN algorithm to obtain a word center point as a clustering feature;

[0038] S6-3, generating a clustering feature of each sentence according to the word clustering feature;

[0039] S6-4, inputting a word vector matrix of a search sentence;

[0040] S6-5, convoluting the hidden layer coefficient of the search sentence in the digital economic industrial park knowledge fusion model with the clustering feature of the searched sentence to obtain a sentence similarity;

[0041] S6-6, if the sentence similarity is greater than a preset threshold, starting deduplication and fusion, adding the attribute of str2 to the attribute of str1, removing str2 and keeping str1; if the sentence similarity is less than the preset threshold, keeping both data.

[0042] The present application completes a comprehensive and structured knowledge system of "digital economic industrial park enterprise cognitive graph-digital economic industrial park industry knowledge graph-digital economic industrial park knowledge graph" by the above method, improves the information analysis and processing efficiency, and facilitates the formation of large-scale industrial analysis results.

[0043] In the second aspect of the present application, a system for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data is disclosed, which comprises:

[0044] A data acquisition module is used to acquire digital economic knowledge and digital economic industrial park data from different sources;

[0045] A data preprocessing module is used to integrate the knowledge and data into structured knowledge centered on the digital economic industrial park through data cleaning;

[0046] An entity labeling module is used to label the entity relationships in the structured knowledge according to a labeling strategy;

[0047] A fusion model module is configured to construct a digital economic industrial park knowledge fusion model based on a convolutional neural network and train the model; and the digital economic industrial park knowledge fusion model is used to preliminarily extract entity relationships in the structured knowledge.

[0048] A data fusion module is configured to search for repeated entity relationship triples and extract entity relationships, thereby constructing entity-relation-entity triples of a digital economic industrial park knowledge graph and performing knowledge fusion.

[0049] In a third aspect, the present application discloses an electronic device, which comprises at least one processor, at least one memory, a communication interface and a bus; wherein the processor, the memory and the communication interface can communicate with each other through the bus; the memory stores a program of a method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data, which can be executed by the processor; and the program of the method for constructing the industrial park knowledge graph by fusing multi-source heterogeneous data is configured to implement the method for constructing the industrial park knowledge graph by fusing multi-source heterogeneous data according to the first aspect of the present application.

[0050] In a fourth aspect, the present application discloses a computer readable storage medium, which stores a program of a method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data; and the program of the method for constructing the industrial park knowledge graph by fusing multi-source heterogeneous data is configured to implement the method for constructing the industrial park knowledge graph by fusing multi-source heterogeneous data according to the first aspect of the present application when executed.

[0051] The method for constructing the industrial park knowledge graph by fusing multi-source heterogeneous data has the following beneficial effects compared with the prior art:

[0052] (1) The method takes the digital economic industrial park as the core, establishes the mutual relationship among the related economic elements of the digital economic industrial park, quantifies the relationship by using big data, and finally finds the hidden influence and transmission effect among the elements of the digital economic industrial park by using machine learning, so as to finally sort out a complete logical relationship chain of the digital economic industrial park knowledge graph;

[0053] (2) By applying NLP, knowledge graph and neural network processing methods to a large amount of comprehensive and structured knowledge, a large-scale digital economic industrial park knowledge graph is rapidly constructed from bottom to top, which helps to improve the information analysis and processing efficiency, facilitates the finding of the hidden influence and transmission effect among the elements of the digital economic industrial park, and forms a large-scale industrial analysis result. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description only represent some embodiments of the present application, and other accompanying drawings can be obtained by those skilled in the art without any creative effort on the basis of these accompanying drawings.

[0055] Figure 1 The workflow diagram of the method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data according to the present application is shown in DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the scope of protection of the present application.

[0057] EMBODIMENT

[0058] The workflow of the method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data according to the present application is shown in Figure 1 The processing steps are described as follows.

[0059] In the first step, data about digital economic knowledge and digital economic industrial park from different sources are collected.

[0060] It should be understood that, on the basis of the above solutions, the digital economic knowledge includes digital industrialization knowledge and industrial digitization knowledge, the digital industrialization includes digital product manufacturing industry, digital product service industry, digital technology application industry and digital factor driven industry, and the industrial digitization includes digital efficiency improvement industry.

[0061] Among them, digital product manufacturing industry, digital product service industry, digital technology application industry and digital factor driven industry are the core industries of digital economy, which refers to the provision of digital technology, products, services, infrastructure and solutions for industrial digitization development, as well as various economic activities that rely entirely on digital technology and data elements. It corresponds to 26 major categories, 68 medium categories and 126 small categories in the National Economic Industry Classification, and is the basis for the development of digital economy. Core industries mainly include computer communication and other electronic equipment manufacturing industry, telecommunications broadcasting television and satellite transmission services, Internet and related services, software and information technology services, etc. Industrial digitization covers smart agriculture, intelligent manufacturing, intelligent transportation, smart logistics, digital finance, digital commerce, digital society, digital government and other digital application scenarios, corresponding to 91 major categories, 431 medium categories and 1256 small categories in the National Economic Industry Classification. And digital economy industrial park policy, enterprise upstream and downstream relationship, industry classification and directory, development and status of the park, main enterprises in the park, main products, main equipment, investment and financing, intellectual property, talent distribution, technological transformation, output value and profit, etc.

[0062] The digital economy industrial park data includes: digital economy industrial park policy, enterprise upstream and downstream relationship, industry classification and directory, development and status of the park, main enterprises in the park, main products, main equipment, investment and financing, intellectual property, talent distribution, technological transformation, output value and profit, etc.

[0063] Due to the professional nature of the economic industrial park database, multiple methods are used to collect and acquire knowledge and data. The collection methods include: using a crawler framework to crawl semi-structured public data from industrial park websites; summarizing and organizing data related to digital economy industrial parks, including relevant policies, industry reports, industry analysis, related materials, and books; correcting and deleting data of digital economy industrial parks by experts to improve the professional level of data knowledge.

[0064] In the second step, the knowledge and data are integrated into structured knowledge centered on digital economy industrial parks through data cleaning; and in the third step.

[0065] It should be understood that, on the basis of the above scheme, the knowledge and data are subjected to information desensitization and text segmentation preprocessing, and at the same time, authoritative books and experts are used to correct, supplement and delete redundant information, to integrate structured knowledge centered on digital economy industrial parks, and to obtain entity information in the knowledge and data.

[0066] For example, after processing the digital business text document, the focus is on "digital wholesale, digital retail, digital accommodation, digital catering, digital leasing, digital business services", etc. Smart agriculture focuses on "digital facility planting, digital forestry, automated breeding, new technology breeding, and other smart agriculture".

[0067] In the third step, the entity relationship in the structured knowledge is labeled according to the labeling strategy; go to the fourth step.

[0068] It should be understood that, on the basis of the above scheme, the entity relationship in the structured knowledge is determined, and for the records x and y of two entities, the values of x and y on the i-th attribute are x i ,y i , x and y are recorded and connected by the following steps:

[0069] (5) Get the attribute similarity vector by integrating the similarity of individual attributes

[0070] [sim(x1,y1),sim(x2,y2),…sim(x n ,y n ),];

[0071] (6) Get the similarity between entities according to the attribute similarity vector;

[0072] According to the similarity between entities, the entity relationship is labeled, and the label is composed of the position of the entity, the relationship type and the order of the entity;

[0073] The calculation method of the similarity between entities according to the attribute similarity vector includes but is not limited to edit distance, set similarity calculation, vector-based similarity calculation and synonym-based word similarity calculation.

[0074] In the processing process of this step, the part of the structured knowledge that does not belong to the entity relationship is "ignored", and only the content belonging to the entity relationship component is processed. In the graph database, the relationship between entities and other entities is represented by edges. The more edges, the more complex the relationship between the entity and the network.

[0075] In the fourth step, a digital economy industrial park knowledge fusion model is constructed based on a convolutional neural network and trained; go to the fifth step.

[0076] It should be understood that, on the basis of the above scheme, a digital economy industrial park knowledge fusion model including a convolutional layer, a pooling layer, a hidden layer and a conversion layer is constructed, and the model is trained;

[0077] The training method of the digital economy industrial park knowledge fusion model is to first perform forward propagation training, and then perform back propagation verification training;

[0078] The digital economy industrial park knowledge fusion model uses a similarity calculation method based on a convolutional neural network. After dividing the search sentence into words, a word vector is generated as input. In the convolution layer, different sizes and different convolution kernels are used to convolve the word vector of the sentence to obtain convolution values. The maximum value is selected and retained through a pooling layer. The output coefficient of the hidden layer is calculated. The output of the conversion layer is finally obtained by calculating the output coefficient of the hidden layer and the sentence clustering feature, which is the similarity value of the search sentence and the searched sentence. The convolution features and clustering features of the search sentence are mapped through the model to calculate the similar feature description sentence.

[0079] In the fifth step, the entity relationship in the structured knowledge is preliminarily extracted through the digital economy industrial park knowledge fusion model; and the process proceeds to the sixth step.

[0080] It should be understood that, on the basis of the above scheme, the entity relationship in the structured knowledge is preliminarily extracted through the digital economy industrial park knowledge fusion model that has completed training;

[0081] Some of the preliminarily extracted entity relationships are not complete, only the name of the industry, but no corresponding enterprise (or product, technology, etc.), because of the error in the joint extraction of entity relationships, the corresponding enterprise (or product, technology, etc.) of the industry is not extracted, resulting in data loss, so it is necessary to collect the corresponding information through online crawling to supplement the data:

[0082] First, the url rule of the target page is analyzed, and then the directly crawled content is processed, all punctuation and conjunctions are replaced with “*”, and then added to the data after cutting with a regular expression.

[0083] In the sixth step, the digital economy industrial park knowledge fusion model is used to search for repeated entity relationship triples and extract entity relationships, so as to construct the entity-relation-entity triples of the digital economy industrial park knowledge graph and perform knowledge fusion.

[0084] It should be understood that, on the basis of the above scheme, the sixth step includes:

[0085] S6-1, generating a word vector after dividing the feature description sentence of all industrial parks into words;

[0086] S6-2, clustering the word vectors using the DBSCAN algorithm to obtain word center points as clustering features;

[0087] S6-3, generating the clustering features of each sentence according to the word clustering features;

[0088] S6-4, inputting the word vector matrix of the search sentence;

[0089] S6-5, the search statement is calculated in the hidden layer coefficient of the digital economic industrial park knowledge fusion model and the clustering characteristics of the searched statement to obtain a statement similarity;

[0090] S6-6, if the statement similarity is greater than a preset threshold, the de-duplication fusion is started, the attributes of str2 are added to the attributes of str1, str2 is removed and str1 is reserved; if the statement similarity is less than the preset threshold, both data are reserved.

[0091] It should be understood that, on the basis of the above scheme, the knowledge fusion is further performed on the data, and the construction of the comprehensive and structured knowledge system of the "digital economic industrial park enterprise cognitive graph-digital economic industrial park industry knowledge graph-digital economic industrial park knowledge graph" is completed.

[0092] The industrial park knowledge graph construction method fusing multi-source heterogeneous data takes the digital economic industrial park as the core, establishes the mutual connection between the related economic elements of the digital economic industrial park, quantifies the relationship by using big data, finally finds the implicit influence and transmission effect between the elements of the digital economic industrial park by using machine learning, and finally combs out a complete digital economic industrial park knowledge graph logical relationship chain. Instead of the previous single-point data structure taking the investment enterprise as the core, a comprehensive and structured knowledge system of the "digital economic industrial park enterprise cognitive graph-digital economic industrial park industry knowledge graph-digital economic industrial park knowledge graph" for the digital economic industrial park is built by fusing multi-source heterogeneous data and knowledge of the digital economic industrial park, which is helpful for the government, financial institutions, group enterprises, commercial institutions and research institutions to make industrial layout and decision, data risk control, procurement optimization, precise marketing and research analysis.

[0093] The application further discloses an industrial park knowledge graph construction system fusing multi-source heterogeneous data.

[0094] The data acquisition module is used for acquiring digital economic knowledge and digital economic industrial park data from different sources.

[0095] The data preprocessing module is used for integrating the knowledge and data into structured knowledge with the digital economic industrial park as the center by data cleaning.

[0096] The entity labeling module is used for labeling the entity relationship in the structured knowledge according to a labeling strategy.

[0097] The fusion model module is used for constructing a digital economic industrial park knowledge fusion model based on a convolutional neural network and training the model, and the entity relationship in the structured knowledge is preliminarily extracted by using the digital economic industrial park knowledge fusion model.

[0098] Data fusion module: used for retrieving repeated entity relationship triplets and performing entity relationship extraction, thereby constructing entity-relation-entity triplets of the digital economic industrial park knowledge graph, and performing knowledge fusion.

[0099] The application further discloses an electronic device, which comprises at least one processor, at least one memory, a communication interface and a bus; wherein the processor, the memory and the communication interface complete communication with each other through the bus; the memory stores a multi-source heterogeneous data fusion industrial park knowledge graph construction method program which can be executed by the processor, and the multi-source heterogeneous data fusion industrial park knowledge graph construction method program is configured to implement the multi-source heterogeneous data fusion industrial park knowledge graph construction method as described in the embodiments of the application.

[0100] The application further discloses a computer readable storage medium, which stores a multi-source heterogeneous data fusion industrial park knowledge graph construction method program; when the multi-source heterogeneous data fusion industrial park knowledge graph construction method program is executed, the multi-source heterogeneous data fusion industrial park knowledge graph construction method as described in the embodiments of the application is implemented.

[0101] The above merely describes preferred embodiments of the application and is not intended to limit the application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data, characterized in that, The method comprises the following steps: S1, collecting digital economic knowledge and digital economic industrial park data from different sources; S2, integrating the knowledge and data into structured knowledge centered on digital economic industrial parks through data cleaning; S3, labeling the entity relationships in the structured knowledge according to the labeling strategy; S4, constructing a digital economic industrial park knowledge fusion model including convolution layers, pooling layers, hidden layers, and conversion layers, and training the model; the digital economic industrial park knowledge fusion model uses a similarity calculation method based on a convolutional neural network, generates word vectors after dividing the search sentence, and inputs the word vectors into the convolution layer to obtain convolution values through different sizes and different convolution kernels of the convolution window; the maximum value is selected through a pooling layer and is reserved; the output coefficients of the hidden layer are calculated, and the output of the conversion layer is finally obtained by calculating the output coefficients of the hidden layer and the sentence clustering features, which is the similarity value of the search sentence and the searched sentence; the convolution features and clustering features of the search sentence are mapped through the model to calculate the similar feature description sentence; S5, preliminarily extracting the entity relationships in the structured knowledge through the digital economic industrial park knowledge fusion model; S6, retrieving repeated entity relationship triples and extracting entity relationships to construct entity-relation-entity triples of the digital economic industrial park knowledge graph and perform knowledge fusion: S6-1, generating word vectors after dividing the feature description sentences of all industrial parks; S6-2, clustering the word vectors using the DBSCAN algorithm to obtain word center points as clustering features; S6-3, generating clustering features for each sentence according to the word clustering features; S6-4, inputting the word vector matrix of the search sentence; S6-5, calculating the similarity of the search sentence in the hidden layer of the digital economic industrial park knowledge fusion model and the clustering features of the searched sentence; S6-6, if the sentence similarity is greater than the preset threshold, start deduplication and fusion, add the attributes of str2 to the attributes of str1, remove str2, and keep str1; if the sentence similarity is less than the preset threshold, both data are kept. 2.The industrial park knowledge graph construction method of fusing multi-source heterogeneous data according to claim 1, wherein, The step S1 specifically comprises: The digital economic knowledge includes digital industrialization knowledge and industrial digitization knowledge, the digital industrialization includes digital product manufacturing industry, digital product service industry, digital technology application industry, and digital factor driven industry, and the industrial digitization includes digital efficiency improvement industry; The digital economic industrial park data includes digital economic industrial park policies, enterprise upstream and downstream relationships, industry classification and directory, park development and status, main enterprises in the park, main products, main equipment, investment and financing, intellectual property, talent distribution, technological transformation, and output value and profit; The collection mode comprises: using a crawler framework to crawl semi-structured open data of an industrial park website; summarizing and sorting data of a digital economy industrial park, including relevant authoritative policies, industrial reports, industry analysis, relevant materials and books; correcting and adding or deleting data of the digital economy industrial park by experts to improve the professional level of data knowledge. 3.The industrial park knowledge graph construction method of fusing multi-source heterogeneous data according to claim 1, wherein, The step S2 specifically comprises: The knowledge and data are preprocessed by information desensitization and text segmentation, and are supplemented, corrected and deleted by experts to obtain entity information in the knowledge and data. 4.The method of claim 3, wherein, The step S3 specifically comprises: determining entity relationships in the structured knowledge, for two entity records x and y, the values of x and y on the i-th attribute are x i ,y i , and y, x and y are record-linked by the steps of: ① obtaining an attribute similarity vector by comprehensively combining single attribute similarities [sim(x1,y1),sim(x2,y2),…sim(x n ,y n ),] ② obtaining similarity between entities according to the attribute similarity vector; adding labels to the entity relationship according to the similarity between entities, wherein the labels are composed of positions in the entities, relationship types and orders of the entities; The calculation method of the similarity between entities according to the attribute similarity vector comprises but is not limited to an edit distance, a set similarity calculation, a vector-based similarity calculation and a synonym-based word similarity calculation. 5.The method of claim 1, wherein, The step S5 specifically comprises: preliminarily extracting entity relationships in the structured knowledge by using a trained digital economy industrial park knowledge fusion model, and supplementing data by collecting corresponding information through online crawling: firstly, analyzing url rules of a target page, and secondly, processing directly crawled content, replacing all punctuation and conjunctions with "*", and then cutting and adding to data by using a regular expression.

6. An industrial park knowledge graph construction system fusing multi-source heterogeneous data, characterized in that, The system comprises: a data collection module for collecting digital economy knowledge and digital economy industrial park data from different sources; a data preprocessing module for integrating the knowledge and data into structured knowledge with the digital economy industrial park as the center by data cleaning; an entity labeling module for labeling entity relationships in the structured knowledge according to a labeling strategy; a fusion model module for constructing a digital economy industrial park knowledge fusion model comprising a convolution layer, a pooling layer, a hidden layer and a conversion layer, and training the model; the digital economy industrial park knowledge fusion model uses a convolutional neural network-based similarity calculation method, generates a word vector as input after dividing a search sentence, convolves the word vector of the sentence through different sizes and different convolution kernels of a convolution window in the convolution layer to obtain convolution values respectively, filters out a maximum value through a pooling layer and retains the maximum value, calculates an output coefficient of the hidden layer, and finally obtains an output of the conversion layer by calculating the output coefficient of the hidden layer and a sentence clustering feature, which is a similarity value of the search sentence and a searched sentence; maps the convolution feature and the clustering feature of the search sentence through the model to calculate a similar feature description sentence; and preliminarily extracts entity relationships in the structured knowledge by using the digital economy industrial park knowledge fusion model. The data fusion module is used for retrieving repeated entity relationship triples and performing entity relationship extraction, so as to construct entity-relation-entity triples of the digital economic industrial park knowledge graph and perform knowledge fusion. S6-1, generating a word vector after dividing all feature description sentences of the industrial park into words; S6-2, clustering the word vector by using a DBSCAN algorithm to obtain a word center point as a clustering feature; S6-3, generating a clustering feature of each sentence according to the word clustering feature; S6-4, inputting a word vector matrix of a search sentence; S6-5, calculating a sentence similarity by convoluting a hidden layer coefficient of the search sentence in the digital economic industrial park knowledge fusion model and a clustering feature of a searched sentence; S6-6, if the sentence similarity is greater than a preset threshold, starting deduplication and fusion, adding an attribute of str2 to an attribute of str1, removing str2 and keeping str1; if the sentence similarity is less than the preset threshold, keeping both data.

7. An electronic device, comprising: The device comprises at least one processor, at least one memory, a communication interface and a bus; wherein the processor, the memory and the communication interface complete mutual communication through the bus; the memory stores a program of a method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data, which is configured to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program of a method for constructing an industrial park knowledge graph by fusing multi-source heterogeneous data, which is configured to implement the method according to any one of claims 1 to 5 when executed.

Images