CN110362660B - Electronic product quality automatic detection method based on knowledge graph - Google Patents
Electronic product quality automatic detection method based on knowledge graph Download PDFInfo
- Publication number
- CN110362660B CN110362660B CN201910667154.9A CN201910667154A CN110362660B CN 110362660 B CN110362660 B CN 110362660B CN 201910667154 A CN201910667154 A CN 201910667154A CN 110362660 B CN110362660 B CN 110362660B
- Authority
- CN
- China
- Prior art keywords
- product
- data
- detection
- entity
- knowledge graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention relates to an electronic product quality automatic detection method based on a knowledge graph, which belongs to the field of electronic product quality detection and knowledge graph, and comprises the following steps: the construction field knowledge graph and the quality detection method are realized. Constructing a knowledge graph of the product field through standard documents and technical specifications related to electronic products; the method comprises the steps of detecting the quality of an electronic product from four aspects, managing a mode layer of a knowledge graph by adopting a body modeling method, constructing a data layer on the basis of the mode layer, extracting entity and entity relations by adopting a Bi-LSTM+attribute+CRF method from a plurality of heterogeneous data sources, generally adopting a similarity calculation method for data fusion, storing knowledge in a neo4j graph database by using facts as a unit by the data layer, detecting the quality of the product by utilizing a knowledge graph in the product standard field, and adding a front-end module for directly inquiring quality detection data.
Description
Technical Field
The invention belongs to the field of electronic product quality detection and knowledge graph, and relates to an electronic product quality automatic detection method based on the knowledge graph.
Background
The electronic products become an indispensable part in life, the quality of the electronic products is the primary requirement of people, the quality detection of the electronic products is mostly carried out by adopting a manual detection method, the detection personnel sample and detect the products, and supervise the personnel to carry out self-detection, mutual detection and first detection on the production of the products, the manual detection can prolong the development period, the development efficiency is reduced, and a plurality of defects appear. In addition, a large amount of data can be generated in the detection process, but all detection systems are mutually independent, so that a large amount of data redundancy can be generated, the data structures are inconsistent, the association between data knowledge cannot be established, and the overall management of the quality detection systems is not facilitated. Therefore, it is necessary to design an automatic detection method for quality of electronic products capable of realizing data association to solve these problems.
The knowledge graph is a method for describing each entity, relation and theory in the real world in a structured mode, can describe the relation among the entities by using a visual technology, is a new generation semantic net realization from the steps of constructing, analyzing, displaying and mining the association among data, and provides possibility for solving the problem of data redundancy by using the knowledge graph. For fault detection of electronic products, the knowledge graph is applied to a very small number, so that the scheme innovatively utilizes the knowledge graph scheme to construct a product field graph, heterogeneous data fusion and data sharing are realized, and product detection data are analyzed to further realize product quality detection.
Disclosure of Invention
Therefore, the invention aims to provide an automatic detection method for the quality of electronic products based on a knowledge graph, which solves the problems of excessive manual participation, low detection efficiency, high requirements on the technical level of detection personnel and less data relevance among detection units in the quality detection process. In order to achieve the above purpose, the present invention provides the following technical solutions:
a quality detection method of electronic products based on a knowledge graph comprises the following steps:
s1: constructing a knowledge graph in the field of electronic products according to the relevant standard as a basis, wherein the graph is used as a basis for quality detection;
s2: monitoring the name, the detected attribute and the value range of the product to be detected;
s3: matching the product attribute to be detected with the attribute in the domain knowledge graph, determining a detection target, inquiring the corresponding attribute value, and determining the product quality according to the matching result of the attribute value;
s4: and adding a front-end module for directly inquiring the quality detection data.
Further, in step S1, the following steps are included:
aiming at the detection process of the electronic product, relevant standards are derived from national standards, industry standards, enterprise standards and product technical specifications, and particularly, the detection is carried out from four aspects of the product technical specifications, product appearance detection standards, product function detection standards and product reliability detection standards aiming at the quality detection process of the electronic product, so as to determine a mode layer of a knowledge graph;
performing entity identification on a data source by adopting a method of a circulating neural network Bi-LSTM+attribute+CRF, and constructing a data layer on the basis of a mode layer;
disambiguating the entity identification result by adopting a clustering-based entity disambiguation method to obtain disambiguated entity information;
the obtained data comprise corresponding entities, entity relations and attribute values in the field of electronic products, the data in the graph database are selected for storage, namely, the data are imported into the graph database neo4j to form the relation between nodes, and the knowledge graph of the whole product field is completed.
The attribute requirements for the field of electronic products are different, and the detection attribute for mobile phone electronic products is common in: appearance, function, reliability, etc., so the extracted entity relates to aspects of product name, color, pixel, memory, operating system, processor, memory, etc.;
and determining the triplet entity and the relation thereof, and obtaining the product name-attribute value triplet by using a Bi-LSTM+attribute+CRF method for the characteristics of mobile phone electronic products. The specific method comprises the following steps:
(1) Converting text vectors in the knowledge base into character vectors by using a word2vec method;
(2) Inputting the character vector into a Bi-LSTM network, obtaining past information through a forward LSTM, obtaining future information through a backward LSTM, and simultaneously containing context information through a Bi-LSTM model;
(3) An Attention mechanism is added, so that effective keywords are highlighted, and the recognition efficiency is improved;
(4) And inputting the trained vector into the CRF, and compensating the limitation of each tag sequence after Bi-LSTM training to increase the recognition rate.
Further, step S1 includes the steps of:
s11: detecting from four aspects of a product technical specification, a product appearance detection standard, a product function detection standard and a product reliability detection standard aiming at the quality detection process of the electronic product, and determining a mode layer of a knowledge graph;
s12: performing entity identification on a data source by adopting a method of a circulating neural network Bi-LSTM+attribute+CRF, and constructing a data layer on the basis of a mode layer;
s13: disambiguating the entity identification result by adopting a clustering-based entity disambiguation method to obtain disambiguated entity information;
s14: fusing data from different databases by using a frame matching method, namely a body alignment method, so as to obtain the complement of the knowledge graph;
s15: through the steps S11-S14, the obtained data comprise corresponding entities, entity relations and attribute values in the field of electronic products, the data in the graph database are selected for storage, namely, the data are imported into the graph database neo4j to form the relations among nodes, and the knowledge graph of the whole product field is completed.
Further, step S11 specifically includes the following steps:
s111: according to the current quality detection mode of the electronic product, detecting from four aspects of a product technical specification, a product appearance detection standard and a product function detection standard, and performing system modeling according to a product detection object to form a product detection body structure;
s112: the concept and the classification relation of the product detection object are defined, and the concept of the detection object is integrated and objectively formed into a whole;
s113: defining a knowledge description system, a category of each detection object and a possible relation between each detection object;
s114: ontology for product detection by knowledge representation.
Further, the step S12 specifically includes the steps of:
s121: the data of the data layer comprises an electronic product standard detection document and a technical specification of product production, belongs to unstructured data, and takes the unstructured data as input by taking a text;
s122: converting into corresponding character vector by word2vec method, and obtaining target word by CBOW model in which original sentence is presumed, i.e. by knowing current word w n The probability P of the existence of the current word is presumed by utilizing the front and back 2k words:
P(w n |SUM(w n-k ,w n-k+1 …,w n-1+k ,w n+k )) (1)
thereby calculating the probability of the sentence S using the CBOW model, the calculation expression is as follows:
t is the input text length and P (S) represents the sentence in (w 1 ,w 2 ,…,w T ) The probability of occurrence is the joint probability of each word; likelihood functions are then constructed for the entire text:
deriving a log-likelihood function:
the aim of CBOW model training is to make the objective likelihood function reach the maximum value;
s123: when the input character is Word2vec to obtain a corresponding character vector, the corresponding character vector is input into a Bi-LSTM network, the input character vector is subjected to forward LSTM to obtain past information, backward LSTM is subjected to backward LSTM to obtain future information, and the context information can be simultaneously contained through a Bi-LSTM model;
s124: in order to improve the recognition efficiency, focusing on important entities, adding an Attention mechanism;
s125: after passing through the Attention mechanism, the character vector is input into the CRF to play a role of a multi-classifier to make up for the limitation of each word label sequence after Bi-LSTM training;
for an input sequence z= (Z) 1 ,z 2 ,…,z n ) Where n is the input sequence length, z i If the i-th word of the input vector is represented, the output sequence label corresponding to Z is y= (Y) 1 ,y 2 ,…,y n ) The method comprises the steps of carrying out a first treatment on the surface of the For a given sequence Z, the conditional probability of labeling the sequence Y with a value of Y is p (y|z)
S(z)＝∑ y exp(∑ N,k λ k t k (y N-1 ,y N ,z,N)+∑ N,l μ l s l (y N ,z,N)) (6)
N＝1,2,…,n，s l (. Cndot.) and t k (. Cndot.) is a characteristic function, μ l And lambda (lambda) k Respectively corresponding weights, and S (z) is a normalization factor; in the training stage, the maximum likelihood estimation is used for solving the optimal sequence label, and the corresponding likelihood logarithm is: sigma (sigma) N The input sequence is marked by the highest conditional probability y obtained through training of the lovp (y|z):
further, step S13 specifically includes:
adopts the base ofMethod for disambiguating clustered entities, given the set o=o of entities to be disambiguated 1 ,O 2 ,…,O n Extracting the feature of each entity name O, expressing the feature vector as a feature vector, calculating the similarity among the entities, clustering the entity name items by adopting a clustering algorithm, enabling each category in a clustering result to correspond to a specific target entity, and calculating the similarity by adopting a cosine similarity calculation method, wherein the calculation formula is as follows:
the above equation represents the magnitude of the similarity between the two n-dimensional vectors a, B, with closer values to 1 indicating more similarity.
Further, step S2 includes the steps of:
s21: for different detected contents in the field of electronic products, taking mobile phone electronic products as examples, determining the detected product types
S22: determining attributes of a product includes: the types of the determined attributes are in aspects of appearance, functions, reliability and the like, and the types of the determined attributes are in aspects of product names, colors, pixels, memories, operating systems, processors, memories and the like;
s23: from each attribute, an attribute value is determined.
Further, step S3 includes the steps of:
s31: according to the actual index obtained by the detection of the electronic product in the actual production process, the actual index is used as an unstructured data type in a database, and is transmitted into a standard knowledge graph of the product, converted into a graph database type and stored in a Neo4j graph database;
s32: and storing the data into a map, searching and inquiring by using the map, and obtaining whether the detection result accords with the standard by searching and comparing whether the standard data and the detection data are consistent, thereby finishing the quality detection of the product.
Further, the step S4 specifically includes: adding a front-end module, providing a basic display page for a user, loading measurement drawing data and required drawing data by a data interface, then directly requesting data from the data interface, calling a drawing JS to complete drawing on the page, visually seeing related relation and attribute values of the queried items, wherein the data in a neo4j graph database exist, nodes and edges are endowed with corresponding key value pair attributes, when the http request is sent out by the front-end page, related condition parameters are transferred to the graph database to serve as query subjects, the corresponding data is read from the neo4j database by using JavaScript, the JavaScript data is processed through a background, then the corresponding node data of the query subjects is labeled and transferred to a web browser, thus effectively carrying out accurate query and display on a product detection object, and visually finding whether the product quality is wrong to complete a product quality detection task.
The invention has the beneficial effects that: the invention uses standard documents and technical specifications of electronic products as standard sources of knowledge maps, adopts a mode layer ontology construction method to limit and define the detection range and content of the products, then uses a Bi-LSTM+attribute+CRF method to extract entity and entity relations in the standard documents and specifications, uses a similarity judgment method to solve the problem of data fusion, stores the extracted entity and relation into a map database neo4j, finally forms a domain knowledge map, and uses the electronic product standard domain knowledge map to detect the quality of the products.
When each product detection module is detected in the production process, the detected index is directly input into a knowledge base, an inference engine compares rules in the knowledge base with data corresponding to the modules in a knowledge graph, compares the attribute and attribute value corresponding to the detection module, and if the actually measured data is within the standard range of the graph, the quality reaches the standard, otherwise, the quality does not meet the standard. By using the method, a large amount of manpower resources can be saved, four aspects of quality detection can be simultaneously associated, the detection efficiency is improved, finally, a front end is set by using a JavaScript technology, when a http request is sent out by a front end page, related condition parameters are transferred to a graph database to serve as a query subject, corresponding data are read from a neo4j database by using JavaScript, the JavaScript data are processed through a background, and then the node data corresponding to the query subject are labeled and transferred to a web browser, so that accurate query and display are effectively carried out on a product detection object, and whether the product quality is wrong or not is intuitively found, and a product quality detection task is completed. The whole detection system can be connected by utilizing the strong data processing capability of the knowledge graph, so that enterprises can perform better data management.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.
Drawings
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:
FIG. 1 is a schematic diagram of a logic structure of a knowledge graph constructed by the invention;
FIG. 2 is a flow chart of the construction of the detection body of the electronic product according to the invention;
FIG. 3 is a flow chart of a method of named entity recognition according to the present invention;
FIG. 4 is a schematic diagram of a knowledge graph construction flow;
FIG. 5 is a flow chart of a method for detecting product quality using a constructed standard knowledge graph of a product.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.
The invention aims to design an electronic product quality detection method based on a knowledge graph, which solves the problems of excessive manual participation, low detection efficiency, high requirements on technical level of detection personnel and less data relevance among detection units in the quality detection process. The technical scheme of the invention is as follows:
firstly, a knowledge graph of the product field is constructed based on standard documents and technical specifications related to electronic products. The method comprises the steps of detecting from four aspects aiming at the quality detection process of the electronic product, determining a mode layer of a knowledge graph, wherein the mode layer of the graph determines the normalization of an upper concept of the detection of the electronic product. The model layer of the knowledge graph is managed by adopting an ontology modeling method, so that the relation among a plurality of objects such as the extracted entity, entity relation, entity type, attribute and the like is standardized by means of the supporting capability of the ontology base on axiom, rules and constraint conditions. And then constructing a data layer on the basis of the mode layer, extracting entity and entity relations from a plurality of heterogeneous data sources by adopting a Bi-LSTM+attribute+CRF method, wherein the data fusion generally adopts a similarity calculation method, and determining whether the entity similarity and the attribute similarity are the same entity to be described or not by utilizing the entity similarity and the attribute similarity so as to perform the data fusion. The data layer uses facts as units to store knowledge in the neo4j graph database, the extracted knowledge is formed into triples in the type of entity-relation-entity or entity-attribute value, and according to the data layer constructed by the mode layer, redundant knowledge of the knowledge base can be reduced, and the structure of the graph is clearer. Finally, all data are connected to form a huge entity relation network, namely a product field knowledge graph.
The method is characterized in that the product quality detection is carried out by utilizing a knowledge graph in the product standard field, the actual index obtained by the detection of the electronic product in the actual production process is firstly used as an unstructured data type in a database, the unstructured data type is transmitted into the product standard knowledge graph and is converted into a graph database type to be stored into Neo4j, neo4j is a novel graph database, the storage supported by the traditional relational database can be realized, and mass data are processed based on graph theory besides the function of analyzing and processing the data. The data are stored in the atlas, the atlas is utilized for inquiring and reasoning, the inference engine compares the rules in the knowledge base with the facts in the knowledge atlas, and whether the detection result accords with the standard is obtained by comparing whether the standard data are consistent with the detection data, so that the product quality detection is completed.
A front-end module is added to provide a basic display page for a user, a basic data interface is provided for loading the measurement drawing data and the required drawing data, then the data is directly requested from the data interface, the drawing JS is called to complete drawing on the page, and related relation and attribute values of the inquired item can be visually seen, so that whether the detection quality meets the standard can be visually seen, and the method is also beneficial to non-professional personnel to directly inquire the data.
Constructing a knowledge graph in the field of electronic products is logically divided into a mode layer and a data layer, wherein the mode layer guides the type and the relation between the data of the data layer. The logic structure diagram of the knowledge graph is shown in fig. 1.
The method mainly comprises the steps of mainly designing content objects for product quality detection in a mode layer, mainly detecting from four aspects of product technical specifications, product appearance detection standards and product function detection standards according to the current electronic product quality detection mode, performing system modeling according to product detection objects to form a product detection body structure, enabling concepts and classification relations of the product detection objects to be clear, enabling the concepts of the detection objects to be integrated and objectively formed into a whole, defining knowledge description systems, categories and possible relations among the knowledge description systems, the categories and the possible relations among the knowledge description systems, and finally representing the body of the product detection in a form of knowledge. The specific flow is shown in fig. 2.
After the schema layer is constructed, a data layer is constructed on the basis. The data of the data layer mainly depends on electronic product standard detection documents such as: "YD/T1539-2006 Mobile communication handset reliability technical requirement and test method", "YS/T711-2009 aluminium alloy plate, strip for mobile phone and digital product housing", "GB/T9298-1998 cross-cut experiment of paint and varnish film", "GB/T5170.1-2016 electrical and electronic product environmental test device inspection method part 1: general rule "and the like, and specifications for product production, most of which are unstructured data. The data source is processed correspondingly, and corresponding entity identification is needed to be performed on the data source first. Entity identification employing recurrent neural networks
The Bi-LSTM+Attention+CRF method and the whole training flow are shown in figure 3. The unstructured data mainly takes text as input, in order to be recognized by a computer, the unstructured data is forwarded to corresponding character vectors by a word2vec method, a CBOW model is utilized, in which original sentences are presumed to obtain target words, namely the target words are obtained through knowing the current word w n The probability P of the existence of the current word is presumed by utilizing the front and back 2k words:
P(w n |SUM(w n-k ,w n-k+1 …,w n-1+k ,w n+k )) (1)
thereby calculating the probability of the sentence S using the CBOW model, the calculation expression is as follows:
t is the input text length and P (S) represents the sentence in (w 1 ,w 2 ,…,w T ) The probability of occurrence is the joint probability of each word. Likelihood functions are then constructed for the entire text:
deriving a log-likelihood function:
the goal of CBOW model training is to maximize the objective likelihood function.
When the input character is Word2vec, the corresponding character vector is obtained, and the character vector is input into the Bi-LSTM network. The input character vector is used for obtaining past information through a forward LSTM, future information is obtained through a backward LSTM, and the context information can be contained simultaneously through a Bi-LSTM model. In order to improve the recognition efficiency, focusing on important entities, adding an Attention mechanism;
when character vectors are input into the CRF after passing through the Attention mechanism, the purpose of adding the CRF layer is to label words in sentences, the whole text sequence can be labeled on the sentence level, and the character vectors can play the role of a multi-classifier to make up the limitation of each word label sequence after Bi-LSTM training. The CRF study sequence characteristics from sentence level can not only obtain each optimal entity tag sequence, but also automatically learn the limit rule of the following tags in the training process. For an input sequence z= (Z) 1 ,z 2 ,…,z n ) Where n is the input sequence length, z i If the i-th word of the input vector is represented, the output sequence label corresponding to Z is y= (Y) 1 ,y 2 ,…,y n ). For taking value in a given sequence ZIn the condition of z, the conditional probability of the value Y on the labeling sequence Y is p (y|z)
S(z)＝∑ y exp(∑ N,k λ k t k (y N-1 ,y N ,z,N)+∑ N,l μ l s l (y N ,z,N)) (6)
N＝1,2,…,n，s l (. Cndot.) and t k (. Cndot.) is a characteristic function, μ l And lambda (lambda) k Respectively, the corresponding weights, S (z) being a normalization factor. In the training stage, the maximum likelihood estimation is used for solving the optimal sequence label, and the corresponding likelihood logarithm is: sigma (sigma) N The input sequence is marked by the highest conditional probability y obtained through training of the lovp (y|z):
the named entity can be identified through the process, and the relation extraction and the attribute extraction are also completed in the whole training process. The attribute extraction is to extract attribute information of the entity from the text, and the attribute of the entity can be regarded as a noun relation between the entity and the attribute, so that the problem of attribute extraction is regarded as a relation extraction. The relation extraction and named entity recognition are part of information extraction, and belong to a multi-classification problem in essence, bi-LSTM is utilized to extract advanced features from a word embedding layer, then an Attention mechanism generates a weight vector, vocabulary-level features in each iteration are combined into sentence-level features through multiplication of the output value of the Bi-LSTM and the weight vector, and finally the sentence-level feature vectors are used for relation classification. The method can avoid the dependence on some existing vocabulary resources, system or manually extracted features in the traditional method, and reduces the calculation complexity.
Related entities have been obtained but the entities have ambiguous directions and thus identified junctionsIf the result is difficult to be directly put into a gallery, disambiguation must be performed on the result of entity identification to obtain unambiguous entity information. Giving the entity to be disambiguated a set o=o by adopting a clustering-based entity disambiguation method 1 ,O 2 ,…,O n Extracting the characteristic of each entity index O, expressing the characteristic vector as a feature vector, calculating the similarity between the entities, and clustering the entity index items by adopting a clustering algorithm so that each category in a clustering result corresponds to a specific target entity. The calculation similarity adopts a cosine similarity calculation method, and the calculation formula is as follows:
the above equation represents the magnitude of the similarity between the two n-dimensional vectors a, B, with closer values to 1 indicating more similarity.
Knowledge fusion: the method is characterized in that data from different databases are processed, when a standard electronic product knowledge graph is constructed, the data sources are mainly field standards and product specifications, but when the quality of a product is detected, the produced specific type electronic product and required data are fused, and because the targets in the process of product detection are clear, the frame matching method, namely the body alignment method, is utilized for fusion to further obtain the completion of the knowledge graph aiming at four aspects, namely the product technical specification, the product appearance detection, the product functional detection and the product reliability detection.
Through the technology, the corresponding entity, entity relation and attribute value in the field of electronic products are obtained, and in the field, the detection is carried out according to the attribute value of the entity, so that the graph database is selected for data storage, the internal structure of the knowledge graph can be directly and accurately reflected based on the storage mode of the graph structure, the query and reasoning of knowledge are facilitated, the data are imported into the graph database neo4j to form the relation between nodes, the knowledge graph in the whole field of products is completed, and the specific flow is shown in fig. 4.
The knowledge graph constructed according to the steps is a graph of an electronic product detection standard, items and indexes during product detection are detected in production, the product quality detection based on the knowledge graph is realized by searching and inquiring the graph, and a flow chart is shown in fig. 5.
In the knowledge graph, all aspects of the product detection standard types, indexes and attribute values are established, namely the range of normal test values or other attributes, in the actual product detection process, whether the product reaches the standard is mainly detected from four aspects, so that actually obtained data are stored into a knowledge base, then the fact matching is carried out on the data in the knowledge base and the data in the graph in a graph searching mode, if the data match with the attribute values in the graph, the data do not reach the standard, otherwise, the data in the detected four aspects can be queried at the same time, so that the detection efficiency and timeliness can be improved, the product quality detection is realized, when all aspects of detection are finished, if the data do not meet the standard, the graph can be utilized to integrate the data together for reasoning, the finding of a fault occurrence unit is facilitated, and the detection time is reduced for detection personnel.
However, the content of the knowledge graph in the graph database is unfavorable for direct query of non-professional staff, so that a front-end module is added, a basic display page is provided for a user, a basic data interface is provided for loading the data of the graph and the required data of the graph, then the data is directly requested from the data interface, the graph JS is called to complete the graph on the page, and the related relation and attribute values of the queried items can be visually seen, so that whether the detection quality accords with the standard can be more intuitively seen.
When the data in the neo4j graph database exists, the nodes and the edges are endowed with the corresponding key value pair attribute, when the front-end page sends an http request, the related condition parameters are transferred to the graph database to be used as a query subject, the corresponding data is read from the neo4j database by using JavaScript, the JavaScript data is processed through a background, then the node data corresponding to the query subject is labeled and transferred to a web browser, thus the accurate query and display are effectively carried out on the product detection object, and whether the product quality is mistakenly finished or not is intuitively found.
The scheme can be applied to the field of electronic products, the thought and the method are universal, a standard knowledge graph in the field can be constructed for a specific electronic product, then in the actual production process, the data generated by detection are subjected to reasoning and inquiry, and whether the quality meets the related requirements can be detected. For example, in the mobile phone quality detection, the requirements of the production manufacturer and the existing industry and national standards are used as standard data sources, the existing data are extracted according to the standards of the mode layer, and the entity and attribute values corresponding to the requirements are extracted by using a neural network method and stored in a database. In the actual production detection process, according to each specific detection index data, the detection index is determined whether to be qualified by searching and inquiring through the map through front-end input, so that quality detection is completed.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.
Claims (4)
1. The method for detecting the quality of the electronic product based on the knowledge graph is characterized by comprising the following steps of:
s1: constructing a knowledge graph in the field of electronic products according to the relevant standard as a basis, wherein the graph is used as a basis for quality detection;
s2: monitoring the name, the detected attribute and the value range of the product to be detected;
s3: matching the product attribute to be detected with the attribute in the domain knowledge graph, determining a detection target, inquiring the corresponding attribute value, and determining the product quality according to the matching result of the attribute value;
s4: the front-end module is added and used for directly inquiring the quality detection data;
the step S1 includes the following steps:
aiming at the detection process of the electronic product, relevant standards are derived from national standards, industry standards, enterprise standards and product technical specifications, and particularly, the detection is carried out from four aspects of the product technical specifications, product appearance detection standards, product function detection standards and product reliability detection standards aiming at the quality detection process of the electronic product, so as to determine a mode layer of a knowledge graph;
performing entity identification on a data source by adopting a method of a circulating neural network Bi-LSTM+attribute+CRF, and constructing a data layer on the basis of a mode layer;
disambiguating the entity identification result by adopting a clustering-based entity disambiguation method to obtain disambiguated entity information;
the obtained data comprise corresponding entities, entity relations and attribute values in the field of electronic products, the data in a graph database are selected for storage, namely, the data are imported into the graph database neo4j to form the relation between nodes, and the knowledge graph of the whole product field is completed;
determining a triplet entity and a relation thereof, and obtaining a product name-attribute value triplet by using a Bi-LSTM+attribute+CRF method for the characteristics of mobile phone electronic products, wherein the specific method comprises the following steps:
(1) Converting text vectors in the knowledge base into character vectors by using a word2vec method;
(2) Inputting the character vector into a Bi-LSTM network, obtaining past information through a forward LSTM, obtaining future information through a backward LSTM, and simultaneously containing context information through a Bi-LSTM model;
(3) An Attention mechanism is added, so that effective keywords are highlighted, and the recognition efficiency is improved;
(4) Inputting the trained vector into CRF, making up the limitation of each label sequence after Bi-LSTM training to increase the recognition rate;
step S1 comprises the steps of:
s11: detecting from four aspects of a product technical specification, a product appearance detection standard, a product function detection standard and a product reliability detection standard aiming at the quality detection process of the electronic product, and determining a mode layer of a knowledge graph;
s111: according to the current quality detection mode of the electronic product, detecting from four aspects of a product technical specification, a product appearance detection standard and a product function detection standard, and performing system modeling according to a product detection object to form a product detection body structure;
s112: the concept and the classification relation of the product detection object are defined, and the concept of the detection object is integrated and objectively formed into a whole;
s113: defining a knowledge description system, a category of each detection object and a possible relation between each detection object;
s114: ontology through knowledge representation form product detection;
s12: performing entity identification on a data source by adopting a method of a circulating neural network Bi-LSTM+attribute+CRF, and constructing a data layer on the basis of a mode layer;
s121: the data of the data layer comprises an electronic product standard detection document and a technical specification of product production, belongs to unstructured data, and takes the unstructured data as input by taking a text;
s122: converting into corresponding character vector by word2vec method, and obtaining target word by CBOW model in which original sentence is presumed, i.e. by knowing current word w n The probability P of the existence of the current word is presumed by utilizing the front and back 2k words:
P(w n |SUM(w n-k ，w n-k+1 …，w n-1+k ，w n+k )) (1)
thereby calculating the probability of the sentence S using the CBOW model, the calculation expression is as follows:
t is the input text length and P (S) represents the sentence in (w 1 ，w 2 ，…，w T ) The probability of occurrence is the joint probability of each word; likelihood functions are then constructed for the entire text:
deriving a log-likelihood function:
the aim of CBOW model training is to make the objective likelihood function reach the maximum value;
s123: when the input character is Word2vec to obtain a corresponding character vector, the corresponding character vector is input into a Bi-LSTM network, the input character vector is subjected to forward LSTM to obtain past information, backward LSTM is subjected to backward LSTM to obtain future information, and the context information can be simultaneously contained through a Bi-LSTM model;
s124: in order to improve the recognition efficiency, focusing on important entities, adding an Attention mechanism;
s125: after passing through the Attention mechanism, the character vector is input into the CRF to play a role of a multi-classifier to make up for the limitation of each word label sequence after Bi-LSTM training;
for an input sequence z= (Z) 1 ，z 2 ，…，z n ) Where n is the input sequence length, z i If the i-th word of the input vector is represented, the output sequence label corresponding to Z is y= (Y) 1 ,y 2 ，…，y n ) The method comprises the steps of carrying out a first treatment on the surface of the For a given sequence Z, the conditional probability of labeling the sequence Y with a value of Y is p (y|z)
N＝1，2，…，n，s l (. Cndot.) and t k (. Cndot.) is a characteristic function, μ l And lambda (lambda) k Respectively corresponding weights, and S (z) is a normalization factor; in the training stage, the maximum likelihood estimation is used for solving the optimal sequence label, and the corresponding likelihood logarithm is: sigma (sigma) N The input sequence is marked by the highest conditional probability y obtained through training of the lovp (y|z):
s13: disambiguating the entity identification result by adopting a clustering-based entity disambiguation method to obtain disambiguated entity information; giving the entity to be disambiguated a set o=o by adopting a clustering-based entity disambiguation method 1 ,O 2 ，…，O n Extracting the feature of each entity name O, expressing the feature vector as a feature vector, calculating the similarity among the entities, clustering the entity name items by adopting a clustering algorithm, enabling each category in a clustering result to correspond to a specific target entity, and calculating the similarity by adopting a cosine similarity calculation method, wherein the calculation formula is as follows:
the above formula represents the magnitude of the similarity between two n-dimensional vectors a, B, the closer the value is to 1, the more similar it is;
s14: fusing data from different databases by using a frame matching method, namely a body alignment method, so as to obtain the complement of the knowledge graph;
s15: through the steps S11-S14, the obtained data comprise corresponding entities, entity relations and attribute values in the field of electronic products, the data in the graph database are selected for storage, namely, the data are imported into the graph database neo4j to form the relations among nodes, and the knowledge graph of the whole product field is completed.
2. The electronic product quality detection method based on the knowledge graph according to claim 1, wherein the method is characterized in that: step S2 comprises the steps of:
s21: for different detected contents in the field of electronic products, taking mobile phone electronic products as examples, determining the detected product types
S22: determining attributes of a product includes: appearance, function, reliability, and determining the type of the attribute, wherein the type of the attribute comprises a product name, a color, a pixel, a memory, an operating system, a processor and a memory;
s23: from each attribute, an attribute value is determined.
3. The electronic product quality detection method based on the knowledge graph according to claim 1, wherein the method is characterized in that: step S3 comprises the steps of:
s31: according to the actual index obtained by the detection of the electronic product in the actual production process, the actual index is used as an unstructured data type in a database, and is transmitted into a standard knowledge graph of the product, converted into a graph database type and stored in a Neo4j graph database;
s32: and storing the data into a map, searching and inquiring by using the map, and obtaining whether the detection result accords with the standard by searching and comparing whether the standard data and the detection data are consistent, thereby finishing the quality detection of the product.
4. The electronic product quality detection method based on the knowledge graph according to claim 1, wherein the method is characterized in that: the step S4 specifically comprises the following steps: adding a front-end module, providing a basic display page for a user, loading measurement drawing data and required drawing data by a data interface, then directly requesting data from the data interface, calling a drawing JS to complete drawing on the page, visually seeing related relation and attribute values of the queried items, wherein the data in a neo4j graph database exist, nodes and edges are endowed with corresponding key value pair attributes, when the http request is sent out by the front-end page, related condition parameters are transferred to the graph database to serve as query subjects, the corresponding data is read from the neo4j database by using JavaScript, the JavaScript data is processed through a background, then the corresponding node data of the query subjects is labeled and transferred to a web browser, thus effectively carrying out accurate query and display on a product detection object, and visually finding whether the product quality is wrong to complete a product quality detection task.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910667154.9A CN110362660B (en) | 2019-07-23 | 2019-07-23 | Electronic product quality automatic detection method based on knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910667154.9A CN110362660B (en) | 2019-07-23 | 2019-07-23 | Electronic product quality automatic detection method based on knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110362660A CN110362660A (en) | 2019-10-22 |
| CN110362660B true CN110362660B (en) | 2023-06-09 |
Family
ID=68219834
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910667154.9A Active CN110362660B (en) | 2019-07-23 | 2019-07-23 | Electronic product quality automatic detection method based on knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110362660B (en) |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110888808B (en) * | 2019-11-16 | 2023-01-31 | 云南湾谷科技有限公司 | Web intelligent test method based on knowledge graph |
| CN111160847B (en) * | 2019-12-09 | 2023-08-25 | 中国建设银行股份有限公司 | Method and device for processing flow information |
| CN110990748B (en) * | 2019-12-18 | 2023-06-27 | 成都迪普曼林信息技术有限公司 | Public opinion data collection and release system |
| CN111159421A (en) * | 2019-12-25 | 2020-05-15 | 中国建设银行股份有限公司 | Knowledge graph-based fund query method and device |
| CN111191851B (en) * | 2020-01-03 | 2023-06-23 | 中国科学院信息工程研究所 | Knowledge graph-based data center energy efficiency optimization method |
| CN111639498A (en) * | 2020-04-21 | 2020-09-08 | 平安国际智慧城市科技股份有限公司 | Knowledge extraction method and device, electronic equipment and storage medium |
| CN111695771B (en) * | 2020-05-07 | 2024-02-27 | 国网安徽省电力有限公司淮南供电公司 | Electric power material intelligent detection management and control system and method based on Internet of things technology |
| CN111667147A (en) * | 2020-05-11 | 2020-09-15 | 北京方升互联信息技术有限公司 | Professional definition method for detecting service products |
| CN111506722B (en) * | 2020-06-16 | 2024-03-08 | 平安科技（深圳）有限公司 | Knowledge graph question-answering method, device and equipment based on deep learning technology |
| CN112084302B (en) * | 2020-08-24 | 2024-04-30 | 江苏易达捷信数字科技有限公司 | Method, system, device and storage medium for detecting inventory data of cost file |
| CN112364173B (en) * | 2020-10-21 | 2022-03-18 | 中国电子科技网络信息安全有限公司 | IP address mechanism tracing method based on knowledge graph |
| CN112381438A (en) * | 2020-11-24 | 2021-02-19 | 北京三维天地科技股份有限公司 | Product quality and standard matching model based on fuzzy similarity algorithm |
| CN112541600A (en) * | 2020-12-07 | 2021-03-23 | 上海电科智能系统股份有限公司 | Knowledge graph-based auxiliary maintenance decision method |
| CN112598563B (en) * | 2020-12-29 | 2023-11-17 | 中国科学技术大学 | Smart city data construction method based on knowledge graph |
| CN112732939B (en) * | 2021-01-15 | 2022-11-29 | 中国科学院空天信息创新研究院 | Spatiotemporal knowledge graph construction method, device, medium and equipment based on GraphDB |
| CN112801492B (en) * | 2021-01-22 | 2023-07-25 | 中国平安人寿保险股份有限公司 | Knowledge-hierarchy-based data quality inspection method and device and computer equipment |
| CN112883197B (en) * | 2021-02-08 | 2023-02-07 | 广东电网有限责任公司广州供电局 | Knowledge graph construction method and system for closed switch equipment |
| CN113626574B (en) * | 2021-08-19 | 2023-08-29 | 成都数联云算科技有限公司 | Information query method, system and device and medium |
| CN114417015B (en) * | 2022-01-26 | 2023-05-12 | 西南交通大学 | High-speed train maintainability knowledge graph construction method |
| CN115544265A (en) * | 2022-09-13 | 2022-12-30 | 南京航空航天大学 | Bearing fault diagnosis method based on bearing fault knowledge graph |
| CN117171367B (en) * | 2023-09-26 | 2024-04-12 | 北京泰策科技有限公司 | Specification detection method for specified attribute values of different database tables |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3407151A1 (en) * | 2017-05-24 | 2018-11-28 | Tata Consultancy Services Limited | Systems and methods for cognitive control of data acquisition for efficient fault diagnosis |
| CN109146611A (en) * | 2018-07-16 | 2019-01-04 | 浙江大学 | A kind of electric business product quality credit index analysis method and system |
| CN109378053A (en) * | 2018-11-30 | 2019-02-22 | 安徽影联云享医疗科技有限公司 | A kind of knowledge mapping construction method for medical image |
| CN109460010A (en) * | 2018-12-18 | 2019-03-12 | 彩虹无线（北京）新技术有限公司 | The vehicle fault detection method, apparatus and storage medium of knowledge based map |
| CN109948911A (en) * | 2019-02-27 | 2019-06-28 | 北京邮电大学 | A kind of appraisal procedure calculating networking products Information Security Risk |
| CN110032647A (en) * | 2019-03-12 | 2019-07-19 | 埃睿迪信息技术（北京）有限公司 | Method, apparatus and storage medium based on industrial circle building knowledge mapping |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11392598B2 (en) * | 2016-10-19 | 2022-07-19 | Ebay Inc. | Applying a quantitative range for qualitative terms |
| CN108647791B (en) * | 2018-03-30 | 2020-12-29 | 中国标准化研究院 | Multi-source automobile safety information processing method, device and system |
| CN108447534A (en) * | 2018-05-18 | 2018-08-24 | 灵玖中科软件（北京）有限公司 | A kind of electronic health record data quality management method based on NLP |
| CN109145122A (en) * | 2018-08-02 | 2019-01-04 | 北京仿真中心 | A kind of product know-how map construction and querying method and system |
| CN109064318A (en) * | 2018-08-24 | 2018-12-21 | 苏宁消费金融有限公司 | A kind of internet financial risks monitoring system of knowledge based map |
-
2019
- 2019-07-23 CN CN201910667154.9A patent/CN110362660B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3407151A1 (en) * | 2017-05-24 | 2018-11-28 | Tata Consultancy Services Limited | Systems and methods for cognitive control of data acquisition for efficient fault diagnosis |
| CN109146611A (en) * | 2018-07-16 | 2019-01-04 | 浙江大学 | A kind of electric business product quality credit index analysis method and system |
| CN109378053A (en) * | 2018-11-30 | 2019-02-22 | 安徽影联云享医疗科技有限公司 | A kind of knowledge mapping construction method for medical image |
| CN109460010A (en) * | 2018-12-18 | 2019-03-12 | 彩虹无线（北京）新技术有限公司 | The vehicle fault detection method, apparatus and storage medium of knowledge based map |
| CN109948911A (en) * | 2019-02-27 | 2019-06-28 | 北京邮电大学 | A kind of appraisal procedure calculating networking products Information Security Risk |
| CN110032647A (en) * | 2019-03-12 | 2019-07-19 | 埃睿迪信息技术（北京）有限公司 | Method, apparatus and storage medium based on industrial circle building knowledge mapping |
Non-Patent Citations (2)
| Title |
|---|
| Yachen Tang等.Enhancement of Equipment Mangement Using Knowledge Graph.《2019 IEEE Innovative Smart Grid Technologies-ASIA》.2019,第1-6页. * |
| 李秀芳等.基于生产设备监测数据的故障诊断仪的开发与应用.《河南中烟工业有限责任公司》.2018,第1页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110362660A (en) | 2019-10-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110362660B (en) | Electronic product quality automatic detection method based on knowledge graph | |
| US20180232443A1 (en) | Intelligent matching system with ontology-aided relation extraction | |
| CN110019732B (en) | Intelligent question answering method and related device | |
| CN110309268B (en) | Cross-language information retrieval method based on concept graph | |
| CN108932342A (en) | A kind of method of semantic matches, the learning method of model and server | |
| CN110222045A (en) | A kind of data sheet acquisition methods, device and computer equipment, storage medium | |
| US20200334249A1 (en) | Automatic transformation of complex tables in documents into computer understandable structured format and providing schema-less query support data extraction | |
| CN116795973B (en) | Text processing method and device based on artificial intelligence, electronic equipment and medium | |
| KR100980579B1 (en) | Method and system for searching by using natural language query in ontology | |
| Chang et al. | Integrating a semantic-based retrieval agent into case-based reasoning systems: A case study of an online bookstore | |
| CN112270188B (en) | Questioning type analysis path recommendation method, system and storage medium | |
| CN116775847A (en) | Question answering method and system based on knowledge graph and large language model | |
| US20200334250A1 (en) | Automatic transformation of complex tables in documents into computer understandable structured format with mapped dependencies and providing schema-less query support for searching table data | |
| Geng et al. | Cross-domain ontology construction and alignment from online customer product reviews | |
| CN112182145A (en) | Text similarity determination method, device, equipment and storage medium | |
| Liu et al. | Open intent discovery through unsupervised semantic clustering and dependency parsing | |
| Miao et al. | A dynamic financial knowledge graph based on reinforcement learning and transfer learning | |
| CN113761208A (en) | Scientific and technological innovation information classification method and storage device based on knowledge graph | |
| CN113901224A (en) | Knowledge distillation-based secret-related text recognition model training method, system and device | |
| CN115952770B (en) | Data standardization processing method and device, electronic equipment and storage medium | |
| Angermann et al. | Taxonomy Matching Using Background Knowledge | |
| CN116151258A (en) | Text disambiguation method, electronic device and storage medium | |
| US20220215493A1 (en) | Method, Apparatus, and Electronic Device for Obtaining Trademark Similarity | |
| Deng et al. | Covidia: COVID-19 Interdisciplinary Academic Knowledge Graph | |
| CN114547313A (en) | Resource type identification method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |