A method for constructing supply chain online knowledge graph based on multivariate heterogeneous data mining

Abstract

The present invention relates to a supply chain online knowledge graph construction method based on multivariate heterogeneous data mining, comprising the following steps: knowledge graph construction framework; knowledge graph pattern layer construction, dividing the platform carrying online reviews into e-commerce websites, social media and professional websites, combining with the demand analysis of the online review knowledge graph, dividing the description dimension of the online reviews into three dimensions: source characteristics, content characteristics and form characteristics; data layer construction is mainly based on the constructed pattern layer, extracting and fusing the corresponding content in the multi-source online review data, and identifying the corresponding instances, attributes and relationships; storage and drawing of the knowledge graph.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graphs, and in particular to an online knowledge graph method for a supply chain based on multivariate heterogeneous data mining. Background Art

[0002] In the era of big data, discovering and using competitive intelligence from massive data to support decision-making is an important issue facing enterprises and individuals. The massive online review data accumulated on the Internet is an important data source for consumers and businesses to obtain information. At present, online reviews are characterized by multi-heterogeneous distribution, huge data volume, redundancy and lack of retrieval support, which brings great challenges to the organization and utilization of online review information in the big data environment; and users have different information needs. Some users focus on price, and some users focus on performance. The current online review display method and review usefulness ranking related research assume that user needs are consistent, without considering the personalized needs of users; in addition, the demand for intelligent information services in the era of artificial intelligence is more urgent, and the existing online review research is not in-depth enough, lacking systematic representation and modeling of multi-heterogeneous online review domain knowledge.

[0003] Knowledge graph is a very important way of organizing information. On the one hand, it can dynamically associate commodity information and review information with different structures and wide distribution; on the other hand, knowledge graph can improve the information organization of online reviews, solve the problem of information overload, and use an efficient data management model to handle fragmented and redundant online review information; in addition, knowledge graph can organize online review data from a semantic perspective, and then conduct intelligent search, providing solutions for the deep application of online reviews and users' personalized information needs. The continuous development of network technology has led to the rapid generation of multivariate heterogeneous data. Mining and processing complex multivariate heterogeneous data can effectively obtain data potential information and rules. Cluster analysis is an interdisciplinary subject in the fields of machine learning, data mining, and statistics. It can not only play an effective demonstration role, that is, determine the classification standard or category criteria of things, but also the role of clustering is induction, without the need to determine the classification standard and analyze the data object. Therefore, it is of great significance to effectively mine and analyze complex multivariate heterogeneous data. Summary of the invention

[0004] In view of the problem that online reviews are multivariate and heterogeneous in distribution, have huge data volume, are redundant and do not support retrieval, and cannot be effectively used by enterprises, the present invention proposes a supply chain online knowledge graph method based on multivariate and heterogeneous data mining. The technical solution of the present invention is implemented through the following steps:

[0005] A supply chain online knowledge graph construction method based on multivariate heterogeneous data mining includes the following steps:

[0006] Step 1: Build a knowledge graph framework, construct an enterprise design resource ontology library, and establish standardized descriptions of design requirements and design resources from multiple levels;

[0007] Step 2: Construct the model layer of the knowledge graph. The platforms that carry online reviews are divided into e-commerce websites, social media, and professional websites. Combined with the demand analysis of the online review knowledge graph, the description dimensions of online reviews are divided into three dimensions: source characteristics, content characteristics, and form characteristics. The method is as follows:

[0008] Step 2.1: Source features are used to describe the external information of the review; content features are used to describe the attitude and satisfaction of users involved in the review towards the product and service; word count, number of views, and review time are all formal features of the review;

[0009] Step 2.2, set up 7 core categories, including comments, comment content, feature views, users, comment objects, platforms and products; the comment object concept is further divided into two sub-concepts: services and products, and more abundant data features of descriptive classes are defined through attributes;

[0010] Step 2.3: Model the knowledge of the defined classes and relationships and visualize the pattern layer, centering on the comment class and constructing the relationships between entities outwards;

[0011] Step 3: The data layer is constructed mainly based on the constructed pattern layer, extracting and integrating the corresponding content in the multi-source online review data, and identifying the corresponding instances, attributes and relationships;

[0012] Step 3.1: Take the mobile phone reviews on e-commerce websites, related term encyclopedias, reviews, and news content as data sources, and capture the mobile phone review data, encyclopedia data, and review news data as the corpus for word vector training;

[0013] Step 3.2: Extract and classify product feature words by constructing a professional corpus and training word vectors based on word2vec;

[0014] Step 3.3: The word2Vec-based word vector model trains the corpus through a neural network algorithm, then converts words into word vectors and maps them to a high-dimensional space, predicting the probability of surrounding words based on the current word;

[0015] Step 3.4, obtain the vector representation of the words, measure the similarity between words by cosine distance, use the concepts defined in the pattern layer as seed words, sort the related words according to similarity, and perform manual screening and classification;

[0016] Step 3.5, extract feature-opinion pairs from the comment text;

[0017] Step 4: Storage and drawing of knowledge graph.

[0018] Step 1 specifically includes:

[0019] Step 1.1, crawl data by analyzing the existence and specific data of online reviews;

[0020] Step 1.2, design the concepts, relationships and attributes of the model layer by analyzing the source characteristics, content characteristics and form characteristics of online comments;

[0021] Step 1.3, based on the construction foundation of the pattern layer, Word2vec word vector training, opinion extraction, and sentiment analysis are used to extract entities, relationships, and attributes;

[0022] Step 1.4: Integrate the processed structured online review data and store it in the graph database, and use the graph database query language to complete the visualization, semantic retrieval and deep mining of the knowledge graph.

[0023] The beneficial effects of the present invention are as follows: a knowledge graph is constructed using multi-source heterogeneous online review data, and a framework for constructing a knowledge graph based on multi-source heterogeneous data is proposed. The model layer is constructed around the source, content and form of the online reviews, and finally a conceptual framework of the knowledge graph is formed. Word2vec is used to obtain entities, relationships and attributes from multi-source heterogeneous texts, and data fusion and knowledge graph analysis are performed. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Build a pipeline for online review knowledge graphs.

[0025] Figure 2 Describing dimensions for online reviews.

[0026] Figure 3 Visualize part of the knowledge graph DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0028] The technical solution of the present invention is achieved through the following steps:

[0029] Step 1: Build a knowledge graph framework, construct an enterprise design resource ontology library, and establish standardized descriptions of design requirements and design resources from multiple levels.

[0030] Step 1.1, by analyzing the existence and specific data of online comments, data crawling is performed by writing a Python program.

[0031] Step 1.2, design the concepts, relationships and attributes of the model layer by analyzing the source characteristics, content characteristics and form characteristics of online comments.

[0032] Step 1.3, based on the construction foundation of the pattern layer, use Word2vec word vector training, opinion extraction, sentiment analysis and other methods to extract entities, relationships and attributes.

[0033] Step 1.4: Integrate the processed structured online review data and store it in the graph database Neo4j. Use the graph database query language Cypher to complete the visualization, semantic retrieval, and deep mining of the knowledge graph.

[0034] Step 2: construct the model layer of the knowledge graph. Based on the extensive collection and research of online review information, the platforms that carry online reviews are divided into e-commerce websites, social media, and professional websites. Combined with the demand analysis of the online review knowledge graph, the description dimensions of online reviews are divided into three dimensions: source characteristics, content characteristics, and form characteristics.

[0035] Step 2.1, where the source features are used to describe the external information of the review; the content features are used to describe the attitude and satisfaction of the users involved in the review towards the products and services; the number of words, number of opinions, and review time are all formal features of the review.

[0036] Step 2.2, the present invention sets 7 core categories, including comments, comment content, feature viewpoints, users, comment objects, platforms and products. The comment object concept is further divided into two sub-concepts: services and products. At the same time, more abundant data features of the description class are defined through attributes.

[0037] Step 2.3, use the Protege5.5.0 ontology building tool to perform knowledge modeling on the defined classes and relationships, and use the OntoGraf function to visualize the pattern layer. Centered on the comment class, we build relationships with other entities outward, and a total of 17 relationships are constructed.

[0038] Step 3: The data layer is constructed mainly based on the constructed pattern layer, extracting and fusing the corresponding content in the multi-source online review data, and identifying the corresponding instances, attributes and relationships.

[0039] Step 3.1: Use mobile phone reviews on e-commerce websites, related term encyclopedias, reviews, and news content as data sources, and capture more than 30,000 mobile phone review data, encyclopedia data, and review news data totaling more than 4.2 million words as the corpus for word vector training.

[0040] Step 3.2: Extract and classify product feature words by building a professional corpus and training word vectors based on word2vec.

[0041] Step 3.3: The word2Vec-based word vector model trains the corpus through a neural network algorithm, then converts words into word vectors and maps them to a high-dimensional space. Write a Python program to call the Skip-gram model in the Gensim library to predict the probability of surrounding words based on the current word.

[0042] Step 3.4, after model training is completed, the vector representation of the words can be obtained, and the similarity between words is measured by cosine distance. The concepts defined in the pattern layer are used as seed words, and the related words are sorted according to similarity, and manually screened and classified.

[0043]

[0044] Where Sim(p1,p2) is the semantic similarity; d represents the path length between words p1 and p2; α is the semantic distance value when the similarity is 0.5.

[0045] In step 3.5, the deep semantic sentiment analysis model provided by Baidu AI Open Platform is used to extract feature-opinion pairs from the comment text, and fine-grained sentiment analysis is performed based on this.

[0046] Step 4: Select Neo4j as the storage and drawing tool for the knowledge graph.

[0047] Step 4.1: Present the online review knowledge graph through Cypher query statements.

[0048] like Figure 1 As shown, this embodiment involves a supply chain online knowledge graph method based on multi-heterogeneous data mining, including: a data source and collection module, a model layer construction module, a data layer construction module, a knowledge graph drawing module and a knowledge graph application module.

[0049] The data source and collection module includes writing a Python program to crawl data, and the crawling sources are professional forums, encyclopedia website product official websites and e-commerce official websites.

[0050] The model layer construction module includes defining classes and class structure, semantic relationships between classes, and class attributes. Combined with the demand analysis of online review knowledge graph, the description dimensions of online reviews are divided into three dimensions: source features, content features, and form features. Figure 2 shown.

[0051] The source feature is used to describe the external information of the review; the content feature is used to describe the attitude and satisfaction of the user involved in the review towards the product and service. The form feature includes the number of words, the number of viewpoints, and the time of the review.

[0052] According to the dimensions, 7 core categories are set, among which the concept of review object is further divided into two sub-concepts: service and product. At the same time, more abundant data features of the description class are defined through attributes. The classes and attributes defined in the model layer are shown in Table 1.

[0053] Table 1 Online review knowledge graph model layer classes and attribute definitions

[0054]

[0055]

[0056] The defined classes and relationships are modeled using the Protege5.5.0 ontology building tool, and the OntoGraf function is used to visualize the pattern layer. With the comment class as the center, the relationships with other entities are constructed outward. A total of 17 relationships are constructed, as shown in Table 2.

[0057] Table 2 Online review knowledge graph relationship table

[0058]

[0059] The data layer construction includes concept extraction, relationship extraction, and attribute extraction. Word vector training based on word2vec completes the extraction and classification of product feature words, takes mobile phone reviews on e-commerce websites, related term encyclopedias, reviews and news content as data sources, and captures more than 30,000 mobile phone review data, encyclopedia data, and review news data totaling more than 200,000 words as the corpus for word vector training. The word vector model based on word2Vec trains the corpus through a neural network algorithm, then converts words into word vectors and maps them to a high-dimensional space. Write a Python program and call the Skip-gram model in the Gensim library for word vector training.

[0060] After the model training is completed, the vector representation of the words can be obtained, and the similarity between words is measured by cosine distance. The concepts defined in the pattern layer are used as seed words, and the cosine distance is used to sort the related words according to the relevance, and manual screening and classification are performed.

[0061]

[0062] Where Sim(p1,p2) is the semantic similarity; d represents the path length between words p1 and p2; α is the semantic distance value when the similarity is 0.5.

[0063] With the help of the deep semantic sentiment analysis model provided by Baidu AI Open Platform, feature-opinion extraction is performed on the comment text, and fine-grained sentiment analysis is performed based on this.

[0064] The knowledge graph drawing module selects Neo4j as the storage and drawing tool of the knowledge graph, and its essence is to completely convert nodes, relationships and attributes into the knowledge graph.

[0065] The knowledge graph application modules include query, visualization and knowledge discovery.

[0066] The query is to present the online review knowledge graph through Cypher query statements after the nodes and relationships are imported into Neo4j.

[0067] The visualization is to directly view the relationship between nodes from the graph data. Some knowledge graphs are as follows Figure 3 shown.

[0068] The knowledge discovery includes querying the attributes of an object through the Cypher language, such as querying the advantages and disadvantages of a battery; and includes performing a Pearson correlation test through SPSS Statics 24 software to determine the correlation and difference of multivariate heterogeneous data.

[0069] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A supply chain online knowledge graph construction method based on multivariate heterogeneous data mining, comprising the following steps: Step 1: Build a knowledge graph framework, construct an enterprise design resource ontology library, and establish standardized descriptions of design requirements and design resources from multiple levels; Step 2: Construct the model layer of the knowledge graph. The platforms that carry online reviews are divided into e-commerce websites, social media, and professional websites. Combined with the demand analysis of the online review knowledge graph, the description dimensions of online reviews are divided into three dimensions: source characteristics, content characteristics, and form characteristics. The method is as follows: Step 2.1: Source features are used to describe the external information of the review; content features are used to describe the attitude and satisfaction of users involved in the review towards the products and services; word count, number of opinions, and review time are all formal features of the review; Step 2.2, set up 7 core categories, including comments, comment content, feature views, users, comment objects, platforms and products; the comment object concept is further divided into two sub-concepts: services and products. At the same time, more abundant data features of descriptive classes are defined through attributes; Step 2.3: Model the knowledge of the defined classes and relationships and visualize the pattern layer, centering on the comment class and constructing the relationships between entities outwards; Step 3: The data layer is constructed mainly based on the constructed pattern layer, extracting and integrating the corresponding content in the multi-source online review data, and identifying the corresponding instances, attributes and relationships; Step 3.1: Take the mobile phone reviews on e-commerce websites, related term encyclopedias, reviews, and news content as data sources, and capture the mobile phone review data, encyclopedia data, and review news data as the corpus for word vector training; Step 3.2: Extract and classify product feature words by building a professional corpus and training word vectors based on word2vec. Step 3.3: The word2Vec-based word vector model trains the corpus through a neural network algorithm, then converts words into word vectors and maps them to a high-dimensional space, predicting the probability of surrounding words based on the current word; Step 3.4, obtain the vector representation of the words, measure the similarity between words by cosine distance, use the concepts defined in the pattern layer as seed words, sort the related words according to similarity, and perform manual screening and classification; Step 3.5, extract feature-opinion pairs from the comment text; Step 4: Storage and drawing of knowledge graph.

2. According to claim 1, a method for constructing an online knowledge graph of a supply chain based on multivariate heterogeneous data mining is characterized in that: Step 1 specifically includes: Step 1.1, crawl data by analyzing the existence and specific data of online reviews; Step 1.2, design the concepts, relationships and attributes of the model layer by analyzing the source characteristics, content characteristics and form characteristics of online comments; Step 1.3, based on the construction foundation of the pattern layer, Word2vec word vector training, opinion extraction, and sentiment analysis are used to extract entities, relationships, and attributes; Step 1.4: Integrate the processed structured online review data and store it in the graph database, and use the graph database query language to complete the visualization, semantic retrieval and deep mining of the knowledge graph.

Images