A knowledge graph-based multi-source data difference traceability retrieval method

Abstract

The application relates to the technical field of multi-source data difference traceability retrieval, in particular to a multi-source data difference traceability retrieval method based on a knowledge graph, which comprises the following steps: knowledge graph construction, sorting and merging multi-element data with differences, extracting effective information, and realizing construction of the knowledge graph, specifically comprising the following steps: knowledge extraction and knowledge merging; multi-source data traceability retrieval, specifically comprising the following steps: knowledge graph data retrieval and traceability, requesting a Neo4j database through a Django server to obtain key-value pair information of each node in the graph database. The application solves the problems of difficult knowledge acquisition and relationship embodiment through knowledge graph construction, constructs field knowledge ontology from top to bottom by acquiring professional data of different sources and analyzing and sorting the professional data, and then extracts and fuses the data, so that the knowledge semantic information is more complete, and the knowledge graph is constructed.

Description

Technical Field

[0001] This invention relates to the field of multi-source data difference tracing and retrieval technology, specifically a multi-source data difference tracing and retrieval method based on knowledge graphs. Background Technology

[0002] The primary difference between traditional semantic networks and knowledge graphs lies in their scale. Knowledge graphs are large-scale semantic networks, and compared to various semantic networks of the 1970s and 80s, the most significant difference is their scale. Extending this further, the fundamental difference between knowledge representations in the big data era, represented by knowledge graphs, and traditional knowledge representations is primarily reflected in their scale. Traditional knowledge engineering's series of knowledge representations are typically "small knowledge."

[0003] Traditional methods for processing multi-source heterogeneous data utilize techniques such as data mining, machine learning, and statistical analysis to analyze and process integrated multi-source heterogeneous data, thereby extracting valuable information and knowledge. Currently, most multi-source heterogeneous data processing falls under the category of data analysis. Data association analysis aims to discover relationships between seemingly irregular data points, thereby revealing patterns and trends. Commonly used traditional data processing algorithms include the Apriori algorithm and the FP-Growth algorithm. Multivariate heterogeneous data processing algorithms mainly fall into two categories: classification algorithms and clustering algorithms. Classification refers to dividing data into predefined categories using algorithms. Commonly used classification algorithms include decision tree algorithms, rule-based classification methods, artificial neural network algorithms, deep learning algorithms, support vector machine (SVM) algorithms, and Bayesian algorithms. Clustering groups similar data into one category, aiming to maximize the similarity of data within each category. Commonly used clustering algorithms include four main categories: partition-based clustering methods, hierarchical clustering methods, density-based clustering methods, and model-based clustering methods.

[0004] The representation of knowledge relationships is not comprehensive enough. Among existing knowledge acquisition models, the most common is the Euclidean space point set model, which mainly reflects the strength of associations between entities, with less emphasis on relational structure and semantic information. Most expressions of knowledge structure are vector model, mapping knowledge relationships to polar coordinates; however, this model only addresses the structural relationships between knowledge entities, offering limited semantic information. The recently popular complex vector space model performs well with multi-relationship structures, but it incurs significant computational overhead.

[0005] Knowledge graph source tracing is challenging, and searching through massive amounts of data is inherently inefficient. This is especially true in knowledge graphs, where the problem often extends beyond a single entity to include its relational nodes and even its long-tail ancestor nodes. Current knowledge graph solutions for long-tail path queries largely rely on improvements to random walks, utilizing reinforcement learning to access and locate target entity nodes. However, when faced with complex relationships such as many-to-many or closed-loop relationships, reinforcement learning-based random walks become increasingly prone to getting stuck in local optima and path loops, wasting computational resources and failing to find the correct entity. Therefore, this paper proposes a multi-source data difference-based source tracing and retrieval method based on knowledge graphs. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-source data difference tracing and retrieval method based on knowledge graphs to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A knowledge graph-based method for tracing and retrieving the differences in multi-source data includes the following steps:

[0009] Step S1, knowledge graph construction, involves organizing and merging the diverse data to extract valid information and construct the knowledge graph. This includes the following steps:

[0010] a. Knowledge extraction: Knowledge extraction uses named entities in the data corpus as nodes. An entity corresponds to a thing in the real world. From an object-oriented perspective, an entity can also be regarded as an object. Each entity has its own attributes. Each entity represents a knowledge unit in the graph. When searching, relevant information with the entity as the search unit will also be returned. Entities are used in the text through their names. The part of speech of an entity is a noun or a noun phrase.

[0011] Based on the principle of top-down name entity extraction, a professional-level vocabulary list for the domain was first compiled by combining the opinions of domain experts. In addition, after extraction, it was found that many words in the corpus data are also suitable as knowledge units in the knowledge graph. Therefore, this project will use keyword extraction algorithms to extract keywords from the corpus and filter out noun keywords as named entities in the graph.

[0012] b. Knowledge merging: Merging entities from different data that describe the same goal to obtain a hierarchical node with a specific target.

[0013] First, the data is filtered using the existing domain data system and its standardized descriptions to exclude entities that do not conform to the original system plan. Then, the similarity of the entity strings is used for alignment.

[0014] First, alignment is performed using alias attributes. For entities lacking alias information, the entity name string features, entity description, attribute information, and semantic web information are used sequentially to align the entities. The string similarity between two entities can be vectorized using the Word2Vec model and then calculated using the cosine similarity of the vectors. The entity description similarity can be calculated using the Doc2Vec model, edit distance, and TF-IDF algorithm. Semantic web information refers to the structural information of entities in the semantic web, which can also be considered as the related entity information. The entity's embedded vector is used as its semantic web structural feature vector, and then the matching score is calculated using cosine similarity.

[0015] Step S2, multi-source data tracing and retrieval: Based on the constructed knowledge graph, extract the knowledge features to be traced, and input these features into the tracing model to achieve tracing and retrieval. Specifically, this includes the following steps:

[0016] a) Knowledge graph data retrieval and tracing: Source tracing is used to determine the problems and sources of suspicious data. Source tracing is used for data with characteristics such as large volume, multiple levels, multiple dimensions, multiple sources, and inconsistent descriptions. It constructs professional knowledge graphs to solve the knowledge-level representation and analysis of large-scale professional data in the big data environment. On this basis, source tracing is carried out by combining semantics and graph structure to help users understand and analyze the relationships between data.

[0017] The source tracing retrieval section is divided into source tracing retrieval of nodes and source tracing retrieval of natural language query statements, and includes the following two cases:

[0018] If the input content is a node in a knowledge graph, you can directly use the Cypher statement in Neo4j to perform nested queries on the source information of the node based on the two relationships of provider and source;

[0019] If the input is natural language, the HanLp toolkit performs dependency parsing and semantic dependency parsing on the user's natural language query. Entity relations are extracted as question features based on a custom dictionary. Then, based on the semantic understanding of these feature words and the entity relations in the knowledge graph, different question types are established. Next, referring to the feature words of the input question, the Naive Bayes classification algorithm is used to match the category of the query question. Finally, parameters are passed to a preset Cypher query template, and the results are retrieved from the Neo4j graph database and returned to the user.

[0020] b. Through the Django server, requests are made to the Neo4j database to obtain the key-value pair information of each node in the graph database. The Jackjson package is used to convert the semi-structured map key-value pair data into JSON data required by the front end. The front end uses the D3 front-end development toolkit to render the JSON data.

[0021] The knowledge extraction in step S1 uses the open-source software package HanLP for named entity recognition. The recognition process includes the following steps: multi-source feature data input, data acquisition, ontology construction, knowledge extraction and fusion, knowledge storage, feature knowledge graph, data processing, problem classification, source tracing retrieval, and result presentation.

[0022] The detailed retrieval process for knowledge graph data retrieval and source tracing in step S2 includes: data preparation, Bayesian classifier, and source tracing.

[0023] The data preparation includes the following process: professional knowledge graph, data preprocessing, collection of problem description statements, definition of problem categories, and selection of features to build a feature vocabulary. The steps of professional knowledge graph, data preprocessing, collection of problem description statements, definition of problem categories, and selection of features to build a feature vocabulary are performed sequentially.

[0024] The Bayesian classifier includes a question sample set, 70% training samples, 30% test samples, a classification model trained, a Naive Bayes classification model loaded, and test questions classified according to feature values ​​to obtain classification labels.

[0025] The question sample set is connected with the collected question description statements; the 70% training samples are used to train a classification model, and the 30% test samples are used to process data with the trained classification model to load a Naive Bayes classification model. The loaded Naive Bayes classification model is connected to obtain a classification of the test questions based on feature values, and a classification label is obtained by classifying the test questions based on feature values.

[0026] The source tracing includes the following steps;

[0027] a) Determine if the user input is a sentence. If it is, proceed to step b; otherwise, proceed to step e.

[0028] b. Sentence semantic analysis;

[0029] c. Feature words, and classify the test question based on the feature values ​​using a Bayesian classifier;

[0030] d. Obtain the classification labels from the Bayesian classifier, match the question template based on the classification labels, and apply the Cypher statement corresponding to the template;

[0031] e, query the data source;

[0032] Return the query results.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. This paper presents a multi-source data difference tracing and retrieval method based on knowledge graphs. By constructing a knowledge graph, it addresses the difficulties in knowledge acquisition and the representation of relationships. To avoid insufficient expression of structural relationships and semantic information in the knowledge graph, hierarchical structure is extracted during the data preprocessing stage before knowledge graph construction. Data is pre-classified and assigned different hierarchical relationships. This enables the display of data structure during knowledge graph construction. By acquiring professional data from different sources, analyzing and organizing it, a domain knowledge ontology is constructed from top to bottom. Then, knowledge extraction and fusion are performed on these data to make their semantic information more complete, thereby constructing the knowledge graph.

[0035] 2. This invention presents a multi-source data differential tracing and retrieval method based on knowledge graphs. Through knowledge tracing and retrieval, it addresses the difficulties of tracing the source of data in knowledge graph problems. It not only enables searching of single nodes but also displays the related nodes and source paths of those nodes. The tracing and retrieval module addresses the characteristics of professional data, such as large volume, multi-level, multi-dimensional, multi-source, and inconsistent descriptions, by constructing a knowledge graph to solve the knowledge-level representation and analysis of large-scale data in a big data environment. Based on this, it combines semantics and graph structure for tracing and retrieval, helping users understand and analyze the relationships between data. The knowledge graph is stored in a graph database, and then, based on user operations, it enables tracing and retrieval of nodes and natural language queries. Finally, the retrieval results are presented to the user, helping them understand the sources of different data descriptions and the relationships between them. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the knowledge graph construction and tracing technology process of the present invention;

[0037] Figure 2 This is a schematic diagram illustrating the knowledge extraction process of the present invention.

[0038] Figure 3 This is a schematic diagram illustrating the knowledge alignment of the present invention;

[0039] Figure 4 This is a structural diagram of the traceability module of the present invention;

[0040] Figure 5 This is a detailed retrieval process diagram for knowledge graph data retrieval and tracing according to the present invention;

[0041] Figure 6 This is a schematic diagram of the knowledge graph display technology process of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] In the description of this invention, it should be understood that the terms "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0044] In the description of this patent, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "setting" should be interpreted broadly. For example, they can refer to a fixed connection or setting, a detachable connection or setting, or an integral connection or setting. Those skilled in the art can understand the specific meaning of the above terms in this patent according to the specific circumstances.

[0045] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a number" means two or more, unless otherwise explicitly specified.

[0046] Please see Figures 1-6 As shown, the present invention provides a technical solution:

[0047] A knowledge graph-based method for tracing and retrieving the differences in multi-source data includes the following steps:

[0048] Step S1, knowledge graph construction, involves organizing and merging the diverse data to extract valid information and construct the knowledge graph. This includes the following steps:

[0049] a. Knowledge extraction: Knowledge extraction uses named entities in the data corpus as nodes. An entity corresponds to a thing in the real world. From an object-oriented perspective, an entity can also be regarded as an object. Each entity has its own attributes. Each entity represents a knowledge unit in the graph. When searching, relevant information with the entity as the search unit will also be returned. Entities are used in the text through their names. The part of speech of an entity is a noun or a noun phrase.

[0050] Based on the principle of top-down name entity extraction, a professional-level vocabulary list for the domain was first compiled by combining the opinions of domain experts. In addition, after extraction, it was found that many words in the corpus data are also suitable as knowledge units in the knowledge graph. Therefore, this project will use keyword extraction algorithms to extract keywords from the corpus and filter out noun keywords as named entities in the graph.

[0051] b. Knowledge merging: Merging entities from different data that describe the same goal to obtain a hierarchical node with a specific target.

[0052] First, the data is filtered using the existing domain data system and its standardized descriptions to exclude entities that do not conform to the original system plan. Then, the similarity of the entity strings is used for alignment.

[0053] First, alignment is performed using alias attributes. For entities lacking alias information, the entity name string features, entity description, attribute information, and semantic web information are used sequentially to align the entities. The string similarity between two entities can be vectorized using the Word2Vec model and then calculated using the cosine similarity of the vectors. The entity description similarity can be calculated using the Doc2Vec model, edit distance, and TF-IDF algorithm. Semantic web information refers to the structural information of entities in the semantic web, which can also be considered as the related entity information. The entity's embedded vector is used as its semantic web structural feature vector, and then the matching score is calculated using cosine similarity.

[0054] Step S2, multi-source data tracing and retrieval: Based on the constructed knowledge graph, extract the knowledge features to be traced, and input these features into the tracing model to achieve tracing and retrieval. Specifically, this includes the following steps:

[0055] a) Knowledge graph data retrieval and tracing: Source tracing is used to determine the problems and sources of suspicious data. Source tracing is used for data with characteristics such as large volume, multiple levels, multiple dimensions, multiple sources, and inconsistent descriptions. It constructs professional knowledge graphs to solve the knowledge-level representation and analysis of large-scale professional data in the big data environment. On this basis, source tracing is carried out by combining semantics and graph structure to help users understand and analyze the relationships between data.

[0056] The source tracing retrieval section is divided into source tracing retrieval of nodes and source tracing retrieval of natural language query statements, and includes the following two cases:

[0057] If the input content is a node in a knowledge graph, you can directly use the Cypher statement in Neo4j to perform nested queries on the source information of the node based on the two relationships of provider and source;

[0058] If the input is natural language, the HanLp toolkit performs dependency parsing and semantic dependency parsing on the user's natural language query. Entity relations are extracted as question features based on a custom dictionary. Then, based on the semantic understanding of these feature words and the entity relations in the knowledge graph, different question types are established. Next, referring to the feature words of the input question, the Naive Bayes classification algorithm is used to match the category of the query question. Finally, parameters are passed to a preset Cypher query template, and the results are retrieved from the Neo4j graph database and returned to the user.

[0059] b. Through the Django server, requests are made to the Neo4j database to obtain the key-value pair information of each node in the graph database. The Jackjson package is used to convert the semi-structured map key-value pair data into JSON data required by the front end. The front end uses the D3 front-end development toolkit to render the JSON data.

[0060] The knowledge extraction in step S1 uses the open-source software package HanLP for named entity recognition. The recognition process includes the following steps: multi-source feature data input, data acquisition, ontology construction, knowledge extraction and fusion, knowledge storage, feature knowledge graph, data processing, problem classification, source tracing retrieval, and result presentation.

[0061] The detailed retrieval process for knowledge graph data retrieval and source tracing in step S2 includes: data preparation, Bayesian classifier, and source tracing.

[0062] Data preparation includes the following steps: professional knowledge graph, data preprocessing, collection of problem description statements, definition of problem categories, selection of features to build a feature vocabulary.

[0063] The Bayesian classifier consists of a question sample set, 70% training samples, and 30% test samples. The classifier is trained to obtain a classification model. The Naive Bayes classification model is then loaded, and the test questions are classified according to the feature values ​​to obtain the classification labels.

[0064] The question sample set is connected with the collected question description statements; 70% of the training samples are used to train a classification model, and 30% of the test samples are used to process the data of the trained classification model to load a Naive Bayes classification model. The Naive Bayes classification model is then connected to obtain the classification of the test questions based on the feature values, and the classification labels are obtained by classifying the test questions based on the feature values.

[0065] The process of tracing the source includes the following steps;

[0066] a) Determine if the user input is a sentence. If it is, proceed to step b; otherwise, proceed to step e.

[0067] b. Sentence semantic analysis;

[0068] c. Feature words, and classify the test question based on the feature values ​​using a Bayesian classifier;

[0069] d. Obtain the classification labels from the Bayesian classifier, match the question template based on the classification labels, and apply the corresponding Cypher statement to the template;

[0070] e, query the data source;

[0071] f returns the query results.

[0072] Constructing a hierarchical knowledge graph is a significant advancement. Traditional knowledge graph construction techniques often focus on preserving semantic information and extracting local structural information. However, when dealing with complex relationships, especially hierarchical ones, traditional techniques often fail to preserve the structure. This patent, however, demonstrates superior performance in knowledge graph construction with hierarchical data, enabling hierarchical knowledge graph construction. Furthermore, in knowledge graph source retrieval, traditional techniques often fall into local optima and path loops when searching for paths in complex structures. This technology leverages the hierarchical distribution during construction to easily achieve path finding and node source retrieval.

[0073] Knowledge graph construction addresses the difficulties in knowledge acquisition and the representation of relationships. To avoid insufficient expression of structural relationships and semantic information in the knowledge graph, hierarchical structure is extracted during the data preprocessing stage before knowledge graph construction. Data is pre-classified and assigned different hierarchical relationships. This allows for the visualization of the data structure during knowledge graph construction. By acquiring, analyzing, and organizing professional data from different sources, a domain knowledge ontology is constructed from top to bottom. Then, knowledge extraction and fusion are performed on this data to make its semantic information more complete, thereby constructing the knowledge graph.

[0074] Knowledge tracing and retrieval addresses the challenges of tracing the origins of knowledge graphs. It not only enables searching of individual nodes but also displays the related nodes and source paths of those nodes. The tracing and retrieval module addresses the characteristics of professional data—large volume, multi-layered, multi-dimensional, multi-source, and inconsistent descriptions—by constructing a knowledge graph to solve the knowledge-level representation and analysis of large-scale data in a big data environment. Based on this, it combines semantics and graph structure for tracing and retrieval, helping users understand and analyze the relationships between data. The knowledge graph is stored in a graph database, and based on user operations, it enables tracing and retrieval of nodes and natural language queries. Finally, the retrieval results are presented to the user, helping them understand the origins of different data descriptions and the relationships between them.

[0075] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for knowledge graph-based multi-source data difference traceability retrieval, characterized in that, Comprise the following steps: Step S1, knowledge graph construction, the multi-element data with difference is sorted and combined, the effective information is extracted, the construction of knowledge graph is realized, specifically comprising the following steps: a, knowledge extraction, the named entity in the data corpus is selected as the node, one entity corresponds to a thing in the real world, from the object-oriented point of view, the entity is regarded as an object, each entity has its own attribute, each entity represents a knowledge unit in the graph, and the related information is returned in the form of entity retrieval unit, the entity is used in the text through its name, and the entity word is a noun or noun phrase; Based on the principle of from top to bottom, first, combined with the opinions of field experts, the professional hierarchical vocabulary table of the field is sorted out, in addition, through extraction, it is found that many words in the corpus data are also suitable as knowledge units in the knowledge graph, so the key word extraction algorithm is used to extract the key words in the corpus, and the noun key words are filtered out as the named entities in the graph; b, knowledge merging, the entities describing the same target obtained by learning different data are merged to obtain a hierarchical node with specific direction; First, the data is screened by means of the existing field data system and the standard description, and the entities not conforming to the original system planning are excluded, and then the entity string similarity is used for alignment; Firstly, the alias attribute is used for alignment, and for the entity lacking alias information, the entity is aligned in turn by using the string characteristics of entity name, entity description, attribute information and semantic network information, wherein the string similarity of two entities is vectorized through Word2Vec model, and then the matching score is calculated through the cosine similarity of vector, the entity description similarity is calculated through Doc2Vec model, edit distance and TF-IDF algorithm, the semantic network information refers to the structural information of entity in semantic network, which is also regarded as the associated entity information of entity, the embedded vector of entity is used as its semantic network structure feature vector, and then the matching score is calculated through cosine similarity; Step S2, multi-source data traceability retrieval, according to the constructed knowledge graph, the knowledge characteristics to be traced are extracted, the characteristics are brought into the traceability model, the traceability retrieval is realized, specifically comprising the following steps: a, knowledge graph data retrieval and traceability, the problem and data source of suspicious data are determined through traceability retrieval, the traceability retrieval is aimed at large amount, multi-level, multi-dimensional, multi-source and inconsistent description data, a professional knowledge graph is constructed, on the basis of which, the traceability retrieval is combined with semantics and graph structure to help users understand and analyze the relationship between data; The traceability retrieval part is divided into node traceability retrieval and natural language query question traceability retrieval, which is divided into the following two cases: If the input content is a node in the knowledge graph, the Cypher statement is directly used in neo4j to query the source information of the node according to the provider and source two relationships, and the multi-node and multi-relation nested query is carried out; If the input content is natural language, the natural language query sentence input by the user is analyzed by the HanLp tool package for dependency syntax and semantic dependency analysis, entity relationships are extracted as problem features according to a self-defined dictionary, and then different problem types are established according to the semantic understanding of these feature words and the entity relationships in the knowledge graph; then, the category of the query problem is matched by using the Naive Bayes classification algorithm with reference to the feature words of the input query sentence; finally, parameters are passed to the preset Cypher statement query template to retrieve the results from the graph database Neo4j and return them to the user; b. Through the Django server, the Neo4j database is requested to obtain the key-value pair information of each node in the graph database, and the semi-structured map key-value pair data is converted into the required Json data of the front end by using the Jackjson package, and the front end uses the D3 front-end development package to realize the rendering of the Json data.

2. The multi-source data difference traceability retrieval method based on a knowledge graph according to claim 1, characterized in that: The knowledge extraction in step S1 uses the open source software package HanLP for named entity recognition, and the recognition process includes the following steps: multi-source feature data input, data acquisition, ontology construction, knowledge extraction and fusion, knowledge storage, feature knowledge graph, data processing, problem classification, traceability retrieval, and result presentation.

3. The method of claim 1, wherein the method further comprises: The detailed retrieval process of the knowledge graph data retrieval and traceability in step S2 includes: data preparation, Bayes classifier, and traceability search; The data preparation includes the following processes: professional knowledge graph, data preprocessing, collecting problem description sentences, defining problem categories, selecting features to build a feature word library, and the professional knowledge graph, data preprocessing, collecting problem description sentences, defining problem categories, and selecting features to build a feature word library steps are performed in sequence; The Bayes classifier includes a query sample set, 70% training samples, 30% test samples, a classification model trained, a Naive Bayes classification model loaded, classification of test query sentences according to feature values, and classification labels obtained; The query sample set is connected with the collected problem description sentences; the 70% training samples are used to train the classification model, the 30% test samples are used to load the Naive Bayes classification model after processing data with the trained classification model, and the Naive Bayes classification model is connected to obtain classification of test query sentences according to feature values, and classification labels are obtained by classifying test query sentences according to feature values; The traceability search includes the following steps: a. Determine whether the user input is a sentence, if yes, go to step b, if not, go to step e; b. Syntax semantic analysis; c. Feature words, and classification of test query sentences according to feature values by the Bayes classifier; d. Classification labels are obtained by the Bayes classifier, problem templates are matched according to the classification labels, and the Cypher statement corresponding to the template is applied; e. Query data sources; f. Return the query results.

Images