An audit guideline intelligent question and answer method, system and device based on a knowledge graph

Abstract

This invention discloses a knowledge graph-based intelligent question-and-answer method for audit guidance, relating to the field of digital auditing. Through knowledge graph framework design and ETL data warehouse technology, this invention extracts, cleans, and transforms audit business data relevant to business environment optimization audits, achieving the fusion of various business data to provide data support for intelligent question-and-answer. By constructing an intelligent question-and-answer system, and designing a multi-task-oriented approach based on real-world audit scenarios, it utilizes word segmentation, matching, and query modules to achieve knowledge graph-based intelligent question-and-answer for business environment optimization audit guidance. This reduces the workload of the audit process, freeing auditors from tedious and repetitive tasks, allowing them to focus more on risk identification, problem analysis, and root cause analysis, effectively improving audit efficiency and quality.

Description

Technical Field

[0001] This invention relates to the field of digital auditing technology, and in particular to an intelligent question-and-answer method, system and device for audit guidance based on knowledge graphs. Background Technology

[0002] Compared to traditional auditing, digital auditing offers advantages such as ensuring audit quality and enabling audit institutions to effectively fulfill their responsibilities. In recent years, audit systems have increasingly leveraged the research findings of information theory and systems theory to conduct audit work, making the construction of digital audit platforms an effective and necessary path to deepen the development of audit informatization. Therefore, how to integrate the concept of business-audit integration into the digital audit process is a key issue that enterprises need to focus on when improving the effectiveness and efficiency of internal auditing. However, in the process of promoting digital auditing, the data generated by audit work is growing exponentially, and the data sources are becoming more diverse. Business data is stored in different business information management systems and their corresponding databases, resulting in problems such as multi-source heterogeneity, data redundancy, poor correlation, and low efficiency in resource sharing and cross-professional reuse. Auditors often need to rely on their professional knowledge to understand and become familiar with various internal management systems of the enterprise before they can carry out audit work and obtain the audit information they need, leading to complex, repetitive, and inefficient audit work. Summary of the Invention

[0003] To address the problems mentioned in the background, this invention provides a knowledge graph-based intelligent question-and-answer method, system, and device for audit guidance, thereby reducing the workload of the audit process and improving the efficiency of internal audit work in enterprises.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] On the one hand, this invention provides an intelligent question-answering method for audit guidance based on knowledge graphs, including:

[0006] Collect audit business data related to optimizing the business environment within enterprises and design an audit knowledge graph structure; the audit business data includes three intelligent audit databases, intermediate tables, and audit model database data;

[0007] By integrating audit business data included in the audit knowledge graph structure through ETL data warehouse technology, a Neo4j graph database is constructed.

[0008] The system acquires user questions and performs word segmentation, then performs dependency parsing on the segmentation results to derive the intended keyword combination.

[0009] By using precise matching and fuzzy matching algorithms, the combination of intent keywords is matched with the graph data in the Neo4j graph database to obtain candidate entities and attribute sets;

[0010] The candidate entities and attribute sets are populated into the query template to form a Cypher query statement, which returns the query results by querying the Neo4j graph database.

[0011] Optionally, the collection of audit business data related to optimizing the business environment within the enterprise and the design of an audit knowledge graph structure specifically include:

[0012] Collect data from the enterprise's internal smart audit database, intermediate tables, and audit model library related to optimizing the business environment as audit business data;

[0013] Audit business data is divided into five types of entities: model library, method library, problem library, fields, and intermediate tables. The fields in each entity table are designed as entity attributes, and the business relationships between entities are designed as edges, thus constructing an audit knowledge graph structure.

[0014] Optionally, the step of integrating audit business data included in the audit knowledge graph structure using ETL data warehouse technology to construct a Neo4j graph database specifically includes:

[0015] The audit business data included in the audit knowledge graph structure is extracted, cleaned and transformed using ETL data warehouse technology. Then, Cypher statements are written to store the data in the Neo4j graph database as graph data, thus constructing the Neo4j graph database.

[0016] Optionally, the step of obtaining the user's question and performing word segmentation, then performing dependency parsing on the segmentation results to derive the intent keyword combination, specifically includes:

[0017] The user's question was segmented using the jieba word segmentation tool by loading a stop word library, a thesaurus, and a domain word library, and the segmentation results were obtained.

[0018] The dependency parsing model is invoked to intelligently analyze the word segmentation results and derive the intent keyword combination.

[0019] Optionally, the step of using precise matching and fuzzy matching algorithms to match the combination of intent keywords with graph data in the Neo4j graph database to obtain candidate entities and attribute sets specifically includes:

[0020] For correctly split keywords in the intent keyword combination, the exact matching algorithm is used to read the graph data in the Neo4j graph database, and the entity attribute words stored in the graph are converted into a word list. A program is written to compare the intent keywords with the entity attribute word list stored in the graph. Those that are exactly the same are stored in the candidate entity and attribute set.

[0021] For keywords that are not correctly split in the intent keyword combination, a fuzzy matching algorithm is used to read the graph data in the Neo4j graph database. A program is written to compare the intent keywords with the entity attribute word list stored in the graph, and those with a similarity of 70% are stored in the candidate entity and attribute set.

[0022] Optionally, before populating the candidate entities and attribute sets into the query template to form the Cypher query statement, the method further includes:

[0023] The query template is constructed by designing a multi-task orientation based on the audit business scenario; the multi-task orientation includes audit model orientation, audit method orientation, audit intermediate table orientation, audit field orientation, audit issue orientation, audit basis orientation, and audit suggestion orientation.

[0024] On the other hand, the present invention also provides an intelligent question-and-answer system for audit guidance based on knowledge graphs, comprising:

[0025] The knowledge graph design module is used to collect audit business data related to optimizing the business environment within enterprises and design the audit knowledge graph structure; the audit business data includes three intelligent audit databases, intermediate tables, and audit model database data;

[0026] The graph database construction module is used to integrate audit business data included in the audit knowledge graph structure through ETL data warehouse technology to build the Neo4j graph database;

[0027] The word segmentation module is used to acquire user questions and segment them into words, perform dependency parsing on the segmentation results, and derive the intent keyword combination.

[0028] The matching module is used to match the combination of intent keywords with the graph data in the Neo4j graph database using precise matching and fuzzy matching algorithms to obtain candidate entities and attribute sets;

[0029] The query module is used to populate the candidate entities and attribute sets into the query template to form a Cypher query statement, and return the query results by querying the Neo4j graph database.

[0030] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned knowledge graph-based audit guidance intelligent question answering method.

[0031] Optionally, the memory is a non-transitory computer-readable storage medium.

[0032] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0033] This invention provides a knowledge graph-based intelligent question-and-answer method, system, and device for audit guidance. Through knowledge graph framework design and ETL data warehouse technology, it extracts, cleans, and transforms audit business data relevant to business environment optimization audits, achieving the fusion of various business data to provide data support for intelligent question-and-answer. By constructing an intelligent question-and-answer system and designing a multi-task-oriented approach based on real-world audit scenarios, it utilizes word segmentation, matching, and query modules to achieve knowledge graph-based intelligent question-and-answer for business environment optimization audit guidance. This reduces the workload of the audit process, freeing auditors from tedious and repetitive tasks, allowing them to focus more on risk identification, problem analysis, and root cause analysis. This effectively improves audit efficiency and quality, meeting audit requirements and corporate strategic needs. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart of an intelligent question-and-answer method for audit guidance based on knowledge graphs, according to the present invention.

[0036] Figure 2 This is a schematic diagram illustrating the overall construction concept of the intelligent question-answering method for audit guidance based on knowledge graphs according to the present invention.

[0037] Figure 3 This is a schematic diagram illustrating the business relationships between entities within an audit knowledge graph structure.

[0038] Figure 4 This is a schematic diagram of dependency parsing of word segmentation results;

[0039] Figure 5 A schematic diagram illustrating relation types analyzed by the dependency parsing model;

[0040] Figure 6 An audit business guidance diagram for the business logic of an intelligent question-and-answer system;

[0041] Figure 7 A schematic diagram illustrating the natural language question processing process of the intelligent question-answering system for audit guidance. 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] The purpose of this invention is to provide an intelligent question-and-answer method for audit guidance based on knowledge graphs, so as to reduce the workload of the audit process and improve the efficiency of internal audit work in enterprises.

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Figure 1 and Figure 2 These are a flowchart and a schematic diagram of the overall construction idea of ​​an intelligent question-answering method for audit guidance based on knowledge graphs, as described in this invention. See also... Figure 1 and Figure 2 A knowledge graph-based intelligent question-answering method for audit guidance includes:

[0046] Step 1: Collect audit business data related to optimizing the business environment within the enterprise and design an audit knowledge graph structure.

[0047] The rapid development of knowledge graphs has provided high-quality knowledge sources for the implementation of intelligent question-answering systems, directly promoting their development in various industries (such as customer service and healthcare). Addressing the challenges in digital auditing, an intelligent question-answering system for business environment audit guidance, built using natural language processing, knowledge graphs, and intelligent question-answering technologies, can help auditors accurately and quickly extract necessary knowledge from massive amounts of audit information. This will help enterprises improve the efficiency of internal audit work, promote digital audit transformation, and advance the digitalization of enterprise audit operations.

[0048] This invention optimizes the construction of the business environment knowledge graph, which is mainly divided into two steps: knowledge graph structure design (step 1) and ETL data integration (step 2). Just as building a conventional database requires designing the data table structure, before importing data into the Neo4j graph database, a knowledge graph framework (schema) needs to be designed. This process mainly involves defining entity categories, attribute categories, and relationship categories based on business logic and data characteristics. Then, based on the business data and the designed knowledge graph structure, ETL data warehouse technology is used to import the data into the Neo4j graph database, completing data storage and integration. ETL is the process of extracting, cleaning, and transforming data from the business system and loading it into a specified database. Its purpose is to integrate scattered, disorganized, and inconsistent data within an enterprise, and it is a crucial step in building a knowledge graph.

[0049] Specifically, this invention first collects audit business data related to optimizing the business environment within enterprises, such as the three intelligent audit databases, intermediate tables, and audit model databases. Audit experts and technical experts analyze the business relationships between the data, transforming data fields and business relationships into entities, attributes, and edges of a knowledge graph, and designing the structure of the audit knowledge graph.

[0050] The audit business data can be viewed as multiple Excel spreadsheets, each designed as an entity, with fields in each spreadsheet representing entity attributes, and business relationships between tables represented as edges. For example, there are five entity types: model library, method library, problem library, field table, and intermediate table. The problem library entity has attributes such as problem description, problem classification, policy basis, and problem characterization. The problem characterization in the problem library has business relationships with the model library and method library, designed as edges. Figure 3 As shown in the table below. The specific entities and their corresponding attributes are shown in Table 1.

[0051] Table 1. Entities and Corresponding Attributes in the Audit Knowledge Graph Structure

[0052]

[0053] Step 2: Use ETL data warehouse technology to integrate the audit business data included in the audit knowledge graph structure to build the Neo4j graph database.

[0054] Based on the audit knowledge graph structure designed in step 1, the audit business data mentioned above is extracted, cleaned, and transformed through ETL data warehouse technology and stored in the Neo4j graph database, thus completing the integration of audit business data.

[0055] ETL data warehouse technology is implemented using Java programs. Before importing data, it needs to be formatted according to a specified Excel format. The Java program reads the data according to the format (extraction), while filtering out characters such as spaces, line breaks, and tabs (cleaning). The formatted data is then converted from tabular format to graph data format and stored as variables (conversion). Finally, Cypher statements (the query language specified by Neo4j database) are used to store the data in the Neo4j graph database. The entire process is implemented using Java programs based on the data business logic.

[0056] Step 3: Obtain the user's question and perform word segmentation. Then, perform dependency parsing on the segmentation results to derive the intent keyword combination.

[0057] The intelligent question-answering mechanism of this invention is mainly constructed through three mechanisms: word segmentation (step 3), matching (step 4), and querying (step 5). The word segmentation mechanism breaks down user questions into combinations of one or more keywords, requiring a stop word library, a thesaurus, a domain-specific thesaurus, and dependency parsing. Stop words are used to filter meaningless characters or words in user questions, such as punctuation marks like commas and colons, and interjections like "ah" and "ne." The thesaurus and domain-specific thesaurus consist of specialized vocabulary specific to the auditing business domain, such as economic responsibility audit (synonyms include "economic responsibility audit"), fund management inspection (synonyms include "fund verification" and "fund management audit"), and engineering audit. Their function is to help the word segmentation tool correctly break down professional domain vocabulary. After breaking down the user question, a dependency parsing model is needed to intelligently analyze the sentence components and the relationships between words, thereby obtaining the most accurate combination of intent keywords.

[0058] Step 3 specifically includes:

[0059] Step 3.1: Use the jieba word segmentation tool to load the stop word library, the thesaurus, and the domain word library to segment the user's question and obtain the segmentation results.

[0060] Specifically, the Chinese stop word library for NLP is downloaded from the GitHub platform on the Internet. Audit experts compile domain-specific terms and their synonyms, which are then converted into domain-specific and synonym-specific lexicons. The stop word library, synonym-specific lexicon, and domain-specific lexicon are loaded using the Jieba word segmentation tool to segment user questions, and the segmentation results are obtained. Then, the dependency parsing model is called to perform intelligent analysis to obtain the intent keyword combination.

[0061] Jieba is a commonly used word segmentation tool in Natural Language Processing (NLP). This tool can segment specified user-defined words by loading a thesaurus / lexicon. It can also segment user-defined questions by loading stop words, synonyms, and domain-specific terms.

[0062] For example:

[0063] User question: How is the fund managed?

[0064] Original breakdown: How are funds / managed / managed?

[0065] Breaking down keywords after loading domain terms: Funds management / is / how / managed?

[0066] Step 3.2: Call the dependency parsing model to perform intelligent analysis on the word segmentation results and obtain the intent keyword combination.

[0067] Dependency parsing is a deep learning model that, through loading, can associate words based on segmentation results. It utilizes dependency relationships between words in a sentence to represent the syntactic structure information of words (such as subject-verb, verb-object, attributive, etc.) and uses a tree structure to represent the structure of the entire sentence (such as subject-verb-object, attributive, adverbial, complement, etc.). Figure 4 As shown. The relation types analyzed by the dependency parsing model are as follows: Figure 5 As shown. Based on the relationships such as ATT / SBV / ADV / HED / POB, write a specified logic program to concatenate / parallelize the word segmentation results, converting 2-3 words into 1 word.

[0068] The dependency parsing model used in this invention is the DDParser open-source model, a Chinese dependency parsing tool developed by Baidu NLP based on large-scale labeled data and the PaddlePaddle deep learning platform. It can help users directly obtain related word pairs, long-distance dependency word pairs, etc. in the input text.

[0069] The specific analysis process of the dependency parsing model is illustrated below:

[0070] Dependency parsing components include 'ATT', 'MT', 'SBV', 'HED', 'VOB', "COO", "ADV", etc. After sentence splitting, data is collected according to HED, SBV, and VOB respectively, resulting in three result sets: HED, SBV, and VOB. 1) When only HED is present, words with "ATT", "ADV", "COO", and "MT" in the remaining components are integrated. 2) When both HED and VOB are present, words with "ATT", "ADV", "COO", "MT", "SBV", "VOB", "IC", and "POB" are integrated. 3) In other cases, words with "ATT", "ADV", "COO", and "MT" are integrated.

[0071] Step 4: Use precise matching and fuzzy matching algorithms to match the combination of intent keywords with the graph data in the Neo4j graph database to obtain candidate entities and attribute sets.

[0072] The design incorporates precise matching and fuzzy matching algorithms to match the intent keyword combinations obtained in step 3 with the graph data in the Neo4j graph database, yielding candidate entities and attribute sets. The matching mechanism primarily consists of two parts: precise matching and fuzzy matching. First, precise matching is performed on the segmentation module results (intent keyword combinations). For correctly segmented results, their meaning can be confirmed by precise matching with the graph data. For incorrectly segmented keywords, a fuzzy matching algorithm (an algorithm combining multiple similarity algorithms) is used to perform fuzzy matching with the graph data to obtain their true meaning. Finally, candidate entities and attribute sets are obtained.

[0073] The precise matching process is as follows:

[0074] The entity attribute words stored in the graph are read and converted into a word list. A program is written to compare the intent keyword with the entity attribute word list stored in the graph. If they are completely identical, they are stored in the candidate entity and attribute set. This invention considers completely identical words to be the candidate entity / attribute to be searched.

[0075] The fuzzy matching process is as follows:

[0076] Fuzzy matching has many algorithms; this invention employs an algorithm based on similarity and edit distance fusion. The input is a word and a list of words to be compared, and the output is the probability (0-1) that the word is similar to the list of words to be compared. A program is written to compare the intent keyword with a list of entity attribute words stored in the graph. Entities with a similarity of 70% or higher are stored in the candidate entity and attribute set. This 70% similarity threshold was adjusted through multiple question-and-answer sessions, and the matching results are used as the candidate entity and attribute set.

[0077] Step 5: Populate the candidate entities and attribute sets into the query template to form a Cypher query statement, and return the query results by querying the Neo4j graph database.

[0078] The query mechanism mainly consists of query statement templates and scenario-based task guidance. Audit experts design multi-task-oriented question-and-answer systems based on audit business scenarios, construct query templates, and then populate the candidate entities and attribute sets from step 4 into the query templates to form Cypher query statements. The results are then returned to the user by querying the Neo4j graph database.

[0079] like Figure 6 As shown, the multi-task orientation of the audit business in this invention mainly includes the following seven directions: audit model direction, audit method direction, audit intermediate table direction, audit field direction, audit issue direction, audit basis direction, and audit suggestion direction.

[0080] (1) Audit model direction refers to the audit rules established by auditors based on specific audit matters related to optimizing the business environment, by setting audit conditions. It is mainly used to verify audit matters. Auditors input keyword combinations and select the listed candidate model names to obtain audit model information related to the model, including the audit intermediate table associated with the audit model, model checking methods, associated audit issue information (issue classification, fact description, audit recommendations, audit basis), and the jump link to the third-party system where the model is located.

[0081] (2) Audit Methodology refers to the various techniques employed by auditors to achieve the audit objective of optimizing the business environment. Auditors input keyword combinations primarily to obtain information from the audit methodology database within the three audit intelligence databases. This includes audit methods, data requirements, data lists, key themes, secondary themes, tertiary themes, division of labor, specific data, audit rules, etc., with a focus on showcasing tertiary themes and audit methodology content.

[0082] (3) The audit intermediate table approach is a digital audit analysis tool built using basic electronic data in the enterprise database, in accordance with the audit requirements for optimizing the business environment. Auditors input keyword combinations to obtain information from the audit intermediate table, combine it with a knowledge graph to obtain the associated audit model, and display the audit model name and the intermediate table name.

[0083] (4) Audit field direction refers to the specific data and type information stored in the audit intermediate table. Auditors input keyword combinations to obtain basic information about the audit fields in the audit intermediate table, and combine this with a knowledge graph to display the associated audit model, audit intermediate table, and audit fields.

[0084] (5) Audit Issues: Auditors identify various issues discovered by enterprises during the process of optimizing their business environment. Auditors can input keyword combinations to obtain basic information from the audit issue database in the three audit intelligence databases, which displays issue classifications, factual descriptions, audit recommendations, and audit basis information.

[0085] (6) The audit basis direction refers to the standards relied upon when auditing the business environment optimization, mainly referring to national and corporate laws, regulations, and rules. Auditors input keyword combinations to obtain audit basis information from the audit question database, and combine it with audit questions associated with the knowledge graph to display the audit questions and audit basis.

[0086] (7) Audit Recommendation Direction refers to the improvement measures proposed by auditors for audit issues related to optimizing the business environment of enterprises. Auditors input keyword combinations to obtain audit recommendation information from the audit issue database, and combine it with the audit issues associated with the knowledge graph to display the audit issues and audit recommendations.

[0087] Figure 6The system displays various audit business directions, and users can select and confirm their intentions through the system interface. For example, selecting "Audit Model" confirms the intention as the audit model direction, and users can continuously confirm the slots based on subsequent options.

[0088] Example of a slot:

[0089] 1. The user enters a question to query the audit model;

[0090] Required slot: Audit model option;

[0091] 2. Users input questions to query audit issues related to the audit model;

[0092] Required slots: Audit model options, Audit model name, Audit issue options.

[0093] After filling in the slots according to the user's selections, write the Cypher query statement. For example, the Cypher query statement for "Audit Issues" in model "A" is:

[0094] "match(n:`model library`)-[r:`model 2 problem`]-(m:`problem library`)where n.name = "A"return distinct m.name"

[0095] That is, query the name of the associated question database in the model database with the name A.

[0096] This invention optimizes the intelligent question-and-answer mechanism for audit guidance on the business environment. It leverages natural language processing technology to understand users' intentions regarding data usage and knowledge graph-based knowledge association to understand the logical relationships between underlying data. Simultaneously, it integrates the experience and knowledge of auditing experts into the constructed application model, comprehensively transforming traditional auditing experience and knowledge. Through a question-and-answer approach, it provides intelligent and convenient guidance for auditors to effectively audit the business environment. This has positive implications for improving audit efficiency and promoting the construction of a smart audit knowledge system based on digital transformation.

[0097] The intelligent question-and-answer mechanism for audit guidance proposed in this invention can also be applied to other audit business scenarios for enterprises. Utilizing the knowledge graph-based intelligent question-and-answer mechanism proposed in this invention, enterprises can achieve the integration of traditional, diverse, and heterogeneous audit knowledge. Auditors can obtain audit work guidance from multiple levels and perspectives through intelligent question-and-answer methods, leveraging the value of various audit knowledge data in guiding auditors. This mechanism plays a key technical support role in improving audit efficiency, resolving trust crises, and enhancing audit real-time performance, further improving audit quality and efficiency, and strengthening the comprehensiveness, scientific rigor, and accuracy of audits.

[0098] Based on the method provided by this invention, this invention also provides an intelligent question-and-answer system for audit guidance based on a knowledge graph, comprising:

[0099] The knowledge graph design module is used to collect audit business data related to optimizing the business environment within enterprises and design the audit knowledge graph structure; the audit business data includes three intelligent audit databases, intermediate tables, and audit model database data;

[0100] The graph database construction module is used to integrate audit business data included in the audit knowledge graph structure through ETL data warehouse technology to build the Neo4j graph database;

[0101] The word segmentation module is used to acquire user questions and segment them into words, perform dependency parsing on the segmentation results, and derive the intent keyword combination.

[0102] The matching module is used to match the combination of intent keywords with the graph data in the Neo4j graph database using precise matching and fuzzy matching algorithms to obtain candidate entities and attribute sets;

[0103] The query module is used to populate the candidate entities and attribute sets into the query template to form a Cypher query statement, and return the query results by querying the Neo4j graph database.

[0104] The modules described in some embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor may be described as including a knowledge graph design module, a graph database construction module, a word segmentation module, a matching module, and a query module. The names of these modules do not necessarily limit the module itself; for example, the knowledge graph design module may also be described as a "module for designing knowledge graph structures."

[0105] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0106] Figure 7 The process of using the knowledge graph-based intelligent question-answering system for audit guidance according to this invention to process natural language questions is illustrated, such as... Figure 7 As shown, the specific process includes:

[0107] S1: The user inputs a natural language question. The Jieba word segmentation tool first loads the stop word library to filter out meaningless stop words, then loads the domain word library and the thesaurus to accurately segment the question and obtain preliminary segmentation results; then loads the dependency parsing model to perform sentence structure and grammatical analysis on the segmentation results and transform them into intent keyword combinations.

[0108] S2: The intent keyword combination from the previous step is precisely matched with the graph data to obtain a partial set of confirmed entities and attributes. Then, fuzzy matching is performed. The fuzzy matching algorithm consists of three algorithms: LSA, LDA, and TF-IDF. The weights of the three algorithms are dynamically allocated according to the length to solve the problem that the matching results of a single similarity algorithm do not conform to the actual results in ultra-long texts. The matching results scores of the three algorithms are recalculated according to the weights, and the final candidate entity and attribute set is obtained by summarizing them.

[0109] The following are examples illustrating three fuzzy matching algorithms: LSA, LDA, and TF-IDF.

[0110] The intent keyword combination includes the word "funds check". The word length is 3. The weights of LSA, LDA and TF-IDF are set to 0.3, 0.3 and 0.4 respectively. The fuzzy match results are scored according to this weight. The overall results are summed and the result ranked first is taken as the final match result, which is "funds check".

[0111] The LSA, LDA, and TF-IDF algorithms are dynamically weighted according to length as follows:

[0112] For lengths up to 10, set the weights to 0.3, 0.3, and 0.4.

[0113] Length is 10-20, weight is set to 0.4, 0.4, 0.2;

[0114] For lengths of 20 or more, the weights are set to 0.6, 0.2, and 0.2.

[0115] S3: Based on the task orientation of the audit business scenario, fill the candidate entities and attribute sets into the query template to form a Cypher query statement, and obtain the answer to the question from the query graph.

[0116] Through the above process, an intelligent question-and-answer system for optimizing audit guidelines for the business environment was implemented. The question-and-answer data originates from a knowledge graph knowledge base resulting from data fusion. Based on specific business scenarios and task-oriented approaches, it provides intelligent guidance to auditors, significantly reducing the time auditors spend obtaining audit information, improving audit efficiency, and enhancing audit quality.

[0117] Furthermore, the present invention also provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor can invoke a computer program stored in the memory to execute the aforementioned knowledge graph-based audit guidance intelligent question-answering method.

[0118] Furthermore, when the computer program in the aforementioned memory is implemented as a software functional unit and sold or used as an independent product, it can be stored in a non-transitory computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0119] This invention proposes a knowledge graph-based intelligent question-and-answer method, system, and device for audit guidance. It outlines a holistic approach of "graph design - data import - intelligent question answering," integrating tabular data and internal database data related to business environment optimization audits through graph structure design and ETL data warehousing technologies. Furthermore, it incorporates a multi-task-oriented design tailored to audit business scenarios to construct an intelligent question-and-answer system. This system provides intelligent and convenient guidance for auditors in effectively auditing the business environment, significantly contributing to improved audit efficiency and promoting the construction of a smart audit knowledge system based on digital transformation.

[0120] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0121] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A knowledge graph-based intelligent question-answering method for audit guidance, characterized in that, include: We collect audit business data related to optimizing the business environment within enterprises and design an audit knowledge graph structure. Specifically, the audit business data is divided into five types of entities: model library, method library, question library, fields, and intermediate tables. The fields in each entity table are designed as attributes of the entity, and the business relationships between the entities are designed as edges, thus constructing the audit knowledge graph structure. The audit business data includes data from the three intelligent audit databases, intermediate tables, and audit model database. By integrating audit business data included in the audit knowledge graph structure through ETL data warehouse technology, a Neo4j graph database is constructed. The process of obtaining user questions and performing word segmentation includes: using the jieba word segmentation tool to load stop word library, thesaurus and domain word library to segment user questions and obtain word segmentation results; The thesaurus and domain thesaurus consist of specialized vocabulary from the audit business domain, and their purpose is to help the word segmentation tool correctly split professional domain terms. Dependency parsing is performed on the word segmentation results to derive the intended keyword combination; Using precise matching and fuzzy matching algorithms, the intent keyword combination is matched against graph data in the Neo4j graph database to obtain candidate entities and attribute sets, including: For keywords that are not correctly split in the intent keyword combination, use the fuzzy matching algorithm to read the graph data in the Neo4j graph database, write a program to compare the intent keywords with the entity attribute word list stored in the graph, and store those with a similarity of 70% into the candidate entity and attribute set; The fuzzy matching algorithm consists of three algorithms: LSA, LDA, and TF-IDF, with weights dynamically allocated according to the length. The query template is constructed using a multi-task orientation designed in conjunction with audit business scenarios; the multi-task orientation includes audit model orientation, audit method orientation, audit intermediate table orientation, audit field orientation, audit issue orientation, audit basis orientation, and audit recommendation orientation. The candidate entities and attribute sets are populated into the query template to form a Cypher query statement, which returns the query results by querying the Neo4j graph database.

2. The intelligent question-answering method for audit guidance based on knowledge graphs according to claim 1, characterized in that, The process involves integrating audit business data from the audit knowledge graph structure using ETL data warehouse technology to construct a Neo4j graph database, specifically including: The audit business data included in the audit knowledge graph structure is extracted, cleaned and transformed using ETL data warehouse technology. Then, Cypher statements are written to store the data in the Neo4j graph database as graph data, thus constructing the Neo4j graph database.

3. The intelligent question-answering method for audit guidance based on knowledge graphs according to claim 1, characterized in that, The process of acquiring user questions and performing word segmentation, then performing dependency parsing on the segmentation results to derive intent keyword combinations, specifically includes: The dependency parsing model is invoked to intelligently analyze the word segmentation results and derive the intent keyword combination.

4. The intelligent question-answering method for audit guidance based on knowledge graphs according to claim 1, characterized in that, The process of using precise matching and fuzzy matching algorithms to match intent keyword combinations with graph data in the Neo4j graph database to obtain candidate entities and attribute sets specifically includes: For correctly split keywords in the intent keyword combination, the graph data in the Neo4j graph database is read using the exact matching algorithm. The entity attribute words stored in the graph are converted into a word list. A program is written to compare the intent keywords with the entity attribute word list stored in the graph. Those that are exactly the same are stored in the candidate entity and attribute set.

5. A knowledge graph-based intelligent question-and-answer system for audit guidance, characterized in that, include: The knowledge graph design module is used to collect audit business data related to optimizing the business environment within the enterprise and design the audit knowledge graph structure. Specifically, the audit business data is divided into five types of entities: model library, method library, question library, fields and intermediate tables. The fields in each entity table are designed as the attributes of the entity, and the business relationships between the entities are designed as edges to construct the audit knowledge graph structure. The audit business data includes data from the three intelligent audit databases, intermediate tables, and audit model database. The graph database construction module is used to integrate audit business data included in the audit knowledge graph structure through ETL data warehouse technology to build the Neo4j graph database; The word segmentation module is used to obtain user questions and perform word segmentation, specifically including: The user's question was segmented using the jieba word segmentation tool by loading a stop word library, a thesaurus, and a domain word library, and the segmentation results were obtained. The thesaurus and domain thesaurus consist of specialized vocabulary from the audit business domain, and their purpose is to help the word segmentation tool correctly split professional domain terms. Dependency parsing is performed on the word segmentation results to derive the intended keyword combination; The matching module uses precise matching and fuzzy matching algorithms to match intent keyword combinations with graph data in the Neo4j graph database, obtaining candidate entities and attribute sets, including: For keywords that are not correctly split in the intent keyword combination, use the fuzzy matching algorithm to read the graph data in the Neo4j graph database, write a program to compare the intent keywords with the entity attribute word list stored in the graph, and store those with a similarity of 70% into the candidate entity and attribute set; The fuzzy matching algorithm consists of three algorithms: LSA, LDA, and TF-IDF, with weights dynamically allocated according to the length. The query template is constructed using a multi-task orientation designed in conjunction with audit business scenarios; the multi-task orientation includes audit model orientation, audit method orientation, audit intermediate table orientation, audit field orientation, audit issue orientation, audit basis orientation, and audit recommendation orientation. The query module is used to populate the candidate entities and attribute sets into the query template to form a Cypher query statement, and return the query results by querying the Neo4j graph database.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the knowledge graph-based intelligent question-answering method for audit guidance as described in any one of claims 1 to 4.

7. The electronic device according to claim 6, characterized in that, The memory is a non-transitory computer-readable storage medium.

Images