Project attribute data intelligent auditing method and system

Abstract

The present application relates to the technical field of data governance and artificial intelligence, in particular to a project attribute data intelligent auditing method and system, the method comprising: constructing a domain knowledge base containing dictionaries of project regions, types and the like, and a polysemy and synonym mapping table; obtaining a project name text to be audited, using a hybrid semantic analysis model based on the knowledge base to extract dimension information such as project regions, years and confidence levels; obtaining corresponding structured attribute values; arbitrating attribute missing, errors or to-be-confirmed through a rule engine combined with confidence levels to determine, and outputting results containing problem dimensions, suggested values and the like; generating work order update data, and feeding back effective mapping optimization of the knowledge base. The system comprises knowledge base management, semantic analysis, data interface, rule engine, closed loop processing and visual display modules, and each module cooperates to realize intelligent auditing. The present application can improve the intelligence and automation level of auditing, realize self-learning, reduce labor costs, and effectively guarantee project data quality.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data governance and artificial intelligence, and particularly relates to a project attribute data intelligent auditing method and system. BACKGROUND

[0002] With the deepening of the process of enterprise digital transformation, the project data accumulated in the long-term operation of core business systems such as ERP and PMS has gradually become an important strategic asset supporting enterprise business decision-making, resource allocation and development planning. However, the current enterprise project data quality is generally facing severe challenges, mainly in two dimensions: on the one hand, as a typical unstructured text, the writing process of the project name has a strong randomness, and there are frequent abbreviations and ambiguous expressions, such as "Inner Mongolia X Pumped Storage Feasibility Study" and "Shanghai 20235G Macro Station Construction", which makes it difficult to directly extract key information from them; on the other hand, the structured attribute fields corresponding to the project name (covering the project location, project year, project phase, industry field and other core dimensions), often have problems such as missing, wrong filling and inconsistency with the name description, further exacerbating the data usability problem.

[0003] In order to improve the above problems, three main data auditing methods have been formed in the industry, but all have obvious limitations: first, manual spot check method can cover part of the key data, but has low efficiency and is prone to data omission due to human negligence, and long-term investment has high labor cost; second, script verification based on simple rules (such as keyword matching, regular expression) lacks flexibility and cannot understand the semantics, making it difficult to handle equivalent expressions such as "pumped storage" and "pumped storage", "feasibility study" and "feasibility study phase", and cannot handle ambiguous reference scenarios such as "X area", resulting in high false positive and false negative rates of data auditing; third, existing general NLP tools are limited by the generality of model training data, lack of industry knowledge related to enterprise projects, and have low accuracy in identifying specific domain entities such as "project type" and "project phase", and can only extract basic information, cannot perform deep comparison and auditing decision with project structured attribute fields, and cannot meet the precise control needs of enterprise project data quality.

[0004] Therefore, the present application is proposed. SUMMARY

[0005] In order to solve the above technical problems in the prior art, the present application provides a project attribute data intelligent auditing method and system, which can not only extract multi-dimensional information from unstructured project names with high precision, but also intelligently judge the quality problems of attribute data through rule engine and confidence mechanism, and can trigger a closed-loop processing flow to automatically recommend or complete data repair, thereby realizing the continuous autonomy and optimization of data quality.

[0006] To achieve the above object, the technical scheme of the present application is as follows:

[0007] In a first aspect, an intelligent auditing method for project attribute data comprises:

[0008] S1, a domain knowledge base is constructed, which comprises a project area dictionary, a project type dictionary, a customer department dictionary, a project stage dictionary, and a polysemy mapping relationship table and a synonym mapping relationship table;

[0009] S2, a project name text of a project to be audited is obtained, and a semantic analysis model based on the domain knowledge base is used to process the project name text, extract at least one extraction value of a dimension, and calculate the confidence of each extraction value; the dimension includes project area, project year, project type, installed capacity, and project stage;

[0010] S3, structured attribute data corresponding to the project to be audited is obtained from a project database, and attribute values corresponding to the dimensions in step S2 are obtained;

[0011] S4, the extraction value, the confidence of the extraction value, and the corresponding attribute value are input into a preset rule engine for comparison and auditing; the execution logic of the rule engine comprises:

[0012] If the attribute value is empty and the confidence of the extraction value is higher than a first preset threshold, it is determined that the attribute of this dimension is missing, and the extraction value is output as a high-confidence suggestion value;

[0013] If the attribute value and the extraction value are both non-empty, and it is determined according to the domain knowledge base that the attribute value and the extraction value are semantically inconsistent, a confidence arbitration mechanism is started: if the confidence of the extraction value is higher than a second preset threshold and higher than the confidence of the attribute value, it is determined that the attribute value is incorrect, and the extraction value is output as a suggestion value; if the confidence of the extraction value is lower than the second preset threshold, a confirmation signal is output;

[0014] S5, an auditing result is output, which includes problem dimension, problem type, attribute original value, suggestion value, and confidence.

[0015] Further, it further comprises:

[0016] S6, for the auditing result of the output suggestion value, a data modification suggestion work order is generated, which is automatically updated to the project database after approval; at the same time, the effective mapping relationship in this audit is fed back to the domain knowledge base to optimize the semantic analysis accuracy of the subsequent.

[0017] Further, the project area dictionary includes province, city, and county level regional entries, the project type dictionary contains industry sub-type entries such as hydropower generation and 5G construction, and the project phase dictionary contains corresponding standard phase entries such as planning, feasibility study, and approval.

[0018] Further, the semantic parsing model based on the domain knowledge base in step S2 adopts a hybrid semantic parsing method, which includes:

[0019] S21, dictionary and rule matching: performing word segmentation processing on the project name text, performing accurate matching between the word segmentation result and the entries in the domain knowledge base, if the matching is successful, returning a standard value and assigning the highest confidence;

[0020] S22, fine-tuning NER model extraction: using a named entity recognition model fine-tuned on a large amount of project corpus to deeply parse the segmented project name text, outputting entity segments and their probability scores, which are the main basis for the confidence of the corresponding extraction value;

[0021] S23, post-processing and standardization: cleaning and converting the extraction results of steps S21 and S22; for fuzzy results containing "X area", assigning a basic confidence according to the preset rules;

[0022] S24, confidence fusion: if the same dimension entity is recognized by steps S21 and S22 at the same time, taking the highest confidence as the final confidence of the dimension extraction value.

[0023] Further, the highest confidence is 0.99, the named entity recognition model is a BERT-BiLSTM-CRF model, and the basic confidence of the fuzzy result is 0.7.

[0024] Further, the first preset threshold is 0.6 and the second preset threshold is 0.85.

[0025] The determination that the attribute value and the extraction value are semantically inconsistent according to the domain knowledge base includes that the attribute value and the extraction value are non-synonymous words and non-hierarchical relationship.

[0026] Second aspect, a project attribute data intelligent auditing system, comprising:

[0027] Knowledge base management module: for storing domain knowledge data, and providing knowledge support for the semantic parsing module and the rule engine module, while receiving external feedback data to update the domain knowledge data stored by itself;

[0028] A semantic analysis module: based on the domain knowledge data provided by the knowledge base management module, the project data transmitted by the data interface module is subjected to semantic analysis, and the extracted value and the corresponding confidence are output, and the extracted value and the confidence are transmitted to the rule engine module;

[0029] A data interface module: used for acquiring relevant data of a project to be audited, and transmitting to the semantic analysis module and the rule engine module respectively;

[0030] A rule engine module: used for receiving another part of the project data transmitted by the data interface module and the extracted value and the confidence transmitted by the semantic analysis module, executing a preset comparison logic to make an audit judgment, outputting an audit result, and transmitting the audit result to the closed-loop processing module and the visual display module respectively;

[0031] A closed-loop processing module: used for performing a closed-loop processing operation based on the audit result output by the rule engine module, and feeding back the effective data generated in the processing process to the knowledge base management module;

[0032] A visual display module: used for visualizing based on the audit result output by the rule engine module.

[0033] Further, the domain knowledge data stored by the knowledge base management module includes a project area dictionary, a project type dictionary, a customer department dictionary, a project phase dictionary, and a polysemy mapping relationship table and a synonym mapping relationship table.

[0034] The knowledge base management module includes: supporting manual addition or deletion of entries in each dictionary, receiving effective mapping relationships fed back by the closed-loop processing module, and supplementing the effective mapping relationships to the polysemy mapping relationship table or the synonym mapping relationship table to update the domain knowledge data.

[0035] Further, the semantic analysis module is built-in with a customized named entity recognition model, and the named entity recognition model is a BERT-BiLSTM-CRF model fine-tuned based on a large amount of project corpus.

[0036] The semantic analysis process of the semantic analysis module adopts a hybrid processing method, including: performing word segmentation on the received project text, performing accurate matching with the entries in the knowledge base management module, performing entity extraction through the BERT-BiLSTM-CRF model, cleaning and converting the extraction result, and fusing the confidence of the same entity extracted in multiple ways.

[0037] Further, the project text content that can be processed by the semantic analysis module through the hybrid processing method includes:

[0038] Abbreviations, ambiguous references and values with units in the project name; among them, the extraction value output by the exact match is given the highest confidence 0.99, the corresponding extraction value of the ambiguous reference is given the basic confidence 0.7, and the value with units is converted into a pure digital form.

[0039] Compared with the prior art, the above-mentioned project attribute data intelligent auditing method and system provided by the application, the method comprises the following steps: constructing a domain knowledge base containing a project area, a type and the like dictionary and a polysemy, synonym mapping table; obtaining a to-be-audited project name text, extracting dimension information such as a project area, a year and the like and a confidence by using a hybrid semantic analysis model based on the knowledge base; obtaining corresponding structured attribute values; determining attribute absence, error or to-be-confirmed through a rule engine combined with the confidence arbitration, and outputting a result containing a problem dimension, a suggested value and the like; generating a work order update data, and feeding back an effective mapping optimization knowledge base. The system comprises a knowledge base management, a semantic analysis, a data interface, a rule engine, a closed-loop processing and a visual display module, and each module cooperates to realize intelligent auditing. The application can improve the intelligent and automatic level of auditing, realize self-learning, reduce labor cost, and effectively guarantee the project data quality. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The flowchart of the intelligent auditing method provided for the embodiments of the application;

[0041] Figure 2 The architecture diagram of the intelligent auditing system provided for the embodiments of the application. DETAILED DESCRIPTION

[0042] The technical solutions of the application will be described clearly below with the aid of the accompanying drawings. Obviously, the described embodiments are not all the embodiments of the application, and all other embodiments obtained by those of ordinary skill in the art without any creative labor fall within the protection scope of the application.

[0043] It should be noted that, unless otherwise specified, the relative arrangement of the components and steps, numerical expressions described in these embodiments should not be understood as limiting the scope of the application.

[0044] The following description of exemplary embodiments is merely illustrative in nature and is in no way intended to limit the application or its application or use in any way. Here, the technologies, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but in the case of applicable technologies, methods and devices, these technologies, methods and devices should be considered as part of this specification.

[0045] Embodiment one

[0046] Reference Figure 1 , Figure 1A flowchart of an intelligent auditing method for project attribute data is provided, and the specific steps can include:

[0047] S1, a domain knowledge base is constructed, which includes a project area dictionary, a project type dictionary, a customer department dictionary, a project phase dictionary, and a polysemy mapping relationship table and a synonym mapping relationship table;

[0048] The project area dictionary includes provincial, municipal, and county-level geographical entries, the project type dictionary includes industry sub-type entries such as hydropower generation and 5G construction, and the project phase dictionary includes standard phase entries corresponding to the stages of preparation, feasibility study, planning, and approval.

[0049] S2, the project name text of the project to be audited is obtained, and a semantic analysis model based on the domain knowledge base is used to process the project name text, extract at least one dimension information extraction value, and calculate the confidence of each extraction value; the dimensions include project area, project year, project type, installed capacity, and project phase;

[0050] The semantic analysis model based on the domain knowledge base adopts a hybrid semantic analysis method, which includes:

[0051] S21, dictionary and rule matching: the project name text is processed by word segmentation, and the word segmentation result is accurately matched with the entries in the domain knowledge base, if the matching is successful (such as "pumping" matching "pumped storage power station"), the standard value is returned, and the highest confidence (0.99) is given;

[0052] S22, fine-tuning NER model extraction: using a named entity recognition model (BERT-BiLSTM-CRF) fine-tuned on a large amount of project corpus, the segmented project name text is deeply analyzed, and the entity segment and its probability score are output, the probability score is the main basis for the confidence of the corresponding extraction value;

[0053] S23, post-processing and standardization: the extraction results of steps S21 and S22 are cleaned and converted (such as "1200MW" to the number 1200); for ambiguous results (such as "X area"), a basic confidence (0.7) is given according to the preset rules;

[0054] S24, confidence fusion: if the same dimension entity is recognized by steps S21 and S22 at the same time, the highest confidence is taken as the final confidence of the dimension extraction value.

[0055] S3, the structured attribute data corresponding to the project to be audited is obtained from the project database, and the attribute values corresponding to the dimensions in step S2 are obtained;

[0056] S4, input the extracted value, the confidence of the extracted value and the corresponding attribute value into a preset rule engine for comparison and auditing; the execution logic of the rule engine comprises:

[0057] If the attribute value is empty and the confidence of the extracted value is higher than a first preset threshold, it is determined that the attribute of the dimension is missing, and the extracted value is output as a high-confidence suggestion value;

[0058] If the attribute value and the extracted value are both not empty, and it is determined according to the domain knowledge base that the attribute value and the extracted value are semantically inconsistent, a confidence arbitration mechanism is started: if the confidence of the extracted value is higher than a second preset threshold and higher than the confidence of the attribute value, it is determined that the attribute value is wrong, and the extracted value is output as a suggestion value; if the confidence of the extracted value is lower than the second preset threshold, a confirmation signal is output;

[0059] The first preset threshold is 0.6, and the second preset threshold is 0.85; determining that the attribute value and the extracted value are semantically inconsistent according to the domain knowledge base comprises: the attribute value and the extracted value are not synonymous words, and are not in a hierarchical relationship

[0060] S5, output the auditing result, the auditing result comprising the problem dimension, the problem type, the attribute original value, the suggestion value and the confidence.

[0061] S6, for the auditing result of the output suggestion value, a data modification suggestion work order is generated, which is automatically updated to the project database after approval; at the same time, the effective mapping relationship in this auditing is fed back to the domain knowledge base, so as to optimize the semantic analysis accuracy in the future.

[0062] Embodiment two

[0063] Referring to Figure 2 , Figure 2 The present application provides a kind of project attribute data intelligent auditing system architecture diagram, comprising:

[0064] M1, knowledge base management module: for storing domain knowledge data, and for the semantic analysis module, rule engine module provides knowledge support, simultaneously receives external feedback data to update the domain knowledge data stored in itself;

[0065] The domain knowledge data stored in the knowledge base management module comprises a project area dictionary, a project type dictionary, a customer department dictionary, a project phase dictionary and a polysemy mapping relationship table, and a synonym mapping relationship table.

[0066] The knowledge base management module comprises: supporting manual addition or deletion of entries in each dictionary, receiving the valid mapping relationship fed back by the closed loop processing module, and supplementing the valid mapping relationship to the polysemy mapping relationship table or the synonym mapping relationship table to update the domain knowledge data.

[0067] M2, semantic analysis module: based on the domain knowledge data provided by the knowledge base management module, performing semantic analysis on the project data transmitted by the data interface module, outputting the extracted value and the corresponding confidence, and transmitting the extracted value and the confidence to the rule engine module;

[0068] The semantic analysis module is built-in with a customized named entity recognition model, which is a BERT-BiLSTM-CRF model fine-tuned on a large amount of project corpus.

[0069] The semantic analysis process of the semantic analysis module adopts a hybrid processing method, including: performing word segmentation on the received project text, performing accurate matching with the entries in the knowledge base management module, performing entity extraction through the BERT-BiLSTM-CRF model, cleaning and converting the extraction results, and fusing the confidence of the same entity extracted in multiple ways.

[0070] The project text content that can be processed by the semantic analysis module through the hybrid processing method includes:

[0071] Abbreviated terms, ambiguous references and numerical values with units in the project name; wherein the extracted value output when the accurate matching is successful is assigned a highest confidence of 0.99, the extracted value corresponding to the ambiguous reference is assigned a basic confidence of 0.7, and the numerical value with units is converted into a pure digital form.

[0072] M3, data interface module: for obtaining related data of the project to be audited, and transmitting to the semantic analysis module and the rule engine module respectively;

[0073] M4, rule engine module: for receiving another part of the project data transmitted by the data interface module and the extracted value and confidence transmitted by the semantic analysis module, performing a preset comparison logic to make an auditing judgment, outputting an auditing result, and transmitting the auditing result to the closed loop processing module and the visual display module respectively;

[0074] M5, closed loop processing module: for performing closed loop processing operation based on the auditing result output by the rule engine module, and feeding back the valid data generated in the processing process to the knowledge base management module;

[0075] M6, visual display module: for visualizing the auditing result output by the rule engine module.

[0076] Example three

[0077] This embodiment adopts the project attribute data intelligent auditing method proposed in the application, and intelligently audits the pumped storage power station project (installed capacity auditing) as the background, which specifically includes:

[0078] The input project name is: "Zhejiang Ninghai 1200MW pumped storage power station project (preparation)"

[0079] The original attribute in the database is: {project location: "Zhejiang Province Ningbo City", total installed capacity: 1000, construction state: "approved"}

[0080] Workflow:

[0081] Semantic analysis:

[0082] Extract "project area": Zhejiang Province Ningbo City Ninghai County (confidence 0.99);

[0083] Extract "installed capacity": 1200 (confidence 1.0);

[0084] Extract "industry domain": hydropower / pumped storage (confidence 0.98);

[0085] Extract "project phase": preparation phase (confidence 0.95).

[0086] Rule engine comparison:

[0087] Project location: the attribute value Zhejiang Province Ningbo City and the extracted value Ninghai County are in a superior-inferior relationship, and the judgment is consistent.

[0088] Total installed capacity: attribute value 1000 ≠ extracted value 1200. The extracted value confidence (1.0) > second threshold (0.85). Judgment: attribute value is wrong. Output suggestion value: 1200.

[0089] Construction state: attribute value approved ≠ extracted value preparation phase. The extracted value confidence (0.95) > second threshold (0.85). Judgment: attribute value to be confirmed.

[0090] Closed loop processing: the system prompts to modify the installed capacity and confirm the construction state. The administrator updates the capacity to 1200 and updates the state to "preparation phase". The system feeds back the mapping relationship of "preparation" -> "preparation phase" to the knowledge base.

[0091] Example four

[0092] This embodiment adopts the project attribute data intelligent auditing method proposed in the application, and intelligently audits the pumped storage power station project (fuzzy information and phase auditing) as the background, which specifically includes:

[0093] Input project name: "Feasibility study of pumped storage power station in X region of Inner Mongolia"

[0094] Original attributes in database: {Project location: null, Project type: "Hydropower project", Project stage: "Planning"}

[0095] Workflow:

[0096] Semantic analysis (S2):

[0097] Extract "project region": Inner Mongolia X region -> (Cannot accurately match, inferred as Inner Mongolia Autonomous Region) (Confidence 0.75 - Lower, as "X region" is a vague reference)

[0098] Extract "project type": Pumped storage -> (Map through knowledge base synonyms) Pumped storage power station (Confidence 0.98)

[0099] Extract "project stage": Feasibility study -> (Map through knowledge base stages) Feasibility study stage (Confidence 0.97)

[0100] Rule engine comparison (S4):

[0101] Project location: Attribute value is null, extracted value Inner Mongolia Autonomous Region has a confidence of 0.75 > first threshold (0.6). Decision: Missing. Output suggested value: Inner Mongolia Autonomous Region (But marked with medium confidence, need human attention).

[0102] Project type: Attribute value Hydropower project ≠ extracted value Pumped storage power station. Extracted value confidence 0.98 > second threshold (0.85). Decision: Attribute value is wrong. Output suggested value: Pumped storage power station.

[0103] Project stage: Attribute value Planning ≠ extracted value Feasibility study stage. Extracted value confidence 0.97 > second threshold (0.85). Decision: Attribute value is wrong. Output suggested value: Feasibility study stage.

[0104] Closed-loop processing (S6):

[0105] System generates an audit report, strongly suggests modifying project type and stage, and prompts that "project location" information is incomplete, suggests supplementing.

[0106] Administrator adopts all suggestions: changes type to "pumped storage power station", stage to "feasibility study", and supplements location to "Inner Mongolia Autonomous Region, Ulanqab City" according to actual information.

[0107] Feedback learning: the system adds the strong mapping relationship of this time confirmed "pumped storage" -> pumped storage power station, "feasibility study" -> feasibility study stage to the knowledge base. At the same time, the information that "X area" is finally determined as "Ulanqab City" is also learned into the database. In the future, when encountering similar "X area" expressions, "Ulanqab City" can be provided as a potential suggestion value.

[0108] In summary, the present application has the following advantages:

[0109] 1. Through the multi-layer technical architecture combining semantic analysis, customized rule engine and confidence arbitration, the project data quality problems can be deeply excavated and intelligently judged, effectively solving the semantic consistency problems that cannot be handled by traditional auditing methods such as manual sampling, simple rule script verification and general NLP tools, especially accurately processing abbreviations in project names, greatly improving the auditing accuracy and applicability;

[0110] 2. The present application solves the limitations of existing auditing tools that can only find problems, innovatively introduces a complete data governance closed loop consisting of problem finding, scheme suggestion, approval execution and feedback optimization, extends the project data governance process from a single auditing link to data repair and system optimization link, realizes the automation and systematization of data governance, significantly reduces the cost of manual intervention, and improves the efficiency of data governance;

[0111] 3. Relying on the closed loop feedback mechanism, the system can actively absorb the correct results of manual confirmation and supplement the effective mapping relationship to the domain knowledge base, continuously enrich the content of the knowledge base; with the progress of the use process, the semantic analysis accuracy and auditing accuracy of the system can be gradually improved, forming a positive optimization cycle of data governance effect, realizing the self-learning effect of getting smarter with use;

[0112] 4. Confidence-based arbitration mechanism. This mechanism provides reliable judgment basis for the decision-making behavior of the machine in project attribute data auditing, effectively solves the problem of how the computer simulates human thinking to judge data quality, and provides key technical support for machine decision-making of project data intelligent auditing.

[0113] The above specific embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the examples, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the scope of the technical solutions of the present application, which should be covered by the scope of the claims of the present application.

Claims

1. A method for intelligent auditing of project attribute data, characterized by, The method comprises the following steps: S1, constructing a domain knowledge base, which comprises a project area dictionary, a project type dictionary, a customer department dictionary, a project stage dictionary, and a polysemy mapping table and a synonym mapping table; S2, obtaining a project name text of a to-be-audited project, processing the project name text by using a semantic analysis model customized based on the domain knowledge base, extracting at least one extracted value of dimension information, and calculating a confidence degree of each extracted value; the dimensions comprise a project area, a project year, a project type, a installed capacity, and a project stage; The semantic analysis model customized based on the domain knowledge base adopts a hybrid semantic analysis method, which comprises the following steps: S21, dictionary and rule matching: performing word segmentation processing on the project name text, performing accurate matching between a word segmentation result and a word in the domain knowledge base, returning a standard value if the matching is successful, and giving the highest confidence degree; S22, fine-tuning of a NER model for extraction: using a named entity recognition model fine-tuned based on a large amount of project corpus to deeply analyze the word segmented project name text, outputting an entity segment and a probability score, and taking the probability score as a main basis for the confidence degree of the corresponding extracted value; S23, post-processing and standardization: cleaning and converting the extracted results of steps S21 and S22; for a fuzzy result containing an "X area", a basic confidence degree is given according to a preset rule; S24, confidence degree fusion: if an entity of the same dimension is recognized by steps S21 and S22 at the same time, the highest confidence degree is taken as the final confidence degree of the extracted value of the dimension; S3, obtaining structured attribute data corresponding to the to-be-audited project from a project database, and obtaining attribute values corresponding to the dimensions in step S2; S4, inputting the extracted values, the confidence degrees of the extracted values, and the corresponding attribute values into a preset rule engine for comparison and auditing; the execution logic of the rule engine comprises: If the attribute value is empty and the confidence degree of the extracted value is higher than a first preset threshold, it is determined that the attribute of the dimension is missing, and the extracted value is output as a high-confidence suggestion value; If the attribute value and the extracted value are both not empty, and it is determined according to the domain knowledge base that the attribute value and the extracted value are semantically inconsistent, a confidence degree arbitration mechanism is started: if the confidence degree of the extracted value is higher than a second preset threshold and higher than the confidence degree of the attribute value, it is determined that the attribute value is wrong, and the extracted value is output as a suggestion value; if the confidence degree of the extracted value is lower than the second preset threshold, a to-be-confirmed signal is output; S5, outputting an auditing result, which comprises a problem dimension, a problem type, an attribute original value, a suggestion value, and a confidence degree.

2. The method of claim 1, wherein, Further comprising: S6, for the auditing result of the output suggestion value, generating a data modification suggestion work order, which is automatically updated to a project database after being approved; meanwhile, feeding back effective mapping relationships in this auditing to the domain knowledge base, so as to optimize the semantic analysis accuracy in the future.

3. The method of claim 1, wherein, The project area dictionary includes province, city, and county level regional entries, the project type dictionary contains industry sub-type entries such as hydropower generation and 5G construction, and the project phase dictionary contains corresponding standard phase entries such as planning, feasibility study, and approval.

4. The method of claim 1, wherein, The highest confidence is 0.99; the named entity recognition model is a BERT-BiLSTM-CRF model; and the basic confidence of the fuzzy result is 0.

7.

5. The method of claim 1, wherein, The first preset threshold is 0.6, and the second preset threshold is 0.

85. The determination that the attribute value and the extracted value are semantically inconsistent according to the domain knowledge base includes that the attribute value and the extracted value are non-synonymous words or non-hierarchical relationship.

6. An intelligent audit system for project attribute data, characterized by, It comprises: The knowledge base management module is used to store domain knowledge data and provide knowledge support for the semantic analysis module and the rule engine module, and also receives external feedback data to update the stored domain knowledge data; The semantic analysis module performs semantic analysis on the project data transmitted by the data interface module based on the domain knowledge data provided by the knowledge base management module, outputs the extracted value and the corresponding confidence, and transmits the extracted value and the confidence to the rule engine module; The semantic analysis module adopts a hybrid semantic analysis method, which comprises: S21, dictionary and rule matching: performing word segmentation processing on the project name text, and performing accurate matching between the word segmentation result and the entries in the domain knowledge base, if the matching is successful, returning the standard value and assigning the highest confidence; S22, fine-tuning NER model extraction: using a named entity recognition model fine-tuned on a large amount of project corpus to deeply analyze the segmented project name text, outputting entity fragments and their probability scores, which are the main basis for the confidence of the corresponding extracted value; S23, post-processing and standardization: cleaning and converting the extracted results of steps S21 and S22; for fuzzy results containing "X area", a basic confidence is assigned according to a preset rule; S24, confidence fusion: if the same dimension entity is recognized by steps S21 and S22 at the same time, the highest confidence is taken as the final confidence of the extracted value of this dimension; The data interface module is used to obtain relevant data of the project to be audited and transmit them to the semantic analysis module and the rule engine module respectively; The rule engine module is used to receive another part of the project data transmitted by the data interface module and the extracted value and confidence transmitted by the semantic analysis module, execute a preset comparison logic to perform auditing and judgment, output the auditing result, and transmit the auditing result to the closed-loop processing module and the visual display module respectively; The closed-loop processing module is used to perform closed-loop processing operations based on the auditing result output by the rule engine module, and feed back the effective data generated in the processing process to the knowledge base management module; The visual display module is used to visually present the auditing result output by the rule engine module.

7. The project attribute data intelligent auditing system of claim 6, wherein, The domain knowledge data stored by the knowledge base management module includes a project area dictionary, a project type dictionary, a customer department dictionary, a project phase dictionary, and a one-word polysemy mapping relationship table and a synonym mapping relationship table. The knowledge base management module comprises: supporting manual addition or deletion of entries in each dictionary, receiving the effective mapping relationship fed back by the closed loop processing module, and supplementing the effective mapping relationship to the polysemy mapping relationship table or the synonym mapping relationship table to update the field knowledge data.

8. The project attribute data intelligent auditing system of claim 6, wherein, The semantic analysis module is built-in with a customized named entity recognition model, which is a BERT-BiLSTM-CRF model fine-tuned on a large project corpus. The semantic analysis process of the semantic analysis module adopts a hybrid processing method, including: performing word segmentation on the received project text, performing accurate matching with the entries in the knowledge base management module, performing entity extraction through the BERT-BiLSTM-CRF model, cleaning and converting the extraction results, and fusing the confidence of the same entity extracted in multiple ways.

9. The project attribute data intelligent auditing system of claim 6, wherein, The project text content that can be processed by the semantic analysis module through the hybrid processing method includes: Abbreviated terms, ambiguous references, and numerical values with units in project names; wherein the extraction value output when the accurate matching is successful is assigned a highest confidence of 0.99, the extraction value corresponding to the ambiguous reference is assigned a basic confidence of 0.7, and the numerical value with units is converted into a pure numerical form.

Images