Method and system for automatically generating budgeting and execution report based on AI reasoning

Abstract

The invention discloses a budgeting and execution report automatic generation method and system based on AI reasoning, and particularly relates to the technical field of artificial intelligence. The method comprises the following steps: constructing a semantic feature data set of scientific research fund application and expenditure; performing semantic classification on scientific research funds according to the semantic feature data set, and generating semantic boundary data and compliance difference analysis data based on a compliance rule base; constructing an attention mechanism neural network, performing semantic differentiation on the semantic boundary data, and outputting a semantic weight feature matrix; based on the compliance difference analysis data, constructing a budget compliance inference network, and generating a compliance classification inference model of scientific research fund subjects; performing feature fusion on the semantic weight feature matrix and the output of the inference model to generate classification data with scientific research fund entries as units; and automatically generating a scientific research budget execution report according to the classification data of the scientific research fund items. And the generation efficiency and accuracy of the scientific research budget execution report are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method and system for automatically generating budget preparation and execution reports based on AI reasoning. Background Technology

[0002] The preparation and execution of research budgets typically require distinguishing the financial attributes and usage boundaries of different research funding items. However, due to the ambiguity of research funding application texts and the semantic overlap of expenditure items, existing AI-based budget preparation and execution report generation methods usually rely on historical samples and simple semantic classification models when identifying and classifying research funding items. This makes it difficult to accurately identify and distinguish budget items that are semantically similar but have significant differences in financial compliance requirements. As a result, deviations occur in the attribution of research funding items, budget occupancy calculations, and overspending risk assessments, making it difficult to achieve accurate, efficient, and compliant research budget execution management. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an AI-based reasoning-based method and system for automatically generating budget preparation and execution reports to address the problems mentioned in the background art.

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

[0005] The AI-based reasoning-based method for automatically generating budget preparation and execution reports includes the following steps:

[0006] S1: Obtain research funding application data and research funding expenditure data, and extract text semantic features and funding usage information respectively, and output the semantic feature dataset of research funding application and expenditure.

[0007] S2: Based on the semantic feature dataset, perform semantic classification of research funding subjects, and use the compliance rule base to analyze the compliance boundary differences between research funding subjects, generating semantic boundary data and compliance difference analysis data for research funding subjects;

[0008] S3: Based on semantic boundary data, construct an attention mechanism neural network for semantic differentiation of research funding subjects, and output a semantic weight feature matrix;

[0009] S4: Based on compliance discrepancy analysis data, a budget compliance reasoning network is constructed using knowledge graph reasoning methods, and a reasoning model for compliance classification of research funding subjects is output;

[0010] S5: The semantic weight feature matrix is ​​fused with the output of the inference model for the compliant classification of research funding items to generate classification data for research funding items.

[0011] S6: Automatically generate research budget execution reports based on the categorized data of research funding items.

[0012] In a preferred embodiment, S1 specifically refers to:

[0013] Obtain data on research grant applications and research grant expenditures;

[0014] Extracting textual semantic features corresponding to research funding application subjects from research funding application data;

[0015] Extract the purpose of funds for each research funding expenditure item from the research funding expenditure data;

[0016] The semantic features of the text and the information on the use of funds are cleaned and standardized to output a semantic feature dataset of research funding applications and expenditures.

[0017] In a preferred embodiment, S2 specifically refers to:

[0018] Word vector encoding is performed on the research funding categories in the semantic feature dataset to obtain the corresponding semantic vectors;

[0019] Construct a semantic similarity matrix for research funding subjects based on cosine similarity calculation between semantic vectors;

[0020] The compliance rule base is invoked to mark the research funding items in the semantic similarity matrix that have a similarity higher than a preset threshold and different financial ownership, thereby generating semantic boundary data for the research funding items.

[0021] For research funding categories, the compliance rule database is searched, and a comparative analysis of differences is conducted in conjunction with semantic boundary data to generate compliance difference analysis data for research funding categories.

[0022] In a preferred embodiment, S3 specifically refers to:

[0023] Based on the semantic boundary data of research funding subjects, an attention mechanism neural network is constructed for semantic differentiation of research funding subjects.

[0024] The input layer receives semantic boundary data of research funding categories and inputs the semantic vectors corresponding to the research funding categories into the attention computation layer.

[0025] The attention calculation layer calculates attention weights based on semantic vectors for similarity and conflict markers in semantic boundary data, and obtains the attention weight score corresponding to each research funding subject.

[0026] The output layer integrates the attention weight scores and outputs a semantic weight feature matrix.

[0027] In a preferred embodiment, S4 specifically refers to:

[0028] Based on compliance discrepancy analysis data of research funding categories, a budget compliance reasoning network is constructed using knowledge graph reasoning methods;

[0029] The entity node layer generates compliance entity nodes with research funding categories as nodes based on compliance difference analysis data;

[0030] The relation representation layer establishes compliance constraint relationships based on the prohibited cross-addition markers and the limit difference thresholds between compliance entity nodes, generating a relation graph structure with compliance difference relationships as edges.

[0031] The reasoning decision-making layer, based on a relational graph structure, uses compliance constraints to perform budget compliance reasoning and outputs a reasoning model for the compliance classification of research funding items.

[0032] In a preferred embodiment, S5 specifically refers to:

[0033] The subject number of the research funding subject in the semantic weight feature matrix is ​​matched with the corresponding attention weight score to form a semantic feature set of the research funding subject.

[0034] The compliance relationships, allowable expenditure ranges, and limit control conditions among research funding items in the reasoning model are mapped to the item numbers of research funding items in the semantic feature set, forming a compliance feature set for research funding items.

[0035] Data is mapped and concatenated based on the subject number between the semantic feature set and the compliance feature set of research funding subjects, and the output is classified data in units of research funding items.

[0036] In a preferred embodiment, S6 specifically refers to:

[0037] Based on the classification data of research funding items, each research funding item is categorized and marked;

[0038] The expenditure amount of the research funding expenditure item is compared with the budget amount of the research funding application item to determine the use of the budget amount and the execution of the budget expenditure for the research funding item.

[0039] The compliance classification results are judged, the difference between the expenditure amount of the research funding item and the budget application amount is calculated, and the overspending of the research funding item is determined.

[0040] Summarize the budget usage, expenditure execution, and overspending of research funding items, organize them according to the item number of the research funding category, and form a research budget execution report.

[0041] On the other hand, the present invention provides an AI-based inference-based automatic generation system for budget preparation and execution reports, comprising:

[0042] Data acquisition module: Acquire research funding application data and research funding expenditure data, extract text semantic features and funding usage information respectively, and output semantic feature datasets of research funding applications and expenditures;

[0043] Semantic classification module: Based on the semantic feature dataset, the module performs semantic classification of research funding subjects and uses the compliance rule base to analyze the compliance boundary differences between research funding subjects, generating semantic boundary data and compliance difference analysis data for research funding subjects;

[0044] Semantic differentiation module: Based on semantic boundary data, an attention mechanism neural network is constructed for semantic differentiation of research funding subjects, and the semantic weight feature matrix is ​​output;

[0045] Compliance Reasoning Module: Based on compliance discrepancy analysis data, a budget compliance reasoning network is constructed using knowledge graph reasoning methods, and a reasoning model for compliance classification of research funding subjects is output;

[0046] Feature fusion module: fuses the semantic weight feature matrix with the output of the inference model for the compliant classification of research funding items to generate classification data for research funding items;

[0047] Report generation module: Automatically generates research budget execution reports based on the categorized data of research funding items.

[0048] The technical effects and advantages of the AI-based reasoning-based automatic generation method and system for budget preparation and execution reports of this invention are as follows:

[0049] Semantic features and funding usage information are extracted from research funding application data and research funding expenditure data, respectively, and then uniformly cleaned and standardized to avoid recognition biases caused by text noise and non-standard expressions. By combining semantic classification with a compliance rule base, semantic boundary data and compliance difference analysis data for research funding subjects are generated, which can accurately characterize the boundaries of research funding subjects with similar semantics but different financial attributions, effectively avoiding classification ambiguities caused by ambiguous descriptions. The constructed attention mechanism neural network can automatically adjust the attention weights between research funding subjects based on semantic boundary data, strengthening the ability to distinguish conflicting research funding subjects. The output semantic weight feature matrix intuitively reflects the semantic difference strength between research funding subjects. A budget compliance reasoning network is constructed using knowledge graph reasoning methods, storing compliance constraints in a graph structure and automatically reasoning to generate a reasoning model for compliance classification of research funding subjects that conforms to financial management standards. Feature fusion of the semantic weight feature matrix and the output of the research funding subject compliance classification reasoning model provides a basis for decision-making. The automatically generated research budget execution report can comprehensively reflect the budget usage, execution progress, and overspending risk of each subject, reducing manual review costs and improving budget management efficiency and compliance level. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the AI-based reasoning-based automatic generation method for budget preparation and execution reports according to the present invention.

[0051] Figure 2 This is a schematic diagram of the structure of the AI-based reasoning-based automatic budget preparation and execution report generation system of the present invention. Detailed Implementation

[0052] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1

[0054] Figure 1 This invention presents a method for automatically generating budget preparation and execution reports based on AI reasoning, which includes the following steps:

[0055] S1: Obtain research funding application data and research funding expenditure data, and extract text semantic features and funding usage information respectively, and output the semantic feature dataset of research funding application and expenditure.

[0056] S2: Based on the semantic feature dataset, perform semantic classification of research funding subjects, and use the compliance rule base to analyze the compliance boundary differences between research funding subjects, generating semantic boundary data and compliance difference analysis data for research funding subjects;

[0057] S3: Based on semantic boundary data, construct an attention mechanism neural network for semantic differentiation of research funding subjects, and output a semantic weight feature matrix;

[0058] S4: Based on compliance discrepancy analysis data, a budget compliance reasoning network is constructed using knowledge graph reasoning methods, and a reasoning model for compliance classification of research funding subjects is output;

[0059] S5: The semantic weight feature matrix is ​​fused with the output of the inference model for the compliant classification of research funding items to generate classification data for research funding items.

[0060] S6: Automatically generate research budget execution reports based on the categorized data of research funding items.

[0061] S1: Obtain research funding application data and research funding expenditure data, and extract textual semantic features and funding usage information respectively, outputting a semantic feature dataset of research funding applications and expenditures, including:

[0062] Obtain data on research grant applications and research grant expenditures;

[0063] Research funding application data refers to the budget application documents submitted by researchers or research institutions when applying for research project funding from research management departments. Research funding expenditure data refers to the text records of actual expenditures recorded by financial management departments during the actual implementation phase of research projects. Both research funding application data and research funding expenditure data are in text format and are obtained from the actual research funding management system or financial management information system. For example, research funding application data can be obtained through the data interface of the research budget system, with a text format such as: "The institute submits a research project budget application for 2025, with a total budget of RMB 1.5 million, including RMB 1 million for equipment purchase (for one high-performance mass spectrometer); RMB 300,000 for travel expenses (for participation in domestic and international academic exchanges); and RMB 200,000 for labor costs (for paying the salaries of research support staff)." Research funding expenditure data is automatically exported through the financial information system interface, in text format, for example: "On April 10, 2025, RMB 980,000 was paid to a scientific instrument company for the purchase of one high-performance mass spectrometer; on April 1, 2025, RMB 180,000 was paid for travel expenses to attend an international biology conference; on May 5, 2025, RMB 170,000 was paid for the labor costs of research support staff."

[0064] Research funding categories represent macro-level financial budget classifications, while research funding items are sub-items describing the specific uses of research funding under each category. Research funding application categories are the major budget categories listed in the research funding application data, used by the finance department for macro-level classification during fund management. Examples include equipment purchase costs, travel expenses, and labor costs, all falling under research funding application categories and reflecting the scope of macro-level financial budget classifications. Each research funding application category contains multiple specific research funding items. Research funding items are defined as specific descriptions of the use of individual expenditures. For example, a specific research funding item under the equipment purchase cost category might be "purchasing one high-performance mass spectrometer for protein analysis," and a research funding item under the travel expenses category might be "transportation and accommodation expenses for attending an international biology conference in the United States." This illustrates the hierarchical difference between research funding categories and research funding items: research funding categories represent macro-level budget classifications in financial management, while research funding items represent the specific uses of funds under each category.

[0065] Extracting textual semantic features corresponding to research funding application subjects from research funding application data;

[0066] From the text content of research funding application data, macro-level research funding application categories are extracted. Based on the text content of the entries under each research funding application category, natural language processing methods are used to form textual semantic features of the research funding application categories. Taking actual research funding application data as an example, if the text corresponding to the equipment purchase expense category is "Requesting a budget of RMB 1 million for the purchase of one high-performance mass spectrometer for protein analysis research," firstly, the text is determined to belong to the equipment purchase expense category. Then, word embedding transformation is performed on the text information to obtain a fixed-dimensional semantic feature vector corresponding to the equipment purchase expense category, for example, the vector dimension is 300, and the value on each dimension reflects the weight value of the subject's semantic features. Similarly, the text for the travel expense category, "Requesting a budget of RMB 300,000 for travel expenses to participate in domestic and international academic exchanges," is also converted into a semantic feature vector corresponding to the travel expense category in a similar way. This reflects that the research funding application categories are budget classifications at the macro level, forming semantic features of research funding application categories suitable for AI analysis and processing.

[0067] Extract the purpose of funds for each research funding expenditure item from the research funding expenditure data;

[0068] The macro-level category of research funding expenditure is determined from the textual information of actual expenditures. Then, textual content reflecting the macro-level purpose classification is extracted from the corresponding textual records of research funding expenditure categories as the funding usage information. For example, in the research funding expenditure data, "Payed 980,000 yuan to supplier scientific instrument company on April 10, 2025, for the purchase of one high-performance mass spectrometer," the macro-level research funding expenditure category is determined to be equipment purchase cost. The extracted macro-level usage information is "equipment purchase," not the equipment model information. Similarly, in the textual data of "Pay 180,000 yuan for travel expenses to attend an international biology conference," the macro-level research funding expenditure category is travel expenses, and the extracted funding usage information is "travel expenses." This reflects the macro-level financial budget classification of research funding expenditure categories, highlighting the difference between the research funding expenditure categories and the usage descriptions of the research funding items.

[0069] The semantic features of the text and the information on the use of funds are cleaned and standardized to output a semantic feature dataset of research funding applications and expenditures.

[0070] The textual semantic features of research funding application categories and the funding purpose information of research funding expenditure categories are standardized and processed in a unified manner. This allows research funding application and expenditure data to be expressed under a unified macro-classification of research funding categories. For example, application data for "purchasing one high-performance mass spectrometer for protein analysis research" and expenditure data for "purchasing one high-performance mass spectrometer" are both standardized and processed into the macro-purpose category "equipment purchase". Similarly, application data for "travel expenses for attending an international biology conference" and expenditure data for "paying travel expenses" are both standardized and processed into the macro-purpose category "travel expenses". After standardization, the final output dataset is a structured semantic feature dataset, such as "equipment purchase fee: [application semantic feature vector, expenditure purpose text 'equipment purchase'], travel expenses: [application semantic feature vector, expenditure purpose text 'travel expenses']".

[0071] S2: Based on the semantic feature dataset, perform semantic classification of research funding categories, and use the compliance rule base to analyze the compliance boundary differences between research funding categories, generating semantic boundary data and compliance difference analysis data for research funding categories, including:

[0072] Word vector encoding is performed on the research funding categories in the semantic feature dataset to obtain the corresponding semantic vectors;

[0073] The semantic feature dataset of research funding applications and expenditures includes multiple macro-level research funding categories, such as equipment purchase costs, travel expenses, labor costs, material costs, and outsourcing costs. As a macro-level budget classification in financial management, each research funding category contains corresponding application and expenditure data. For example, the equipment purchase cost category includes a semantic feature vector of the application text and the expenditure purpose information text "equipment purchase"; the travel expense category includes a semantic feature vector of the application text and the expenditure purpose information text "travel expenses". By encoding the text information contained in the research funding categories using word vectors—that is, by using text embedding methods to convert the text information of the research funding categories into numerical vector representations suitable for AI computation—specifically, based on an existing corpus of research funding, the context window method is used to count the co-occurrence frequency of words, and the numerical vector corresponding to each word is calculated. Typically, the vector dimension is set to 300. For example, for the application and expenditure text information "application for budgeted equipment purchase expenses for the purchase of mass spectrometry instruments" and "equipment purchase" included in the equipment purchase expense item, the vector representations of words such as "equipment," "purchase," "mass spectrometry," and "instrument" in the text are calculated separately using word embedding methods. Then, the vectors of each word are averaged and combined to form vector data representing the overall semantic features of the equipment purchase expense item, i.e., the semantic vector of the equipment purchase expense item. Similarly, the semantic vector representations of the travel expense item and the labor service expense item are calculated in a similar way.

[0074] Construct a semantic similarity matrix for research funding subjects based on cosine similarity calculation between semantic vectors;

[0075] The semantic similarity matrix quantifies the semantic similarity between research funding items to identify potential budget classification conflicts or blurred boundaries. First, a method for calculating the similarity between semantic vectors of research funding items is determined. The cosine similarity method is used, treating the semantic vectors of two research funding items as vectors in a high-dimensional space and calculating the cosine of the angle between them as the numerical representation of similarity. The formula for calculating the cosine similarity between the semantic vectors of two research funding items is: divide the dot product of the two vectors by the product of their magnitudes. The output value ranges from 0 to 1. The closer the output value is to 1, the higher the semantic similarity between the two research funding items; the closer the output value is to 0, the lower the semantic similarity. For example, a calculated cosine similarity of 0.7 between equipment purchase costs and materials costs indicates a high semantic similarity between these two research funding items. However, a cosine similarity of 0.1 between equipment purchase costs and labor costs might indicate a significant semantic difference and clear classification boundaries between the two research funding items. The semantic similarity between all research funding items is calculated using the above method. Finally, a two-dimensional matrix of data is constructed with the research funding item numbers as rows and columns. This is the semantic similarity matrix of research funding items, where each cell in the matrix corresponds to the semantic similarity between two research funding items.

[0076] The compliance rule base is invoked to mark the research funding items in the semantic similarity matrix that have a similarity higher than a preset threshold and different financial ownership, thereby generating semantic boundary data for the research funding items.

[0077] The compliance rule base contains information such as the definition of research funding items, permitted expenditure ranges, prohibited cross-expenditure regulations, and funding ownership boundaries. First, a semantic similarity threshold of 0.7 is set. All research funding items with a similarity higher than 0.7 in the semantic similarity matrix are extracted and subjected to compliance rule retrieval. For example, the semantic similarity between the research funding items "Equipment Purchase Costs" and "Materials Costs" is 0.72, and they are defined as funding items with different financial ownership in the compliance rule base. In this case, conflict marking is performed, generating corresponding semantic boundary data records. These records include item number pairs (e.g., Equipment Purchase Costs item number SB001 and Materials Costs item number CL002), the calculated similarity (e.g., 0.72), and a conflict marker (e.g., "Semantic boundary conflict exists"). Following this method, research funding item pairs with similarity exceeding the semantic similarity threshold are analyzed and conflict marked, forming complete semantic boundary data for research funding items in a structured list format.

[0078] For research funding items, information on the scope, expenditure ratio, and annual limit in the compliance rules database is retrieved, and a comparative analysis of the differences is conducted in conjunction with semantic boundary data to form compliance difference analysis data for research funding items.

[0079] The system sequentially retrieves the budget allocation scope for each research funding item within the compliance rule base. For example, the allocation scope for equipment purchase expenses is defined as the purchase of scientific research instruments and equipment. Next, it retrieves the allowed expenditure ratio, such as allowing all (100%) of equipment purchase expenses to be used for purchasing instruments and equipment. Finally, it retrieves the annual budget limit for each research funding item, such as an annual limit of 1 million yuan for equipment purchase expenses. Simultaneously, it analyzes and compares research funding items with conflict markers in the semantic boundary data to identify differences in financial allocation between items. For example, if the semantic similarity between equipment purchase expenses (SB001) and material expenses (CL002) exceeds a threshold and a conflict marker exists, it needs to be recorded that equipment purchase expenses are prohibited from being used for material expenses, a cross-spending prohibition marker is required, and the threshold for the difference in limits between research funding items needs to be recorded. For example, if the limit for equipment purchase expenses is 1 million yuan and the limit for material expenses is 500,000 yuan, then the threshold for the difference in limits is 500,000 yuan. The above information is recorded in structured data format, for example: [Equipment purchase expense item number SB001, conflicting item number CL002, cross-spending prohibited is marked "Yes", and the limit difference threshold is 500,000 yuan]. Following the above method, all research funding items are analyzed and compared, ultimately forming a complete set of compliance difference analysis data for research funding items.

[0080] S3: Based on semantic boundary data, construct an attention mechanism neural network for semantic differentiation of research funding categories, outputting a semantic weight feature matrix, including:

[0081] Based on the semantic boundary data of research funding subjects, an attention mechanism neural network is constructed for semantic differentiation of research funding subjects.

[0082] Attention-based neural networks are deep learning models designed to process input vector data. Their goal is to improve the ability to identify semantic differences between research funding categories, such as those with ambiguous semantic boundaries or conflicting attributions. For instance, when research funding categories have high semantic similarity but belong to different financial accounts, an attention-based neural network can effectively identify and distinguish these semantic differences.

[0083] The input layer receives semantic boundary data of research funding categories and inputs the semantic vectors corresponding to the research funding categories into the attention computation layer.

[0084] The input layer receives semantic vectors from the semantic boundary data of research funding categories. These semantic vectors are digital representations formed by word vector encoding of the textual semantic features of the research funding category and information on the purpose of funding. Each research funding category corresponds to a unique semantic vector in the input layer. Each semantic vector has the same length, typically 300 dimensions, with the value in each dimension representing the magnitude and intensity of the research funding category's specific semantic features. The input layer reads and receives the semantic vector of each research funding category in the semantic boundary data one by one according to a predefined fixed format, and then passes all vectors to the attention computation layer in matrix form.

[0085] The attention calculation layer calculates attention weights based on semantic vectors for similarity and conflict markers in semantic boundary data, and obtains the attention weight score corresponding to each research funding subject.

[0086] The attention calculation layer employs a multi-head attention mechanism, mapping the semantic vector of each research funding item to a query vector, key vector, and value vector, respectively. Then, attention weight scores are calculated using dot product and the Softmax function. Taking the semantic vector of research funding item "Equipment Purchase Expenses" (SB001) as the query vector, the semantic vectors of other research funding items with semantic similarity and conflict markers (such as "Materials Expenses" (CL002)) are used as the key and value vectors. A preliminary attention score is obtained by performing a dot product operation between the query vector and the key vector. This preliminary attention score is then standardized by dividing by the square root of the vector dimension. Finally, the standardized attention weight score is obtained using the Softmax function, with values ​​ranging from 0 to 1. For example, if "Equipment Purchase Expenses" (SB001) and "Materials Expenses" (CL002) have high semantic similarity and conflict markers, the attention weight score calculated by the attention calculation layer using the above method is 0.92, indicating that this research funding item requires higher attention and differentiated treatment in budget management. Similarly, there are no obvious conflict markers between travel expenses and labor costs, indicating low similarity. The attention weight score might be 0.15, suggesting relatively low attention between these research funding items. Therefore, the attention weight score between each research funding item is output through the attention calculation layer, reflecting the degree of semantic boundary conflict between research funding items.

[0087] The output layer integrates the attention weight scores and outputs a semantic weight feature matrix that distinguishes semantic differences between research funding categories.

[0088] The semantic weight feature matrix includes the subject number of the research funding subject and the corresponding attention weight score.

[0089] Each research funding item number is integrated with its calculated attention weight score to form a semantic weight feature matrix for distinguishing semantic differences between research funding items. The semantic weight feature matrix is ​​represented in a two-dimensional structure. Each row corresponds to a research funding item number, and each column corresponds to other research funding item numbers with which it has a semantic boundary relationship. Each matrix cell records the calculated attention weight score between the corresponding research funding item pairs. For example, the attention weight score cell for equipment purchase cost (SB001) and material cost (CL002) in the matrix records a value of 0.92, indicating high attention; the attention weight score cell for travel expenses (CL003) and labor costs (LW004) records a value of 0.15, indicating low attention.

[0090] S4: Based on compliance discrepancy analysis data, a budget compliance reasoning network is constructed using knowledge graph reasoning methods, outputting a reasoning model for compliance classification of research funding items, including:

[0091] Based on compliance discrepancy analysis data of research funding categories, a budget compliance reasoning network is constructed using knowledge graph reasoning methods;

[0092] The budget compliance reasoning network is an automated logical reasoning tool implemented using knowledge graph reasoning methods. It is used to determine the constraints on budget compliance among research funding categories and automatically derive the compliance classification for each category. Compliance variance analysis data reveals the compliance management constraints between different research funding categories and serves as the basis for the budget compliance reasoning network's inferences.

[0093] The entity node layer generates compliant entity nodes with research funding subjects as nodes based on the subject number, conflicting subject number, prohibited cross-spending mark and limit difference threshold in the compliance difference analysis data of research funding subjects.

[0094] The entity node layer is used to generate nodes representing research funding items one by one based on compliance difference analysis data, i.e., compliance entity nodes. A compliance entity node includes: a research funding item number as the node's unique identifier; conflicting item numbers as the numbers of other research funding items that have compliance conflicts or related relationships with the node record; a prohibition on cross-spending flag to indicate whether a research funding item is subject to prohibition on cross-spending; and a limit difference threshold to record the budget limit difference between the research funding item and other research funding items. Taking actual research funding items as an example, the entity node layer creates an entity node for the equipment purchase expense item numbered SB001, recording the following data: node number SB001, research funding item number CL002 that has a conflicting relationship with it, prohibition on cross-spending flag is "yes," and limit difference threshold is 500,000 yuan; as another example, the entity node record for the travel expense item numbered CL003 has a compliance relationship with the labor cost item numbered LW004 but no prohibition on cross-spending flag, and a limit difference threshold of 200,000 yuan. The entity node layer processes all research funding items in this way, ultimately forming a node set composed of all research funding item nodes.

[0095] The relation representation layer establishes compliance constraint relationships between compliance entity nodes based on the prohibited cross-spending flags and the limit difference thresholds between compliance entity nodes, and generates a relation graph structure with the compliance difference relationships between pairs of compliance entity nodes as edges.

[0096] The relation representation layer establishes a network of relationships between research funding subject nodes based on the compliant entity node data created in the entity node layer. In the relation representation layer, every two research funding subject nodes with financial compliance discrepancies are interconnected by compliance constraints, forming a graph structure with the compliance discrepancy relationship between the node pairs as edges. The relation graph structure is represented in a node-edge format, with research funding subject nodes forming the nodes of the graph structure, and the compliance constraints between each node forming the edges. For example, if there is a prohibited cross-spending relationship between equipment purchase costs (entity node SB001) and material costs (entity node CL002) and a limit difference threshold of 500,000 yuan, the relation representation layer records in the graph structure that there is a prohibited cross-spending relationship between node SB001 and node CL002, and the relationship edge includes data showing a limit difference threshold of 500,000 yuan. Similarly, if there is no prohibited cross-spending marker between travel expenses (entity node CL003) and labor costs (entity node LW004), but a limit difference threshold relationship of 200,000 yuan exists, this is also recorded as a relationship edge between the two nodes in the graph structure. The relationship representation layer analyzes and establishes the relationships between the nodes one by one, ultimately forming a relationship graph structure that reflects the compliance relationships between all research funding items.

[0097] The reasoning and decision-making layer uses a relational graph structure and compliance constraints to perform budget compliance reasoning, and outputs a reasoning model for the compliance classification of research funding items.

[0098] The reasoning decision layer is the top layer of the budget compliance reasoning network. Its main function is to perform automated logical reasoning based on the relational graph structure output by the relational representation layer, using knowledge graph reasoning methods to determine the compliance classification of each research funding item. The reasoning process derives the compliance classification result by analyzing the compliance relationships of each node in the relational graph structure, the prohibition of cross-spending, and the difference between the budget limit and the prohibition of cross-spending. For example, through compliance reasoning on the equipment purchase expense item SB001, the reasoning decision layer analyzes the prohibition of cross-spending and the 500,000 yuan difference in the limit between it and the material expense item CL002. It determines that the permitted scope of equipment purchase expense is limited to the purchase of scientific research instruments and equipment, and the use of material expenses is prohibited. The budget limit control condition is limited to no more than 1 million yuan per year. Similarly, through reasoning, it is determined that the permitted scope of material expense item CL002 is the purchase of experimental materials, and the use of cross-spending for equipment purchase is prohibited. The budget limit control condition is limited to no more than 500,000 yuan per year. Following the same logic, through reasoning on travel expenses CL003 and labor costs LW004, it is concluded that cross-spending is allowed, but each must comply with its own budget limit requirements. The reasoning decision-making layer uses this method to reason at each node, ultimately forming a complete reasoning model that includes the definition of compliance relationships for each research funding item, the scope of allowed budget expenditures, the specific circumstances under which cross-spending is prohibited, and the conditions for budget limit control, thus providing an automated decision-making basis for research funding budget management.

[0099] S5: The semantic weight feature matrix is ​​fused with the output of the inference model for the compliant classification of research funding items to generate classification data for research funding items, including:

[0100] The subject number of the research funding subject in the semantic weight feature matrix is ​​matched with the corresponding attention weight score to form a semantic feature set of the research funding subject.

[0101] Based on the semantic weight feature matrix data, the subject number and corresponding attention weight score of each research funding item are extracted one by one. Then, a one-to-one mapping is used to match the subject number and attention weight score, forming a semantic feature set for each research funding item. This semantic feature set consists of the subject number and corresponding attention weight score of the research funding item, stored in a two-dimensional data table structure. The subject number is the primary key of the data table, and the attention weight score is the data item corresponding to the subject number. For example, the attention weight score for equipment purchase cost (subject number SB001) is 0.92, for material cost (subject number CL002) it is 0.85, for travel expenses (subject number CL003) it is 0.65, and for labor costs (subject number LW004) it is 0.30, and so on. The semantic feature set records the mapping relationship between each research funding item number and its attention weight score.

[0102] The compliance relationships, allowable expenditure ranges, and limit control conditions among research funding items in the reasoning model are mapped to the item numbers of research funding items in the semantic feature set, forming a compliance feature set for research funding items.

[0103] The compliance feature set of research funding items is generated based on the inference model in the budget compliance inference network. The inference model ultimately outputs the definition of each research funding item in budget compliance management, including whether there are prohibited cross-spending compliance relationships between research funding items, the permitted spending scope, and budget limit control conditions. For example, the inference model records that the equipment purchase cost of research funding item SB001 is prohibited from being cross-spended with the material cost of item CL002; the permitted spending scope for the equipment purchase cost of SB001 is limited to the purchase of scientific research instruments and equipment, specifically large-scale scientific research equipment and analytical instruments, with an annual budget limit control condition of 1 million yuan. Similarly, the material cost of research funding item CL002 is prohibited from being cross-spended with the equipment purchase cost, and the permitted spending scope is limited to the purchase of experimental materials, such as chemical reagents and biological reagents, with an annual budget limit control condition of 500,000 yuan. For example, travel expenses under research funding item number CL003 are permitted for attending domestic and international academic exchange conferences, with no obvious restrictions on cross-spending, and an annual limit of 200,000 yuan. Labor costs under research funding item number LW004 are permitted for the salaries of research support staff, with an annual budget limit of 300,000 yuan. The compliance relationships, permitted spending ranges, and budget limit control conditions of the above inference model are mapped and corresponded one by one through the semantic feature set of the item number and research funding item. The mapping method involves searching for identical research funding item numbers in the semantic feature set for each research funding item number in the inference model, and then concatenating the data to form a compliance feature set for research funding items. Each record in the compliance feature set of research funding items contains the research funding item number, the corresponding compliance relationship, the permitted spending range, and the limit control conditions. For example, taking equipment purchase expense number SB001 as an example, the compliance feature set records: [Research funding item number SB001, prohibited cross-spending research funding item CL002, permitted spending scope is research instrument and equipment purchase, annual budget limit of 1 million yuan].

[0104] Data is mapped and concatenated based on the subject number between the semantic feature set and the compliance feature set of research funding subjects, and the output is classified data in units of research funding items.

[0105] From the semantic feature dataset of research funding applications and expenditures, research funding items are read one by one. For example, "purchasing one high-performance mass spectrometer for protein analysis research" belongs to the equipment purchase expense category, and "purchasing experimental materials for protein analysis" belongs to the material expense category. Each research funding item is assigned a corresponding macro-level research funding category. Then, using the research funding category number corresponding to the research funding item as an index, the semantic feature set and compliance feature set of the research funding category are called and mapped, and data mapping and splicing operations are performed on a per-research funding item basis. The classification data of research funding items is a structured dataset with defined mapping relationships. Each research funding item includes the research funding category number corresponding to the research funding item, the attention weight score corresponding to the research funding category number, and compliance classification results such as compliance relationships, allowed expenditure ranges, and budget limit control conditions determined in the inference model. For example, the research funding item "Purchasing one high-performance mass spectrometer for protein analysis research" belongs to the equipment purchase expense category SB001. In the semantic feature set, SB001 has an attention weight score of 0.92. The compliance feature set records the permitted expenditure scope as equipment purchase and the prohibited cross-expenditure item CL002 (materials expenses). The annual budget limit is 1 million yuan. Therefore, the classification data record for this research funding item is: [Research funding item "Purchasing one high-performance mass spectrometer for protein analysis research", category number SB001, attention weight score 0.92, compliance relationship prohibited cross-expenditure CL002, permitted expenditure scope equipment purchase, annual limit 1 million yuan]. The same method is used to process all research funding items to obtain a classification data set of research funding items, reflecting the semantic features and compliance requirements of research funding items.

[0106] S6: Based on the categorized data of research funding items, automatically generate a research budget execution report, including:

[0107] Based on the classification data of research funding items, each research funding item is categorized and marked;

[0108] The classification and labeling of research funding entries is primarily based on the research funding category numbers in the research funding entry data records. All research funding entries under the same category are centrally processed and organized. The classification method involves reading the research funding category number corresponding to each research funding entry in the classification data record one by one, and then grouping the research funding entries according to the number, forming research funding entry groups categorized by research funding category number. Taking actual research funding entries as an example to illustrate the classification method, we first show the single-entry record format of research funding entries before classification. For example, the research funding entry record for "Purchasing one high-performance mass spectrometer for protein analysis research" is only individually labeled with its research funding category number SB001; "Purchasing a fluorescence microscope for cell research" is also only labeled with its corresponding research funding category number SB001; "Purchasing experimental materials for protein analysis" is only recorded with its research funding category number CL002; and "Attending an international biology conference" is recorded with its research funding category number CL003. At this point, the classification relationships between the various research funding items were not yet established. After adopting a classification method, the research funding subject numbers corresponding to the research funding items were uniformly grouped and managed. Items with the same research funding subject number were classified under the same research funding subject. For example, under research funding subject number SB001 (equipment purchase cost), there were research funding items "purchase of one high-performance mass spectrometer for protein analysis research" and "purchase of a fluorescence microscope for cell research"; under research funding subject number CL002 (material cost), there was a research funding item "purchase of experimental materials for protein analysis"; under research funding subject number CL003 (travel expenses), there was a research funding item "attendance at an international biology conference," and so on, completing the classification of all research funding items. Furthermore, each research funding item was also labeled. The attention weight score and compliance classification results in the research funding item data records were recorded using structured labeling, so that the research funding item classification results included corresponding semantic weights and compliance restrictions. For example, the research funding item "Purchasing one high-performance mass spectrometer for protein analysis research" is marked as: [Research funding item number SB001, attention weight score 0.92, permitted expenditure scope: purchase of scientific instruments and equipment, prohibited cross-expenditure CL002: material costs, budget limit 1 million yuan]; the research funding item "Purchasing experimental materials for protein analysis" is marked as: [Research funding item number CL002, attention weight score 0.85, permitted expenditure scope: purchase of experimental materials, prohibited cross-expenditure SB001: equipment purchase costs, budget limit 500,000 yuan].

[0109] The actual expenditure amount of the research funding expenditure item is compared with the budget amount of the research funding application item to determine the use of the budget amount and the execution of the budget expenditure for the research funding item.

[0110] Research funding expenditure items and research funding application items are information items contained in the research funding application data and research funding expenditure data. Research funding application items are descriptions of the budget purpose and budget amount submitted by researchers or research institutions when applying for research project funding, such as "Purchase one high-performance mass spectrometer, with a budget of 1 million yuan." Research funding expenditure items are records of expenditures and actual expenditure amounts incurred during the actual execution phase of the research project, such as "Pay 980,000 yuan to the supplier, a scientific instrument company, for the purchase of one high-performance mass spectrometer." The actual expenditure amount for each research funding expenditure item is compared with the planned budget amount for the corresponding research funding application item. Specifically, the actual expenditure amount for each research funding expenditure item is read, and then the corresponding budget amount is found in the corresponding research funding application item. The difference in amount is compared to determine the current budget allocation for each research funding item and the actual execution status of the research project. For example, the research funding application for "purchasing one high-performance mass spectrometer" had a budget of 1 million yuan, and the actual expenditure was 980,000 yuan. Therefore, the actual expenditure of 980,000 yuan represents 98% of the budget of 1 million yuan, and the budget expenditure execution status is the actual expenditure of 980,000 yuan. Similarly, the research funding application for "participating in an international biology conference" had a budget of 300,000 yuan, and the actual expenditure was 180,000 yuan. Therefore, the actual expenditure of 180,000 yuan represents 60% of the budget of 300,000 yuan, and the budget expenditure execution status is the actual expenditure of 180,000 yuan. By comparing each research funding item using these methods, a data set of budget utilization ratios and budget expenditure execution status is formed.

[0111] Determine the scope of allowable expenditures and the conditions for limit control, calculate the difference between the expenditure amount of the research funding item and the budget application amount, and determine the overspending situation of the research funding item.

[0112] Based on the compliance classification results of research funding items, budget compliance is analyzed and determined. First, the permitted expenditure range and budget limit control conditions for each research funding item are retrieved. Then, the difference between the actual expenditure amount and the corresponding budget application amount for each research funding item is calculated to determine the overspending situation for each item. Specifically, the difference between the actual expenditure amount and the budget application amount is calculated. If the difference is negative or zero, it indicates no overspending or that the budget amount has not been exceeded. If the difference is positive, it indicates that the actual expenditure amount has exceeded the budget application amount, and an overspending situation exists, which needs to be separately noted in the research budget execution report. For example, the actual expenditure of "purchasing one high-performance mass spectrometer" is 980,000 yuan, and the budget application amount is 1,000,000 yuan. The calculated difference is -20,000 yuan, indicating no overspending. The research funding item "purchasing experimental materials for protein analysis" has a budget amount of 500,000 yuan, and the actual expenditure amount reaches 550,000 yuan. The calculated difference is +50,000 yuan, indicating that the budget amount is exceeded by 50,000 yuan, and it is marked as an overspending. In addition, compliance determination needs to be made by combining the permitted expenditure scope and the budget control conditions. For example, if the research funding item "purchase of 1 high-performance mass spectrometer" is within the permitted expenditure scope, it will be determined to be compliant. However, if the research funding item exceeds the permitted expenditure scope or exceeds the budget limit control conditions, it will be marked as non-compliant expenditure.

[0113] Summarize the budget amount used, budget expenditure execution and overspending of research funding items, and organize them according to the item number of research funding items to form a research budget execution report that includes the amount used for research funding items, execution progress and overspending risk.

[0114] The report summarizes the budget usage, expenditure execution, and overspending: Using the research funding item number as an index, it summarizes the usage amount, actual expenditure progress, and overspending for all research funding items under each item number. The summary results include the total budget usage amount for each research funding item, the budget execution progress percentage for each item, and the overspending risk amount. For example, taking the equipment purchase expense item SB001 as an example, the budget usage for "purchasing one high-performance mass spectrometer" is 980,000 yuan, with a budget execution progress of 98% and no overspending. The summary report is: Equipment purchase expense 980,000 yuan, budget execution progress 98%, no overspending risk; Materials expense item CL002 has a budget usage of 550,000 yuan, a budget execution progress of 110%, and an overspending risk of 50,000 yuan. The report states that there is a 50,000 yuan overspending risk for materials expenses. After summarizing all research funding items, a complete research budget execution report is generated, reflecting the overall budget usage amount, execution progress, and overspending risk of the research project.

[0115] Example 2

[0116] The difference between Embodiment 2 and Embodiment 1 is that this embodiment introduces an AI-based reasoning-based system for automatically generating budget preparation and execution reports.

[0117] Figure 2 A schematic diagram of the AI-based inference-based automatic budget preparation and execution report generation system of the present invention is provided. The AI-based inference-based automatic budget preparation and execution report generation system includes:

[0118] Data acquisition module: Acquire research funding application data and research funding expenditure data, extract text semantic features and funding usage information respectively, and output semantic feature datasets of research funding applications and expenditures;

[0119] Semantic classification module: Based on the semantic feature dataset, the module performs semantic classification of research funding subjects and uses the compliance rule base to analyze the compliance boundary differences between research funding subjects, generating semantic boundary data and compliance difference analysis data for research funding subjects;

[0120] Semantic differentiation module: Based on semantic boundary data, an attention mechanism neural network is constructed for semantic differentiation of research funding subjects, and the semantic weight feature matrix is ​​output;

[0121] Compliance Reasoning Module: Based on compliance discrepancy analysis data, a budget compliance reasoning network is constructed using knowledge graph reasoning methods, and a reasoning model for compliance classification of research funding subjects is output;

[0122] Feature fusion module: fuses the semantic weight feature matrix with the output of the inference model for the compliant classification of research funding items to generate classification data for research funding items;

[0123] Report generation module: Automatically generates research budget execution reports based on the categorized data of research funding items.

[0124] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0125] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0126] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0127] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0128] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0129] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0130] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0131] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0133] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI inference-based budgeting and performance reporting auto-generation method, characterized in that, The method comprises the following steps: S1: obtaining scientific research fund application data and scientific research fund expenditure data, and extracting text semantic features and expenditure purpose information respectively, and outputting scientific research fund application and expenditure semantic feature data sets; S2: according to the semantic feature data set, the subject semantic classification of the scientific research fund is carried out, and the compliance boundary difference between the scientific research fund subjects is analyzed by using the compliance rule library, and the semantic boundary data and the compliance difference analysis data of the scientific research fund subjects are generated; S3: based on the semantic boundary data, an attention mechanism neural network for semantic distinction of scientific research fund subjects is constructed, and a semantic weight feature matrix is output; S4: based on the compliance difference analysis data, a knowledge graph reasoning method is used to construct a budget compliance reasoning network, and a reasoning model for compliance classification of scientific research fund subjects is output; S5: the semantic weight feature matrix and the output of the reasoning model for compliance classification of scientific research fund subjects are fused, and classification data of scientific research fund items are generated; S6: based on the classification data of the scientific research fund items, a scientific research budget execution report is automatically generated. 2.The AI-inference-based budgeting and performance reporting automatic generation method of claim 1, wherein, S1, specifically: obtain scientific research fund application data and scientific research fund expenditure data; extract the text semantic features corresponding to the scientific research fund application subjects from the scientific research fund application data; extract the expenditure purpose information corresponding to the scientific research fund expenditure subjects from the scientific research fund expenditure data; the text semantic features and the expenditure purpose information are data cleaned and standardized, and the semantic feature data sets of the scientific research fund application and expenditure are output. 3.The AI-inference-based budgeting and performance reporting automatic generation method of claim 2, wherein, S2, specifically: word vector coding is performed on the scientific research fund subjects in the semantic feature data set to obtain corresponding semantic vectors; a semantic similarity matrix of the scientific research fund subjects is constructed based on the cosine similarity between the semantic vectors; call the compliance rule library, mark the scientific research fund subject pairs with similarity higher than the preset threshold and different financial attributions for conflict, and generate semantic boundary data of the scientific research fund subjects; for the scientific research fund subjects, retrieve the compliance rule library, and combine the semantic boundary data to perform difference comparison and analysis, and form the compliance difference analysis data of the scientific research fund subjects. 4.The AI-inference-based budgeting and performance reporting automatic generation method of claim 3, wherein, S3, specifically: based on the semantic boundary data of the scientific research fund subjects, an attention mechanism neural network for semantic distinction of scientific research fund subjects is constructed; the input layer receives the semantic boundary data of the scientific research fund subjects, and inputs the semantic vectors corresponding to the scientific research fund subjects into the attention calculation layer; the attention calculation layer calculates the attention weight of the similarity and conflict mark in the semantic boundary data based on the semantic vectors, and obtains the attention weight score corresponding to each scientific research fund subject; the output layer integrates the attention weight score, and outputs the semantic weight feature matrix. 5.The AI-inference-based budgeting and performance reporting automatic generation method of claim 4, wherein, S4, specifically: based on the compliance difference analysis data of the scientific research fund subjects, a knowledge graph reasoning method is used to construct a budget compliance reasoning network; the entity node layer generates compliance entity nodes with scientific research fund subjects as nodes according to the compliance difference analysis data; the relationship representation layer establishes compliance constraint relationships according to the prohibited cross-spending mark and the limit difference threshold between the compliance entity nodes, and generates a relationship graph structure with compliance difference relationships as edges; The reasoning and decision-making layer performs budget compliance reasoning based on the relationship graph structure and the compliance constraint relationship, and outputs a reasoning model of the scientific research fund subject compliance classification. 6.The AI-inference-based budgeting and performance reporting automatic generation method of claim 5, wherein, S5, specifically: The subject number of the scientific research fund subject in the semantic weight feature matrix is matched with the corresponding attention weight score to form a semantic feature set of the scientific research fund subject; The compliance relationship between the scientific research fund subjects, the allowed range of expenditure, and the limit control conditions in the reasoning model are corresponded with the subject numbers of the scientific research fund subjects in the semantic feature set to form a compliance feature set of the scientific research fund subjects; Based on the data mapping and splicing between the subject numbers of the semantic feature set and the compliance feature set of the scientific research fund subjects, classification data of the scientific research fund items is output.

7. The AI-inference-based budgeting and performance reporting automatic generation method of claim 6, wherein, S6, specifically: According to the classification data of the scientific research fund items, each scientific research fund item is classified and labeled; The expenditure amount of the scientific research fund expenditure item is compared with the budget amount of the scientific research fund application item to determine the budget amount occupation and budget expenditure execution of the scientific research fund item; The compliance classification result is determined, the difference between the expenditure amount and the budget application amount of the scientific research fund item is calculated, and the overexpenditure of the scientific research fund item is determined; The budget amount occupation, budget expenditure execution, and overexpenditure of the scientific research fund item are summarized and sorted according to the subject number of the scientific research fund subject to form a scientific research budget execution report.

8. The AI inference based budgeting and performance report auto-generation system for implementing the AI inference based budgeting and performance report auto-generation method of any one of claims 1-7, characterized in that, It includes: Data acquisition module: acquire scientific research fund application data and scientific research fund expenditure data, and extract text semantic features and expenditure information respectively to output semantic feature data sets of scientific research fund application and expenditure; Semantic classification module: classify the scientific research fund subject based on the semantic feature data set, and analyze the compliance boundary difference between the scientific research fund subjects using the compliance rule library to generate semantic boundary data and compliance difference analysis data of the scientific research fund subject; Semantic differentiation module: based on the semantic boundary data, construct an attention mechanism neural network for semantic differentiation of the scientific research fund subject, and output a semantic weight feature matrix; Compliance reasoning module: based on the compliance difference analysis data, a budget compliance reasoning network is constructed using a knowledge graph reasoning method to output a reasoning model of the scientific research fund subject compliance classification; Feature fusion module: the semantic weight feature matrix and the output of the scientific research fund subject compliance classification reasoning model are fused to generate classification data of the scientific research fund items; Report generation module: based on the classification data of the scientific research fund items, a scientific research budget execution report is automatically generated.

Images