Financial data identification and accounting method based on artificial intelligence

Abstract

The invention relates to the technical field of financial accounting and artificial intelligence combination, in particular to an artificial intelligence-based financial data identification and accounting method, which comprises the following steps of: firstly, calculating a fractal dimension by using a multi-level grid splitting technology, and constructing a fractal financial semantic map covering transaction and exchange entities; and performing interference operation by using a quantum mixing aggregator, fusing fractal atlas attributes and subject candidate results to obtain interference scores, and performing symbol rule screening or manual auditing confirmation to generate calibrated subject information. Then, the subject information is used for automatically generating an accounting entry and performing account book archiving, and the whole decision process is recorded under the quantum hash technology; by continuously analyzing archived decision data, the system can adaptively adjust fractal dimension calculation and interference parameters, and precise recognition and efficient accounting of complex financial data are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of the combination of financial accounting and artificial intelligence, and particularly relates to a method for identifying and accounting financial data based on artificial intelligence. Background Art

[0002] In current financial management, artificial intelligence is widely used in links such as data collection, intelligent entry, and anomaly identification to reduce the burden of manual operations and improve the accuracy of accounts. In the prior art, multi-level digital employees are generally used in combination with a large number of rule scripts to handle diverse business scenarios (prior literature, publication number: CN118691418A, title: A Digital Employee Intelligent Financial Accounting Method and System). Although the present invention can achieve a certain degree of automation, it still faces bottlenecks in cross-system docking, complex rule maintenance, and handling of unexpected situations. Especially as the business volume and rules continue to expand, the traditional script-based accounting process is difficult to optimize dynamically and has deficiencies in compliance and timeliness. Summary of the Invention

[0003] In view of the above-mentioned many problems existing in the prior art, the present invention provides a method for identifying and accounting financial data based on artificial intelligence. First, the present invention constructs a financial semantic graph by using multi-level grids and fractal dimensions to present the multi-scale attributes of transaction nodes; then, through a quantum hybrid aggregator, an interference operation is performed on the fractal graph vector and the subject candidate vector to screen out credible subjects; finally, the calibrated subject information is output by combining symbolic rule verification and manual review, and quantum hashing technology is used to archive and trace decision records. The present invention takes into account the adaptability of artificial intelligence in complex financial scenarios and the robustness of traditional compliance accounting, and greatly improves automation and accuracy.

[0004] A method for identifying and accounting financial data based on artificial intelligence, comprising the following steps: Obtain financial source data and perform preliminary formatting and verification to obtain available financial data; Associate the nodes and entities in the available financial data to establish an initial semantic graph, and calculate the fractal dimension based on the node attribute vector to generate fractal similar edges or nested relationship edges, so that the transactions and entity nodes form a fractal financial semantic graph; Input the fractal financial semantic graph and the subject candidate results into a quantum hybrid aggregator, and use the quantum state superposition method to evaluate the constructive or destructive effects to determine the interference score. After performing symbolic rule verification, perform review and correction on the transaction nodes with conflicts or anomalies, output the updated subject information, and obtain the calibrated subject data; Generate accounting entries based on the calibrated subject data, perform ledger archiving, and use quantum hashing technology to record the decision-making information in the transaction processing and correction processes. After analyzing the decision-making information, adjust the fractal dimension calculation parameters and the weight factors of the quantum hybrid aggregator.

[0005] Preferably, in the process of obtaining financial source data and performing preliminary formatting and verification, it includes: Use a character detection algorithm to perform optical recognition on paper documents or scanned images to obtain text records; Align the fields of the recognized text records with the records in the structured business system, unify the currency units and set the date format, and screen out the records that do not meet the field integrity requirements; Mark the entries with numerical ranges that do not meet the preset thresholds as abnormal data to be reviewed, and output the available financial data that meets the field integrity.

[0006] Preferably, the steps of associating nodes and entities in the available financial data include: Use the transaction identifier or unique index to map each record to the corresponding entity number to generate the basic nodes of the initial semantic graph; When multiple transactions occur for the same entity number at different times, record the amount, subject, or account information in the node attributes in chronological order; Mark the records that cannot be matched to any entity number as isolated nodes.

[0007] Preferably, the steps of calculating the fractal dimension based on the node attribute vectors include: Use multi-level grid splitting to perform discretized statistics on the amount range, transaction frequency, and the number of associated entities; Calculate the fractal dimension of the nodes for the discretized data. When the difference between the fractal dimension of one node and that of another node is lower than the preset threshold, generate fractal similarity edges; When the node attributes are hierarchically recursively associated on the main feature components, establish nested relationship edges and update this association to the fractal financial semantic graph.

[0008] Preferably, after generating the fractal similarity edges or nested relationship edges, by comparing the transaction frequency and cumulative amount between nodes, the preliminary identification of abnormal transaction nodes is realized, and the preliminary identification results are saved in the fractal financial semantic graph.

[0009] Preferably, the steps of jointly inputting the fractal financial semantic graph and the subject candidate results into the quantum hybrid aggregator include: Construct quantum state sequences for the node attribute vectors of the fractal financial semantic graph and the subject candidate result vectors respectively; Based on the principle of quantum state superposition, perform constructive merging on similar vectors and perform cancellation processing on conflicting vectors to obtain interference scores; Use the candidate subjects with interference scores higher than the threshold as priority attribution suggestions to provide input data for subsequent symbol rule verification.

[0010] Preferably, when performing symbol rule verification, judge each candidate subject with a high interference score according to the transaction amount limit, compliance accounting subject regulations, and internal approval terms, and eliminate and file the candidate subjects that do not meet the rules.

[0011] Preferably, when auditing and correcting transaction nodes with conflicts or anomalies, record the transaction number, the subjects before and after correction, the operator, and the modification time, and associate this correction information with the interference score and write it into the calibrated subject data to form updated subject information.

[0012] Preferably, when generating accounting entries based on the calibrated subject data and performing ledger archiving, configure transaction identification tags, specific subject information, and associated counterparty information for each entry.

[0013] Preferably, use quantum hashing technology to write the decision-making information in the transaction processing process and the correction process, and after analyzing the decision-making information, adjust the grid splitting scale used in the fractal dimension calculation and the interference parameters of the quantum hybrid aggregator.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: The present invention realizes the adaptive aggregation of transaction nodes in multiple dimensions through the technical means of the fractal financial semantic map, effectively avoiding the rule explosion problem caused by the excessive dependence on complex scripts in the comparison scheme; The present invention realizes the deep integration of subject candidate information and fractal attribute vectors through the technical means of the interference operation of the quantum hybrid aggregator, solving the difficulties of poor efficiency and easy misjudgment in the multi-layer digital employee sorting in the comparison scheme; The present invention realizes the safe archiving and self-learning analysis of the decision-making process through the technical means of quantum hash immutability, further solving the defect of the lack of closed-loop optimization in the rule iteration and anomaly traceability in the comparison scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the quantum hybrid aggregator and the fractal financial semantic map in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, numerous specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, obviously, one or more embodiments can also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present disclosure.

[0017] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0019] As Figure 1 shown, an artificial intelligence-based financial data identification and accounting method includes the following steps: Obtain financial source data and perform preliminary formatting and verification to obtain available financial data; Preferably, in the process of obtaining financial source data and performing preliminary formatting and verification, it includes: Use a character detection algorithm to perform optical recognition on a paper document or a scanned image to obtain a text record; Align the fields of the recognized text record with the records of the structured business system, unify the currency units and set the date format, and screen out the records that do not meet the field integrity requirements; Mark the entries whose numerical ranges do not meet the preset thresholds as abnormal data to be rechecked, and output the available financial data that meets the field integrity.

[0020] The present invention combines the character detection algorithm and data normalization strategy in artificial intelligence to realize the preliminary integration and verification of multi-source financial data: In a paper voucher or a scanned image, text information is extracted through an optical recognition process and converted into a machine-readable structured record. The artificial intelligence model mainly uses the principles of character positioning and classification to generate a text mapping for each image block, reducing the errors caused by manual entry. In one embodiment, the character detection algorithm will perform block processing on the scanned image: first divide the image into rows and columns, and output the characters to be classified block by block; then bundle and store the characters with their confidence values to improve the traceability of the subsequent verification of financial data fields.

[0021] Since financial data may contain multiple currencies, for subsequent processing and accounting, it is necessary to convert all original amount fields into a unified currency unit. In one embodiment, if the US dollar is used as the unified unit, the formula is:

[0022] where represents the converted amount, represents the original amount, represents the exchange rate from the original currency to the unified currency.

[0023] The processing of date formats is similar. Different date formats (such as "YYYY / MM / DD" or "DD-MM-YYYY") are re-split, re-labeled, and then stored as the same type of date data structure.

[0024] By setting field integrity rules, it is detected whether each record meets the minimum required fields (for example, it must include the transaction date, amount, and counterparty), and records that do not meet the requirements are marked as "abnormal pending review"; in one embodiment, if the value in the amount field is less than zero or higher than a certain upper limit, this record is marked as an abnormal entry, providing an entry for subsequent manual verification or further screening by the algorithm.

[0025] Through the above process, the system can initially integrate financial data from multiple sources and in multiple formats into a standardized data structure, reducing type conflicts and data omissions that occur in subsequent analysis.

[0026] In practical applications, the specific process of text record extraction includes: Scanning image access: First, the input image is segmented by row, and the possible positions corresponding to each row of characters are identified. The character detection algorithm is used to detect characters row by row, and the coordinates of the detection box and the predicted characters are output; then the predicted characters are merged into a text record in order. If the character detection does not meet the confidence threshold, the label "detection failed" is attached when outputting the text record for subsequent manual proofreading. Through this process, the repetitive workload and error rate during manual entry are greatly reduced.

[0027] In practical applications, the specific process of the currency conversion link includes: The system loads the valid on the current day or period from the exchange rate interface, and converts all original amounts based on the same benchmark. If the currency code in the record cannot be matched (such as an unknown currency), the record enters "abnormal pending review". This method of unifying the currency is beneficial for comparing and correlating the amount attributes between different transaction nodes in the semantic graph later, without the need to process multiple currency types separately again.

[0028] Set the date storage specification, such as "YYYY-MM-DD", and split-combine dates from different sources: If the original date is in the form of "DD / MM / YYYY", slice and recombine the string positions; if non-numeric or out-of-range values are detected in the date field, mark it as an anomaly. This unified date storage method facilitates grasping the timing attributes of transaction nodes during subsequent fractal dimension calculations, such as calculating the number of days between two transactions.

[0029] In practical applications, anomaly item marking and output include: By presetting thresholds, items with amounts exceeding the normal range or lacking key information can be screened.

[0030] The marked items are temporarily removed from the available financial dataset or placed in the "dataset to be reviewed", and are manually confirmed or rejudged by invoking other rule algorithms. Finally, only the data set with complete fields and meeting the threshold requirements is retained to form the available financial data for subsequent use in "semantic graph construction and accounting processes".

[0031] Combining the character detection algorithm based on artificial intelligence with the unified rule verification strategy can automatically identify paper or unstructured data, remove anomaly items or mark defective fields, laying a unified data foundation for subsequent analysis of the fractal financial semantic graph and the quantum hybrid aggregator. By means such as exchange rate conversion and date format standardization, the chaos caused by "messy data source formats, missing fields, and inconsistent amount units" is reduced. The system as a whole reduces the workload of manually collecting, proofreading, and cleaning multi-channel financial data, and explicitly records the uncertain items in the anomaly marks. Under the process of the present invention, there is no need to repeatedly handle issues such as currency and dates when constructing the subsequent fractal semantic graph; the quantum hybrid aggregator can also directly read the standardized amount fields for further interference fraction calculations. In actual deployment scenarios, this step can make the data integration process more orderly and transparent. Once a problem occurs, it can be traced back to the specific scan record or the source of field conflicts, greatly improving the audit and error-checking efficiency.

[0032] Associate the nodes and entities in the available financial data, establish an initial semantic graph, and calculate the fractal dimension based on the node attribute vectors to generate fractal similarity edges or nested relationship edges, so that transactions and entity nodes form a fractal financial semantic graph; In the available financial data, each transaction can be regarded as a "transaction node", while the counterparty, project, or account information can be regarded as "entity nodes". The present invention forms an "initial semantic graph" by connecting transaction nodes and entity nodes to each other in a graph structure. On this basis, an attribute vector is defined for each node to describe the numerical characteristics of the node (such as amount, transaction frequency, time distribution).

[0033] After multiple nodes are mapped into the graph structure, in order to measure their distribution characteristics under multi-dimensional attributes, the present invention introduces the concept of "fractal dimension". Suppose a node attribute vector is as follows:

[0034] where represents the value of the node on the th attribute (such as transaction amount, transaction time span, number of historical transactions, etc.). The fractal dimension is used to evaluate the discrete coverage or distribution level of the node at multiple scales. For example, in a multi-level grid division scheme, if the attribute characteristics of the node show increasing coverage at each scale, it will exhibit a higher fractal dimension.

[0035] If the difference in fractal dimensions and the distance of the main attribute vectors of two nodes (denoted as and ) are both less than a threshold, a "fractal similarity edge" is added for them in the graph structure; this represents that they have approximate distribution characteristics at multiple scales. If a node has a hierarchical nested relationship with another node multiple times in terms of fractal dimension, a "nested relationship edge" is created, indicating a recursive attachment or hierarchical subordination between the nodes, which can be seen in multi-level contracting or cross-layer project financial scenarios.

[0036] In practical applications, each transaction record is marked as , each counterparty or project is marked as , and the corresponding nodes are inserted into the graph structure. If a counterparty identifier appears in a transaction record, connect and to form a preliminary "transaction-entity" mapping. In one embodiment, if a project has multiple associated transactions scattered in different months, they are merged into the project node to fully consider the transaction diversity of the project in subsequent fractal dimension evaluation.

[0037] For the transaction node , can be set, and the specific types of attributes depend on the financial scenario. For example, in procurement transactions, the attributes may include unit price, batch quantity, supplier rating, etc.; in scientific research project fund management, the attributes may be the proportion of matching funds, the batch of funds received, etc. Corresponding attributes can also be defined for the entity node , such as the total historical cooperation amount, the number of transactions involved, the level of associated projects, etc.

[0038] Under the conventional grid subdivision idea, first perform multi-scale division on the space where the attribute vector is located, and each scale is denoted as . At scale , count the number of grids If a box-counting fractal dimension estimation method is defined, the following core expression can be referred to:

[0039] where is the approximate fractal dimension of the node, is the number of grids in which the nodes are distributed at the corresponding scale, and are the grid scales of different levels. Such calculations only require necessary formulae and explanations of parameter meanings in the specification.

[0040] Let node have a fractal dimension , an attribute vector , and node have a fractal dimension , an attribute vector . If: and then in the present invention, a "fractal similarity edge" is added between node and . Where and respectively represent a preset dimension difference threshold and an attribute vector distance threshold. This operation can explicitly show that two nodes tend to be consistent in multi-scale and multi-dimensional features in the graph, which is very helpful for judging transaction attribution or checking similar records in the subsequent quantum hybrid analysis stage.

[0041] If node shows repeated nested coverage of node under multi-scale partitioning, that is, node belongs to the sub-distribution interval of at each level of scale, then it is determined that and have a fractal hierarchical dependence relationship. This situation is common in multi-level internal transfer or sub-fund appropriation scenarios in complex financial networks. By constructing nested relationship edges, it can help financial personnel identify cross-level and cross-project transaction chains.

[0042] Preferably, the step of associating nodes and entities in the available financial data includes: Mapping each record to the corresponding entity number using a transaction identifier or a unique index to generate the basic nodes of the initial semantic graph; When multiple transactions occur for the same entity number at different times, record the amount, subject, or current account information in the node attributes in chronological order; Mark records that fail to match any entity number as isolated nodes.

[0043] Each financial record usually contains a unique transaction identifier, denoted as and the corresponding entity numbers of the counterparty or project party . To represent the association between transactions and entities in the graph structure, it is necessary to map the transaction identifier to its matching entity number, thereby forming two types of nodes (transaction nodes and entity nodes) and their connecting edges within the initial semantic graph.

[0044] If the same entity number accumulates multiple transactions in different time periods, the transactions can be sorted according to the time sequence, and information such as amount, subject, or current account can be appended to the attribute record of the entity node for subsequent artificial intelligence algorithms to utilize the time dimension features for identification and accounting.

[0045] If the entity number in a certain transaction record cannot find the corresponding entity in the system, it is regarded as an "isolated node" and exists alone in the graph structure. This isolated node can prompt the system or financial personnel of possible anomalies such as new suppliers, unregistered projects, or missing records, and attention should be paid in subsequent processes (such as quantum state interference analysis or manual review).

[0046] In practical applications, the system first reads each record in the available financial dataset (such as "TX20230912-001") and (such as "SUP-A001" representing a certain supplier). In the initial graph, for each different Create a transaction node. If it has never appeared in the graph , then create a new entity node; if it has appeared before, use the existing entity node and add a transaction-entity connecting edge.

[0047] If a certain entity number has multiple consecutive transactions that are different, only differing in the date and specific amount and subject, then sort them from the earliest to the latest date:

[0048] In the attribute set of the entity node , record the amounts of these several transactions in chronological order , subject information , current account identifier to provide a serialized data source for subsequent fractal dimension calculation or anomaly detection.

[0049] When the entity number of the transaction record appears When the system fails to match successfully in the registered database or mapping rules, the system automatically marks this transaction node as an "isolated node". In the initial semantic graph, the isolated node has no entity connection edges, which helps to quickly identify scenarios of "unknown transactions" or "manual entry required" during subsequent quantum hybrid analysis or exception handling.

[0050] In one embodiment, an enterprise has three main suppliers, numbered respectively , and more than ten new transaction records are entered into the account on the same day.

[0051] The system reads the transaction identifier and supplier number one by one, directly establishes the connection edge between the transaction node and the entity node for the registered suppliers (such as " "), and appends the amount and subject within the chronological order to the attributes; for the " " that appears and cannot find the corresponding entity number in the system, mark this transaction node as an isolated node.

[0052] Finally, on the initial semantic graph, a part of the transaction nodes converge to the three main supplier nodes to form a star-shaped or chain-shaped structure, and the other part of the nodes (only one with no matching entity) are classified into the "isolated" node queue for subsequent manual verification.

[0053] Transaction centralized mapping can be achieved at the initial graph stage, enabling the management of the same entity's transactions in multiple periods through a single node view. When conducting fractal dimension analysis or quantum state interference later, the system can deeply aggregate the associated transactions, focus on investigating and supplementing the records of isolated nodes, and lay a visual and traceable foundation for the integrity of financial data.

[0054] Preferably, the steps for calculating the fractal dimension based on the node attribute vector include: Using multi-level grid splitting to discretely count the amount range, transaction frequency, and the number of associated entities; Calculating the fractal dimension of the node for the discretized data. When the difference between the fractal dimension of a node and that of another node is lower than the preset threshold, generate a fractal similarity edge; When the node attributes are hierarchically recursively associated on the main feature components, establish a nested relationship edge and update this association to the fractal financial semantic graph.

[0055] In the present invention, each transaction node or entity node has several attributes, such as amount, transaction frequency, and the number of associated entities, etc. These attributes can be integrated into a vector , where represents the a property value (such as the total transaction amount, the cumulative number within a specified period, the number of entities with which there is interaction, etc.).

[0056] Multi-level grid splitting means: setting several levels of scales for each attribute dimension, dividing the node attributes into corresponding intervals at each level, and discretely counting the number or frequency of grids into which the nodes fall. For example, taking the amount field as an example, the value space can be divided into several intervals, and the number of transactions within each interval is recorded; if the transaction frequency range is large, logarithmic scales or other methods can also be used to discretize the frequency values.

[0057] The fractal dimension is used in the present invention to quantify the complexity or self-similarity degree of node attributes at different scales. By statistically analyzing the node coverage on a multi-level grid, the fractal feature value of the node can be obtained. If the distribution of a node at different scales maintains a high degree of recursion or delicate structure, the fractal dimension tends to be higher; if the distribution is single or coarse-grained, the fractal dimension is lower. In the common "box counting" or "grid counting" methods in this field, if a certain node has a grid coverage number of at scale , then the fractal dimension can refer to a similar

[0058] where and are different grid scales. The specific formulas and parameters are listed in detail above and do not need to be written in detail here.

[0059] When the difference in the fractal dimension values between nodes A and B is small, and at the same time the distance of the main attribute vectors is also lower than a certain threshold, a fractal similarity edge is added to them in the fractal financial semantic graph, indicating that these two nodes are highly similar in multiple scales and dimensions.

[0060] If a certain node C shows a recursive inclusion or hierarchical nesting of another node A at multiple grid scales, a nested relationship edge is created so that the master-slave relationship or multi-level attachment relationship between them can be explicitly reflected in the graph.

[0061] In the financial scenario, the fractal similarity edge can help identify similar transaction groups (such as similar amount intervals and interaction frequencies), while the nested relationship can hint at the deep structure of multi-level transfers or cross-project fund management.

[0062] In practical applications, for the amount interval, dividing intervals can be set, such as , count the frequency or cumulative amount of transactions for each node in the interval. The processing of transaction frequency and number of associated entities is similar, and their value ranges can be segmented or split on a logarithmic scale. In one embodiment, if there is a node Amount attribute Larger, but more frequent If the amount is not high, the system will find the grid corresponding to "high amount and low number of times" under more scales. Shows special patterns in fractal dimension assessment.

[0063] After calculating the fractal dimension, the fractal dimension differences between nodes are compared. With Node The dimension difference satisfies:

[0064] in is the preset threshold, which can be considered and It has similarity under multi-level grid splitting. If its main attribute vector distance also meets another threshold , then the fractal similarity edge is established. The specific vector distance metric can be in the form of Euclidean distance or Manhattan distance.

[0065] In the specific implementation, the fractal dimension can be calculated for all nodes first. , and then perform pairwise matching when judging the distance, for those that satisfy:

[0066] and

[0067] Node pair Adding a "fractal similarity edge" indicates that the two are consistent or closely related in multiple dimensions, which is conducive to quickly locking in transactions of the same group during subsequent quantum interference or symbol verification.

[0068] If the node For Node In the grid division, inclusive coverage is shown, that is, at each level of scale Down, The grid and The grid contains recursive inclusion, which can be identified and There is a fractal level dependency.

[0069] So insert "nested relationship edges" into the graph structure ” and updates it to the fractal financial semantic graph to help identify potential hierarchical associations in cross-layer project finance or multi-stage funding chains.

[0070] In one embodiment, in a system, there are ≥10 nodes, where some nodes represent aggregated transaction clusters and some nodes represent single items, and the attributes include three dimensions: "amount range", "transaction frequency", and "number of counterpart entities".

[0071] First, through multi-level grid splitting, the interval coverage frequency of each dimension of each node at different scales is statistically counted, and then the fractal dimension is calculated. If node and node have very close fractal dimensions and similar attributes, then a fractal similarity edge is added ; if node includes node in a higher-level distribution at most scales, then a nested relationship edge is created for and .

[0072] Finally, a fractal financial semantic graph is generated, where some nodes form similar groups, and the other part of the nodes show a multi-level structure through nested relationships, so as to show more refined aggregation in subsequent quantum interference or automatic subject classification.

[0073] Through the fractal dimension calculation and multi-level grid splitting strategy, the system can distinguish the aggregation degree and hierarchical relationship of nodes under multi-dimensional attributes, which is more flexible than single-threshold filtering. The fractal similarity edge reveals strong connections for short-distance nodes, and the nested relationship edge represents hierarchical dependence, which is conducive to discovering potential connections of cross-department, multi-level projects, or recursive fund allocation in financial scenarios. In subsequent artificial intelligence accounting and anomaly detection, the graph structure of this fractal financial semantic graph provides a visual and traceable basic network for quantum state interference calculation or symbolic rule matching.

[0074] Preferably, after generating the fractal similarity edge or nested relationship edge, by comparing the transaction frequency and cumulative amount between nodes, the preliminary identification of abnormal transaction nodes is realized, and the preliminary identification results are saved in the fractal financial semantic graph.

[0075] In the fractal financial semantic graph, if two nodes are already associated by a fractal similarity edge or a nested relationship edge, it means that they show a high degree of approximation or hierarchical dependence in the multi-scale fractal dimension. The present invention further uses the transaction frequency and cumulative amount between nodes for comparison to identify possible abnormal transactions. Suppose node and node have established a fractal similarity edge, and the system can access their transaction times and cumulative transaction amount in the attribute vector. If these values conflict with the overall distribution or the threshold of the node itself, then it can be determined that there are abnormal signs.

[0076] When the system discovers that the frequency of transactions or the cumulative amount of a certain node is too high or too low compared to its associated nodes, it will mark the node with an "abnormal" or "to be reviewed" label in the graph. This preliminary identification is not the final conclusion, but it is sufficient to draw attention in subsequent quantum state interference or symbol rule verification processes, thereby reducing the missed inspection situation in the financial audit process.

[0077] To keep the subsequent data flow complete and traceable, the present invention associates the abnormal identification with the node attributes or edge attributes and stores them in the fractal financial semantic graph. In this way, subsequent analysis can quickly locate the suspected abnormal nodes by simply reading the graph data, further improving the accounting efficiency and accuracy.

[0078] In practical applications, summarize the transaction connection records between node and node : The frequency of transactions can be expressed as , where 1 is an indicator function that counts all the transaction entries belonging to to .

[0079] The cumulative amount can be expressed as , where represents the amount of the th transaction, and the sum of all transactions belonging to is included in this value.

[0080] If several nodes on the fractal similar edges of node show extreme values in terms of frequency or amount distribution, it can be determined that there is a possibility of abnormal transactions for itself or adjacent nodes.

[0081] The system can preset a set of thresholds and for each edge or each node respectively, which are used for the warning detection of the frequency of transactions and the cumulative amount. When or , the system marks the relevant nodes as "abnormal candidate nodes". In one embodiment, if node is connected to multiple similar nodes and the cumulative amount of each connected edge is higher than , then the abnormal level of rises with priority and a "high-risk" status is marked in the graph to guide subsequent algorithms or personnel for review.

[0082] If node forms a hierarchical relationship with node through a nested relationship edge, then compare and Frequency and cumulative amount differences in cross - level funds or transaction records. When the fund flow or the number of transactions under the nested relationship exceeds the expected range, or can be marked as abnormal, or it can be determined that there is an unexplainable large - amount transaction between the two.

[0083] To maintain the integrity of the process, in the fractal financial semantic graph of the present invention, the mark is written into the node or edge attributes, such as "abnormal_flag = true", and the threshold trigger information is recorded. This write - back operation ensures that in subsequent steps (quantum hybrid aggregator or symbolic rule verification), this information can be directly read from the graph without recalculating the transaction frequency or cumulative amount.

[0084] In one embodiment, within a group company, the node has ≥ 5 departments and each department has different fractal associations with supplier or project nodes. A certain department has similar edges with multiple supplier nodes in the fractal semantic graph.

[0085] The system counts the transaction frequency with relevant suppliers and the cumulative amount . If it is found that one of the edges is far higher than the general level, the system marks or as an abnormal candidate node. This preliminary identification record is in the fractal financial semantic graph, and during the subsequent review stage, this potential risk can be immediately noticed during human - machine collaboration or quantum interference evaluation.

[0086] By comparing the actual transaction frequency and amount values extended from the fractal similar edges and nested relationship edges, possible abnormal nodes can be automatically detected, thus focusing the analysis of a large amount of distributed financial data on a few high - risk areas. After visualizing these preliminary identifications in the graph, auditing and accounting personnel can quickly lock in suspicious entities or transaction links to achieve more efficient subsequent investigation.

[0087] As Figure 2 shown, input the fractal financial semantic graph and the subject candidate results into the quantum hybrid aggregator together, and use the quantum state superposition method to evaluate the constructive or destructive effects to determine the interference fraction. After performing symbolic rule verification, audit and correct the transaction nodes with conflicts or anomalies, output the updated subject information and obtain the calibrated subject data; The present invention first matches the "fractal financial semantic map" with the "subject candidate results" internally: in the fractal financial semantic map, each node already has fractal attributes and multi-dimensional transaction records. The subject candidate results often originate from previous artificial intelligence or rule inference links (such as a large language model classifying text descriptions), and a list of possible accounting subjects is given for each transaction node. These two types of information are input into the "quantum hybrid aggregator" together, enabling the system to evaluate the degree of consistency or conflict between them under the principle of quantum state superposition.

[0088] In conventional numerical fusion, most methods are only based on weighting or simple voting. The present invention introduces the idea of quantum state superposition: the node attribute vectors and subject candidate result vectors in the fractal map are regarded as two groups of "quantum states" respectively, and are superimposed in the same Hilbert space; when the two states are similar or in the same direction, a "constructive" effect may occur and the applicability of the subject is improved; if there are obvious contradictions, a "destructive" effect occurs and the priority of this subject is reduced; through this interference operation, the system calculates an interference score, and the higher the score, the stronger the consistency between the fractal map and the subject prediction.

[0089] The interference score is only a measure of the compatibility of quantum states, and it also needs to be re-verified in combination with the symbol rules (including accounting standards and subject usage specifications) built into the present invention. If a certain subject does not conform to the corresponding rules (such as the amount exceeding a certain limit or the subject type conflicting with the project type), it will be excluded from the candidate list. For transaction nodes that still have conflicts or anomalies after the symbol rule verification, the system will automatically prompt to enter the manual review and correction link. The manual reviewer can view the calculation process of the interference score and the rule conflict points and finally confirm the subject or make a revision operation.

[0090] All transaction nodes that pass the quantum interference and symbol rule screening and are manually reviewed are finally selected as the target subjects and updated to the dataset; after this step, the nodes officially have a determined subject information, which is called "calibrated subject information", and is returned to the fractal financial semantic map, and can be directly used in subsequent accounting entry and auditing links.

[0091] In practical applications, if a node is represented by a vector in the fractal map, and the candidate subject list is represented by a vector then the quantum hybrid aggregator will combine and map the two into a "quantum state" and evaluate its superposition effect.

[0092] When the "fractal state" and "subject state" of the node are the same, constructive interference occurs (the interference score increases), and vice versa, destructive interference occurs (the score decreases). The specific quantum mathematical details are generalizable knowledge in this field, and the present invention only needs to list the main logic of the interference calculation in the specification. The system will set an interference score threshold 。If a node has a candidate for a certain subject and the calculated interference score exceeds , it is determined to have a high confidence level; those below the threshold are downgraded. In one embodiment, if a transaction node performs a superposition operation with multiple candidate subjects and only one subject has an interference score significantly higher than , then this subject is intuitively marked as the preferred attribution.

[0093] Symbol rules usually include amounts, subject types, item regulations in the financial system, and may even include internal enterprise processes (such as travel expense limits, department subject availability, etc.). Rule matching is performed item by item on candidate subjects that meet the high interference score. If non-compliance (exceeding the limit, type conflict) is found, they are immediately excluded and the reasons for the conflict are recorded. If there is only one remaining subject that meets the conditions, it is directly determined; otherwise, if there are still multiple subjects passing or all are in conflict, the transaction node is classified into the "requiring manual review" queue.

[0094] In the manual interface, the system centrally displays the interference score, symbol rule matching results, and explanations of conflict points; if the reviewer finds that the subject is reasonable and feasible, it is finally approved; if it is found that the system identification is incorrect, it can be modified to other subjects and remarks are given. The review actions are written back to the fractal financial semantic graph, becoming an important basis for subsequent algorithm self-learning or system traceability.

[0095] After completing this process, each transaction node has a "calibrated subject information" field; nodes without a determined subject or in a dispute state remain in the pending state for further manual intervention. Finally, these calibrated nodes can be automatically referenced by subsequent accounting processing to generate vouchers or financial entries.

[0096] Preferably, the steps of inputting the fractal financial semantic graph and the subject candidate results into the quantum hybrid aggregator include: Constructing quantum state sequences for the node attribute vectors of the fractal financial semantic graph and the subject candidate result vectors respectively; Based on the principle of quantum state superposition, constructive merging is performed on similar vectors, and destructive processing is performed on conflicting vectors to obtain the interference score; The candidate subjects with an interference score higher than the threshold value are used as priority attribution suggestions, providing input data for subsequent symbol rule verification.

[0097] In the fractal financial semantic graph, each node has several numerical attributes, such as amount, fractal dimension, number of transactions, etc. The subject candidate result is the subject information that artificial intelligence or rules in the previous stage judge that a certain transaction node may belong to, and can be expressed in several vector forms. In the present invention, the fractal graph node vector is regarded as the "fractal state", and the subject candidate vector is regarded as the "subject state", and the two each form a column of quantum state sequences, and subsequent state superposition operations are performed in the quantum hybrid aggregator.

[0098] In conventional weighted or simple voting fusion, the system often only does linear addition or maximum value selection. The present invention introduces the idea of quantum superposition: when two vectors (fractal state and subject state) are similar in main features, the system interprets it as a constructive effect, that is, the score increases after interference; when the two conflict in main features, it is regarded as destructive, and the score decreases after interference; the obtained interference score reflects the matching degree between the fractal financial information and the subject candidate, and the larger the value, the more likely the candidate subject is correct.

[0099] If the interference score exceeds the preset threshold, the candidate subject is regarded as the preferred attribution object, providing input data for subsequent symbol rule verification. If the score is lower than the threshold, it means that the candidate subject does not match the fractal graph information and needs to be abandoned or carefully checked.

[0100] In practical applications, for the transaction node its fractal graph vector can contain several elements, such as where represents the fractal dimension, represents the frequency of transactions, represents the total amount or other key indicators. The subject candidate result vector can be denoted as where represents a certain candidate subject, which is the accounting subject information that the node may match, and the dimension can include the vectorized encoding of the subject category, historical attribution success rate, etc.

[0101] In the present invention, inside the quantum hybrid aggregator, and are respectively mapped to two quantum states and . When and are highly similar, they are constructive after superposition, and destructive when the similarity is low. Denote the result after interference as and the system calculates the interference score . The specific state superposition formula can refer to the quantum mechanics interference principle in theory, and it is not necessary to expand all here.

[0102] The interference score measures the node and the subject The degree of fit at the fractal and subject levels. If , where is the threshold value, it is determined as the preferred attribution subject; if it is lower than , then this subject is not adopted for the time being. In one embodiment, the interference score of the fractal form of node and the subject "office expenses" If it is higher than the threshold, the system will use "office expenses" as the most promising subject label for subsequent symbol rule verification.

[0103] The candidate subjects that reach the threshold are recorded in the "preferred attribution suggestion" list. If there is no conflict in the symbol rules subsequently, they can directly become the final selected subject information. If there is a conflict, it will be weighed during manual review or secondary processing.

[0104] In one embodiment, assume that node corresponds to a scientific research project expense. The system gives three candidate subjects, namely "material expenses", "consulting fees", and "equipment fees", through preliminary text analysis, and each is mapped to , , .

[0105] The quantum hybrid aggregator will fractal vectors and the above three groups of subject states are respectively subjected to interference operations to obtain interference scores . If is the highest and exceeds the threshold value, "material expenses" is listed as the preferred attribution suggestion.

[0106] Through quantum state superposition evaluation, rather than relying solely on simple scoring or voting, the system can comprehensively consider the degree of compatibility between fractal attributes (such as transaction frequency, amount range) and subject characteristics, reducing conflicts or omissions that may be caused by traditional linear fusion. The combination of fractal dimension and quantum interference significantly improves the flexibility of multi-dimensional financial information fusion, enabling more accurate identification of suitable subjects and performing fine screening for the next step of symbol rule verification.

[0107] Preferably, when performing symbol rule verification, according to transaction amount limits, compliance accounting subject regulations, and internal approval terms, each candidate subject with a high interference score is judged item by item, and the candidate subjects that do not meet the rules are excluded and archived.

[0108] In the present invention, the quantum hybrid aggregator has selected several candidate subjects with "high interference scores" for each transaction node, and these subjects have a strong theoretical matching degree. However, the financial system must be implemented under the constraints of real compliance and internal audit. This link ensures compliance through symbol rule verification and prevents incorrect subjects or illegal bookkeeping behaviors. The symbol rules may include: Transaction amount limits: upper limit of subject amount, lower limit of special subject spending, etc.; Legitimacy regulations for accounting subjects: For example, the fixed - asset subject is only applicable to transactions with a single - pen amount ≥ a certain amount; Internal approval terms: Such as requiring the signature of the project leader or additional approval for transactions exceeding a certain financial amount, etc.

[0109] When a subject candidate violates the above rules, that is, the transaction amount exceeds the applicable range of the subject, or the transaction attributes do not meet the internal approval requirements, then the subject is removed from the candidate list and archived. The archiving operation can record the "reason for removal" to retain evidence for subsequent review or secondary verification. At the same time, the remaining subjects that meet the rules are retained in the next step. If there is only one left, it can be directly recognized as the final subject information.

[0110] In practical applications, after the quantum hybrid aggregator generates candidate subjects, the system usually lists the subject list in descending order of interference scores:

[0111] Among them has the highest interference score, the lowest. The symbol rule engine judges each one by one, checking elements such as amount and approval terms. When there is a non - compliance, it is immediately removed, and the removal record is written into the log.

[0112] If the subject "fixed assets" has a minimum amount requirement , then the system determines that if the transaction amount , then this subject candidate immediately becomes invalid; if the subject "office expenses" has a maximum amount limit or is only applicable to specified departments, then transactions exceeding the limit or with non - compliant departments are also removed here; in one embodiment, if the node attribute vector shows that the amount is too high and the candidate subject is "office expenses", after the system retrieves the symbol rule table and finds that it exceeds the range, then the "office expenses" subject is removed from the candidate list and the "reason for excess" is recorded.

[0113] If the rule library stipulates that for items with a single - pen amount greater than a certain threshold to be recorded in "conference expenses", a signature from a leader is required, then the system checks whether the node attributes contain a leader - signature field or relevant approval fields; if the necessary approval mark is missing, the candidate subject is invalid, and the system indicates "approval conditions not met" during archiving for subsequent manual review or supplementary recording.

[0114] When the subject is removed, the system will remove the entry Write an audit log or database table for subsequent tracing of why the quantum interference results are high but still do not meet the actual compliance rules. When only 1 or 0 candidate subjects remain in the remaining candidates, the system can automatically perform subsequent actions: if only 1 remains, it is regarded as the final candidate; if 0, it means that the node cannot pass the rules and transfers to the "manual review" process.

[0115] In one embodiment, a certain transaction node The amount is 2,500 currency units. After quantum interference, three candidate subjects of "office expenses", "conference expenses", and "material expenses" are given, and the scores decrease in turn. Symbol rule verification: "Office expenses" has the highest score, but the enterprise stipulates that the upper limit of the amount of "office expenses" is 2,000; since 2,500 exceeds the limit, it is excluded and the reason of "exceeding the limit" is recorded; "conference expenses" requires the signature of the leader and supporting documents for the venue fee, and this node lacks a signature and does not meet the approval terms, so it is excluded again; although "material expenses" has the lowest score, it does not trigger any violation conditions and is finally retained.

[0116] The system retains one valid subject of "material expenses", completes automatic determination and outputs "calibrated subjects".

[0117] This link ensures the compliance and review depth of financial processing, builds a bridge between the AI automation integration score and traditional financial criteria; detailed records are left for the exclusion operation for convenient auditing or manual review; if the rules are upgraded, the previous exclusion records can also be traced back for backfilling; the credibility of the subject information output in the final link is greatly improved, which is convenient for downstream accounting and management decision-making.

[0118] Preferably, when auditing and correcting a transaction node with conflicts or anomalies, record the transaction number, the subjects before and after correction, the operator, and the modification time, and associate this correction information with the interference score and write it into the calibrated subject data to form updated subject information.

[0119] After passing through the quantum hybrid aggregator and symbol rule verification, if a certain transaction node still has problems such as interference score conflicts, multi-subject competition, or non-compliance with audit rules, it is classified as a "conflict or abnormal transaction node". At this stage, further auditing and correction are required by manual or assisted algorithms to ensure the correctness and credibility of the final subject information.

[0120] Once the financial staff or system administrator corrects the subject of the transaction node, it means that the previous suggestions given by artificial intelligence or preliminary rules are not completely suitable for the actual business. In the learning closed-loop of artificial intelligence, retaining elements such as the subject information before and after correction, the reason for correction, and the interference score will provide key information support for subsequent model optimization and audit tracking.

[0121] Write the finally confirmed or corrected subject information back to the "Calibrated Subject Data" set in the system to form a unique version that can be called for downstream bookkeeping, report statistics, or auditing. Artificial intelligence can also use these real correction instances to improve the accuracy of the next-round matching or interference operation, achieving adaptive evolution.

[0122] In practical applications, the interface presented by the system to the reviewers usually includes: Transaction number, such as "TX20230910-001"; interference score, symbol rule verification result; original candidate subjects (several possible options) and their reason explanations. The reviewer clicks or edits in the interface to specify the final subject or perform multi-item splitting, and records the operator identification (such as "UserID=FIN001") and the timestamp together.

[0123] If the subject is changed from "office expenses" to "materials expenses", then record: "Subject before correction = office expenses", "Subject after correction = materials expenses"; interference score and can be saved together. Subsequently, it can be evaluated why the subject with a high interference score was not finally adopted (possibly due to rule conflicts or lack of supporting invoices, etc.).

[0124] In the present invention, the interference score is a key indicator given by the superposition of quantum states. Associating it with the modification information allows the system to see the scenario of "even with a high score, it will be manually corrected" during the next self-learning, thereby optimizing the interference operation or weighting strategy. In actual operation, the system can record the data in a dedicated "correction log" table or write "final_subject=XX" in the fractal financial semantic graph node attributes and record "interference_score=YY", etc.

[0125] After the review operation is completed, the system updates the final subject field of this node in the "Calibrated Subject Data" database to the corrected result, and marks the previous subject as "invalid" or "overwritten". In one embodiment, if the review determines that the subject should be "conference expenses" instead of "travel expenses", then in the calibrated subject data, store "finalSubject=conference expenses" for the transaction node and delete or mark the previously stored "travel expenses" as deprecated.

[0126] In one embodiment, the node shows a relatively high interference score of "materials expenses" after quantum interference, but the symbol check finds that there are "conference" related fields in its transaction description. After manual review, it is determined that it should be classified into the "conference expenses" subject.

[0127] System Record: Transaction ID: "TXID=T_{7}"; Subject Before Correction: "Material Cost"; Subject After Correction: "Conference Expenses"; Interference Score: ; Operator: "FIN007", Modification Time: "2023-09-10T15:20"; When writing back the calibrated subject data, finally set T_{7} to "Conference Expenses" and retain the correction log for self-learning or audit viewing.

[0128] It is difficult for artificial intelligence or preliminary automatic solutions to cover all financial details, so manual intervention and record correction constitute a crucial safety valve. This correction process can be used as a high-value "adversarial example" in the next model training to help the system recognize situations where the interference score is high but still does not conform to business reality or symbol rules in certain scenarios, thus improving the next round of quantum interference or subject prediction. In the future, auditors can also view the log of transaction "T_{7}" to understand why the system initially considered "Material Cost" and why it was changed to "Conference Expenses" during the audit, enhancing the traceability and transparency of the entire financial process.

[0129] Generate accounting entries based on the calibrated subject data and perform ledger archiving, and use quantum hashing technology to record the decision-making information in the transaction processing and correction processes. After analyzing the decision-making information, adjust the fractal dimension calculation parameters and the weight factors of the quantum hybrid aggregator.

[0130] In the previous process, the final subject information has been determined for each transaction node (possibly through a quantum hybrid aggregator, symbol rule verification, and manual review), which is called "calibrated subject data". Based on this data, the system can automatically generate accounting entries for each node. For example, in the debit-credit accounting system, the node corresponds to a debit or credit subject, forming paired accounting entries to meet the requirements of the financial system.

[0131] The present invention introduces the concept of "quantum hashing" when performing ledger archiving: write the decision-making information in the transaction processing and correction processes into a block or a distributed ledger, and perform quantum-secure encryption on the hash value to ensure that it cannot be tampered with or forged subsequently. In this process, the "decision-making information" includes a series of operations from manual review to automatic interference evaluation, and the quantum hash signature can guarantee the integrity of the entire processing chain.

[0132] After the accounting is completed, the system will centrally analyze the decision-making information stored in the quantum hash ledger. By aggregating a large amount of "calibrated subject data" and process records, it is possible to detect the actual performance of the quantum hybrid aggregator in subject evaluation and the accuracy of the fractal dimension calculation in the spectral modeling. If it is found that the system error is relatively high in certain scenarios or node types, the model elements such as the fractal dimension calculation parameters and the quantum interference weight factors can be adjusted automatically or semi-automatically, so as to achieve adaptive correction and optimization of subsequent processing.

[0133] In practical applications, for each transaction node , its established account information (such as "office expenses", "material expenses", etc.) is read, and the corresponding accounting subject code is retrieved in the internal subject mapping table; according to the debit and credit accounting principle, keywords such as amount and transaction time are merged to form the following entry example: debit subject: subject code, amount; credit subject: subject code, amount. In one embodiment, if represents a purchase transaction and its final subject is "material expenses", the system generates a debit entry for material expenses and a credit entry for bank deposits or cash. This information is written into the enterprise financial system to complete the bookkeeping.

[0134] The system collects the decision-making information generated in each link of the "transaction processing process" and the "correction process" (including interference scores, reasons for sign rule conflicts, manual review and modification records, etc.), forms an event sequence and packs it. Under the quantum hashing technology, a hash value is generated for the packed information, and an anti-quantum algorithm (such as lattice-based or multivariate polynomial-based) is used for digital signature to make it have high-strength anti-tampering characteristics. This digital signature can be written into the blockchain or a local decentralized ledger, so as to ensure seamless traceability during subsequent audits or model improvements.

[0135] After the system reaches a certain scale periodically or in terms of the number of nodes, it will automatically read all the processing and correction records in the quantum hashing ledger, and evaluate whether the existing model or rule library needs to be updated by statistically analyzing indicators such as the final correction rate of each node, common conflict reasons, and interference score distribution. If it is found that the fractal dimension calculation shows deviations in high-amount transactions, the multi-level grid splitting strategy may be adjusted; if the quantum hybrid aggregator has a high conflict rate in certain subjects, the corresponding weight factor may be lowered or raised.

[0136] Taking the fractal dimension as an example, if the system statistically finds that the coverage quantity of large-amount transaction nodes is abnormally distributed in the high-scale interval, the threshold value can be fine-tuned or the multi-scale subdivision method can be adjusted; an example of adjusting the weight factor of the quantum hybrid aggregator: if a certain subject is often manually vetoed in the case of high interference scores, it means that the interference weight assigned to this subject is too high, and its superposition gain can be lowered; these parameter changes will take effect automatically during subsequent transaction processing, enabling the system to achieve adaptive correction and improvement and reducing the recurrence of similar errors.

[0137] Preferably, when generating accounting entries based on the calibrated subject data and performing ledger archiving, a transaction identification tag, specific subject information, and associated counterparty information are configured for each entry.

[0138] After processes such as quantum interference, symbol rule verification, and manual review, each transaction node has determined the final accounting subjects, namely, "verified subject data". In the traditional accounting system, to enable the financial system to record accounts, this final information needs to be converted into specific entries to ensure debit-credit symmetry and compliance with accounting principles.

[0139] When the system automatically generates accounting entries, the following key information is added to each entry to facilitate subsequent auditing, retrieval, and statistics: Transaction identification label: Indicates which specific transaction or transaction node the entry originates from (e.g., "TXID=TX20231015-001").

[0140] Subject information: The accounting subject code adopted for this entry, as well as the classification or sub-subject of this subject in the internal financial guidelines.

[0141] Counterparty information: Records the names and numbers of suppliers, customers, or departments associated with this transaction, facilitating subsequent statement analysis and fund flow tracking.

[0142] In the artificial intelligence financial recognition and accounting system, this archiving process is not just about storing records but providing a data source for subsequent algorithm auditing or self-learning: When auditing or reviewing the model performance later, the mapping relationship of this transaction among the graph, quantum interference, and final entry output can be quickly matched.

[0143] In practical applications, the transaction identification label can adopt the form of a unique number within the system (such as "TXID=TX20231015-001"), which is already bound to the transaction node in the previously verified subject data. When generating entries, this label is written into the entry field (a common practice is to add a "Transaction ID" field to the "Voucher Line Table" in the database), enabling a one-to-one or one-to-many traceable relationship between the entry and the graph node.

[0144] The verified subject data indicates the final subjects, such as "Office expenses - 6601" or "Material expenses - 5101". The system queries the complete code according to the internal "Subject Dictionary" and fills in the debit or credit position in the entry. In one embodiment, if a certain transaction node The final subject is "Material expenses", and the system automatically fills in the "Material expenses (subject code 5101), amount X, debit / credit flag D / C" etc. in the "Entry Line" of the accounting software.

[0145] The counterparty information can include supplier numbers, customer numbers, or internal department codes, such as "S001", "Dept02", etc. When generating entries, it is written into the auxiliary accounting items (such as in the debit-credit accounting system, the auxiliary accounting dimensions are often used for projects, suppliers, departments, etc.), facilitating the issuance of counterparty details, statistical statements, or anomaly analysis later.

[0146] When each entry line is complete, the system performs an "archiving" operation, which typically includes: writing the entry into the main ledger or the ERP financial database; recording the generation time of the entry, and if there is a blockchain or quantum hash scheme, attaching an immutable hash record for post - audit. In the scenario of the present invention, the identification and accounting results of artificial intelligence are officially implemented as traceable accounting vouchers here.

[0147] In one embodiment, the node is a transaction that has been calibrated as "conference expenses (account 6503)", with an amount of 1000 currency units, and the counterparty is "Dept05". When generating the entry, the system automatically generates a debit - and - credit line: "Debit: Conference expenses (6503), amount 1000; Credit: Bank deposits (1002), amount 1000"; and adds "TXID = T_{8}" and "Dept = Dept05" to the auxiliary fields. This entry line is successfully archived in the accounting software library, which is very clear for subsequent financial reports, account inquiries, or manual reviews.

[0148] Implementing the fractal financial semantic graph and the AI analysis results into the accounting entry forms a complete end - to - end process: from data cleaning → graph analysis → quantum interference → account verification → final entry, in one go. Since the entry contains transaction identification tags and associated counterparty information, the system can trace back and verify the cause and effect of any specific transaction node after the fact, greatly improving the audit efficiency and transparency. At the same time, it also provides accurate "ground truth" or labeled data for the subsequent iteration of the artificial intelligence model. When encountering transactions with similar attributes again, it can refer to the previous archived results for more accurate account judgment or risk assessment.

[0149] Preferably, quantum hash technology is used to write the decision - making information in the transaction processing process and the correction process, and after analyzing the decision - making information, the grid splitting scale used in the fractal dimension calculation and the interference parameters of the quantum hybrid aggregator are adjusted.

[0150] In the present invention, the decision - making information of each transaction processing and subsequent correction (such as manual review and correction of accounts, anomaly marking, etc.) is written into the ledger. When the ledger uses quantum hash technology, the system performs quantum - secure hashing and signing on this information to make it non - tamperable during transmission or storage, and can also ensure integrity and credibility during post - audit or review. The core of quantum hash technology is to use a quantum - resistant mathematical algorithm (such as lattice - based or multivariate polynomial - based), so that the hash value still has extremely high anti - cracking ability under future quantum computing capabilities.

[0151] The collected decision - making information mainly includes: Transaction process: node attributes, fractal calculation results, quantum interference fractions, symbol rule conflicts, etc.; Correction process: manually or automatically correct the subject information, timestamp, operator, reason for correction, etc. before and after.

[0152] When the system analyzes this information regularly or trigger - based, if it is found that the existing fractal dimension calculation or quantum hybrid aggregator cannot accurately reflect the business reality, then by adjusting the grid splitting scale used in the fractal dimension calculation and the interference parameters in the quantum hybrid aggregator, the next round of processing will be closer to the real scenario.

[0153] In practical applications, the fields written into the quantum hash ledger are: Transaction node identification: For example, "TXID = TX20231101 - 002" Quantum interference stage information: interference fraction, main vector comparison Decision - making process: corrected subject, reason for conflict, manual review remarks Timestamp: record the writing time Perform quantum hash hashing on the above - mentioned information to generate an immutable signature and write it into the distributed ledger for safekeeping.

[0154] Regularly summarize all recently processed transaction nodes, compare their initial fractal dimension evaluations, quantum interference fractions and final results to see which nodes have a higher correction rate or which subjects have frequent conflicts; If multiple nodes show significant inconsistencies in the same amount range or transaction frequency range, it is speculated that the grid division in the fractal dimension calculation performs poorly in processing this range, and it may be necessary to increase more dense splitting scales on the amount axis or time frequency axis; If the quantum interference fractions of multiple nodes are seemingly very high but are often vetoed by the symbol rules, it means that the interference parameter setting of the quantum hybrid aggregator may be too high, and it is necessary to lower the gain or refine the interference determination threshold, etc.

[0155] In the fractal dimension calculation stage, usually perform multi - level grid splitting on attribute dimensions such as amount, transaction frequency, number of counterpart entities, etc.:

[0156] If judged from the decision - making information, To There are often a large number of misjudged transactions in between, then the grid range can be increased, decreased or redistributed so that the fractal dimension can more finely reflect the node attribute distribution in this interval. For example, if the highest frequency interval is concentrated in , the system may locally encrypt and subdivide the original uniform grid, so that the fractal statistics can more accurately capture the node details.

[0157] The quantum hybrid aggregator for the fractal vector And the subject candidate vector When interference occurs, there are often gain coefficients or interference thresholds used to determine constructive or destructive fractions. When the analysis shows that the interference fractions of a certain type of transaction node are generally high but are massively corrected, it indicates that the gain setting is excessive; or if a large number of nodes cannot obtain high enough interference fractions resulting in candidate failures, it indicates that the interference parameters can be appropriately relaxed. By automatically comparing the correction rate and the conflict rate, the system can semi-automatically generate an "interference gain correction" scheme and apply it in the next processing.

[0158] In one embodiment, within one month, there are ≥20 transaction nodes with an amount range of 3000 - 5000. The quantum interference fractions are generally high, but the actual manual review correction rate reaches 50%. Analyzing the correction records in the quantum hash ledger, it is found that these transactions have subtle over-coarseness in the fractal dimension statistics.

[0159] The amount range of 3000 - 5000 is split into finer grid scales for recalculating the fractal dimension; the quantum interference gain parameters are finely tuned to reduce the over-preference for the medium amount range; After the next batch of similar transactions is processed, the manual correction rate drops significantly, indicating an improvement in the system's adaptability.

[0160] Through the above scheme, the present invention provides secure audit traceability for the quantum hash ledger; after reading these "real decisions", the fractal dimension and quantum interference make adaptive fine-tuning to reduce misjudgments; overall, it ensures that the financial automated processing has the ability to continuously evolve in the middle and later stages, continuously conform to the real business scenarios and reduce manual rework.

[0161] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0162] The above are only the embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A financial data identification and accounting method based on artificial intelligence, characterized in that: The following steps are involved: Obtain financial source data and perform preliminary formatting and verification to obtain usable financial data; Associating nodes and entities in the available financial data to establish an initial semantic graph, and calculating fractal dimensions based on node attribute vectors to generate fractal similarity edges or nested relationship edges, so that transactions and entity nodes form a fractal financial semantic graph; The fractal financial semantic graph and the subject candidate results are input into the quantum hybrid aggregator, and the constructive or destructive effect is evaluated by quantum state superposition to determine the interference score. After the symbol rule verification, the conflicting or abnormal transaction nodes are reviewed and corrected, and the updated subject information is output to obtain the calibrated subject data. Accounting entries are generated based on the calibrated account data and ledger archiving is performed. Quantum hashing technology is used to record the decision information of the transaction processing process and the correction process. After analyzing the decision information, the fractal dimension calculation parameters and the weight factors of the quantum hybrid aggregator are adjusted.

2. The method according to claim 1, characterized in that The process of obtaining financial source data and performing preliminary formatting and verification includes: Use character detection algorithms to perform optical recognition on paper documents or scanned images to obtain text records; Align the fields of the recognized text records with those of the structured business system, unify the currency unit and set the date format, and filter out records that do not meet the field integrity requirements; Entries whose value ranges do not meet the preset thresholds are marked as abnormal data to be reviewed, and available financial data that meets the field completeness is output.

3. The method according to claim 1, characterized in that: The step of associating nodes and entities in the available financial data includes: Use transaction identifiers or unique indexes to map each record to the corresponding entity number and generate the basic nodes of the initial semantic graph; When multiple transactions occur under the same entity number at different times, the amount, subject or current account information is recorded in the node attributes in chronological order; Records that fail to match any entity ID are marked as isolated nodes.

4. The method according to claim 1, characterized in that: The steps of calculating fractal dimension based on node attribute vector include: Multi-level grid splitting is used to discretize the amount range, transaction frequency and number of related entities; The fractal dimension of the node is calculated for the discretized data, and when the difference between the fractal dimension and the fractal dimension of another node is lower than a preset threshold, a fractal similarity edge is generated; When node attributes are hierarchically recursively associated on the main feature components, nested relationship edges are established and this association is updated to the fractal financial semantic graph.

5. The method according to claim 1, characterized in that After generating fractal similarity edges or nested relationship edges, the abnormal transaction nodes are preliminarily identified by comparing the transaction frequency and cumulative amount between nodes, and the preliminary identification results are saved in the fractal financial semantic graph.

6. The method according to claim 1, characterized in that The steps of inputting the fractal financial semantic graph and the subject candidate results into the quantum hybrid aggregator include: The node attribute vectors of the fractal financial semantic graph and the subject candidate result vectors are used to construct quantum state sequences respectively; Based on the principle of quantum state superposition, constructive merging of similar vectors and destructive processing of conflicting vectors are performed to obtain interference scores. Candidate subjects with interference scores higher than the threshold are taken as priority attribution suggestions to provide input data for subsequent symbol rule verification.

7. The method according to claim 6, characterized in that When using symbol rule verification, candidate accounts with high interference scores are judged one by one according to transaction amount restrictions, compliance accounting account regulations and internal approval terms, and candidate accounts that do not meet the rules are eliminated and archived.

8. The method according to claim 1, characterized in that: When reviewing and correcting transaction nodes with conflicts or anomalies, record the transaction number, accounts before and after the correction, operator and modification time, associate this correction information with the interference score and write it into the calibrated account data to form updated account information.

9. The method according to claim 1, characterized in that: When generating accounting entries based on the calibrated account data and performing ledger archiving, configure transaction identification tags, specific account information, and related party information for each entry.

10. The method according to claim 9, characterized in that Quantum hashing technology is used to write the decision information of the transaction processing process and the correction process, and after analyzing the decision information, the grid splitting scale used in the fractal dimension calculation and the interference parameters of the quantum hybrid aggregator are adjusted.

Images