Financial information quick consulting method

Abstract

The invention discloses a financial information quick consulting method, which comprises the following steps: step 1, data preprocessing: collecting voucher data, account book data and report data of an accounting system, step 2, user permission and demand configuration, step 3, multi-mode retrieval and data positioning, step 4, associated data display and traceability, and step 5, data query. 5, data updating and retrieval optimization, the system is successfully connected with various core financial systems of an enterprise through an API interface technology, various financial data such as vouchers, invoices and contracts are connected in series through a constructed financial knowledge graph after a data preprocessing process to form a complete association network, a solid data basis is provided for rapid and accurate retrieval, and meanwhile, the data are updated, retrieved and optimized. Through professional data cleaning and standardization processing, repeated data is removed, abnormal data is corrected, and data format standards are unified, so that consulting errors caused by disordered data formats and inconsistent information are effectively avoided, and the overall quality of financial data is improved.

Description

Technical Field

[0001] This invention belongs to the field of information retrieval technology, specifically relating to a method for quickly retrieving financial information. Background Technology

[0002] In the process of enterprise operation, management and decision-making, financial information, as a core data asset, runs through the entire process of business operations, capital flow, tax declaration, and strategic planning. Specifically, it covers a variety of data such as accounting vouchers, detailed ledgers, financial statements, business contracts, tax records, invoice stubs, and payment receipts. With the deepening of enterprise digital transformation and the expansion of business scale, financial data shows the characteristics of exponential growth, multi-source distribution and strong correlation. According to industry survey data, the annual increase in financial data for medium-sized enterprises can reach hundreds of thousands of records, while that for large group enterprises can exceed one million records. Moreover, the data is scattered and stored in multiple independent modules such as ERP accounting systems, contract management platforms, Golden Tax Phase III / IV systems, and invoice management software, forming "data silos".

[0003] Currently, enterprises and related users (such as finance personnel, auditors, and management) have increasingly frequent needs for accessing financial information. This requires not only rapid location of target data but also the ability to retrieve related data and trace the entire process. However, existing methods largely rely on traditional database searches or file directory browsing. Traditional search technologies are mostly based on "precise keyword matching" or "fuzzy field matching," which have significant limitations. On the one hand, search flexibility is extremely poor. If a user only vaguely remembers data characteristics (such as "travel expense vouchers from the sales department in the third quarter of last year" but forgets the specific voucher number, or confuses the difference between "transportation fees" and "logistics fees"), they need to repeatedly try different keyword combinations and adjust search conditions multiple times. On the other hand, the ability to perform cross-type data retrieval is lacking. For example, if a user needs to view "all invoices, vouchers, and payment records corresponding to a certain sales contract," they must first search for the contract number in the contract management system, then manually enter the number in the invoice system and accounting system for a second search. The entire process requires switching between 3-4 systems, making the operation cumbersome and prone to failure due to incorrect number input, further reducing search accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a method for quickly accessing financial information, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for quickly accessing financial information, comprising the following steps: Step 1: Data Preprocessing. Collect multi-source financial data including voucher data, ledger data, and report data from the accounting system; contract data from the contract management system; tax return data from the tax system; and invoice data from the invoice management system. Clean the collected financial data to remove duplicates and correct anomalies, and standardize the data format. Add multi-dimensional association tags, including basic tags and related identifiers, to the standardized financial data. Construct a financial data knowledge graph using core financial data as nodes and association tags as edges to establish relationships between different types of data. Step 2: User Permissions and Requirements Configuration: Establish a three-tiered permission control model of "role-permission-data scope," pre-define the roles of financial operator, financial supervisor, department head, and management, and configure corresponding basic access permissions. Collect users' historical access data, analyze the access habits of frequently accessed data types, access time periods, and related data combinations, and generate personalized access preference profiles. Configure a personalized access interface based on user roles and preference profiles. The interface includes a core data display area, commonly used search entries, and a quick access area for related data. Step 3: Multi-modal Retrieval and Data Location: Receive multi-modal retrieval requests from keyword retrieval, conditional combination retrieval, and natural language retrieval. Perform semantic parsing on the retrieval requests to extract core retrieval elements such as time range, data type, department, and business keywords. Generate retrieval instructions by expanding retrieval elements based on a pre-set thesaurus. Based on the retrieval instructions, query the financial data knowledge graph. First, use associated tags to initially screen and locate candidate data sets. Then, use a semantic similarity algorithm to sort the candidate data based on their matching degree with the retrieval request to obtain the target data. If the retrieval results are empty or the matching degree is below a pre-set threshold, return similar retrieval suggestions and provide an entry point for manual assistance. Step 4: Display and Traceability of Related Data: The target data is displayed in the core data area of ​​the personalized interface according to the display priority in the user's preference profile. Related data is automatically identified based on a financial data knowledge graph. A list of related data is displayed in the quick access area, allowing users to jump to specific points. The generation chain of the target data is visualized in flowchart form. This generation chain includes core nodes such as "contract signing → invoice issuance → voucher entry → ledger registration" and supports node navigation. Step 5, Data Update and Retrieval Optimization: An incremental update mechanism is adopted to monitor data changes from multiple financial systems in real time. Only newly added or modified financial data is cleaned, tags are updated, and knowledge graph associations are adjusted. Retrieval logs are analyzed regularly. Preferably, in step one, the abnormal data includes data with numerical format errors and date logic contradictions. The data after standardization is uniformly formatted with dates in "YYYY-MM-DD" format and amounts in "yuan" unit. The core financial data includes vouchers and contract data. The basic tags include data type, generation date, department, and business type. The associated identification tags include associated contract number, invoice number, and voucher number.

[0006] Preferably, in step two, the basic access permissions of the financial operator are to view the voucher data of their own department, the basic access permissions of the management are to view the report data of the entire department, the analysis cycle of the personalized access preference profile is 3 to 4 months, the core area of ​​the personalized interface of the financial accountant displays voucher details by default and the related area displays the corresponding invoices by default, the core area of ​​the personalized interface of the sales department head displays sales reports by default and the related area displays the corresponding contracts by default.

[0007] Preferably, in step three, the thesaurus contains the correspondences between "invoice" and "receipt," "expense" and "spending," and "payment receipt" and "receipt." The semantic similarity algorithm uses the cosine similarity algorithm, with a preset threshold of 60%–70%. An example of a conditional combination search request is "sales department expense vouchers for Q1-Q3 2024," and an example of a natural language search request is "query all invoices and corresponding vouchers for last year's R&D projects." Preferably, in step four, the display priority is determined based on the user's job position. Financial accountants are given priority in displaying voucher details, while management is given priority in displaying summary data. The associated data includes invoices, contracts, and ledger records associated with the vouchers.

[0008] Preferably, in step five, the search log includes search keywords, search result click-through rate, frequency of related data retrieval, optimized thesaurus, search element weights, and data display priority.

[0009] A financial information quick access system includes a data preprocessing module, a user configuration module, a retrieval and positioning module, a display and traceability module, an update and optimization module, and a data storage module. The data preprocessing module is used to collect multi-source financial data, clean and standardize it, add related tags, and construct a financial data knowledge graph. The user configuration module is used to build a three-level permission control model of "role-permission-data scope", generate personalized user viewing preference profiles and configure personalized viewing interfaces; The retrieval and positioning module is used to parse multi-mode retrieval requests and query the financial data knowledge graph to obtain target data; The display and traceability module is used to display target data and related data according to user preferences, and to visualize the data generation chain. The update and optimization module is used to maintain financial data and knowledge graphs using an incremental update mechanism, and to analyze retrieval logs to optimize retrieval strategies. The data storage module is used to store preprocessed financial data, financial data knowledge graphs, user permission information, and preference profiles.

[0010] Preferably, the data preprocessing module includes a data acquisition unit, a data cleaning unit, and a label and graph construction unit; the data acquisition unit acquires multi-source financial data through API interfaces and direct database connections.

[0011] Preferably, the user configuration module includes a permission management unit, a preference analysis unit, and an interface configuration unit. The permission management unit constructs a three-level model of "role-permission-data range" and dynamically adjusts user permissions. The retrieval and positioning module includes a request parsing unit, an intelligent retrieval unit, and a result optimization unit. The request parsing unit parses multi-mode retrieval requests, expands core retrieval elements based on a preset thesaurus, and generates retrieval instructions.

[0012] Preferably, the display and traceability module includes a personalized display unit, an associated retrieval unit, and a traceability visualization unit; the personalized display unit displays target data according to the user's job priority; the update and optimization module includes an incremental update unit and a retrieval optimization unit; the incremental update unit monitors data changes in real time and performs incremental updates; and the data storage module adopts a hybrid distributed storage architecture of 1TB SSD + 10TB HDD.

[0013] Compared with the prior art, the beneficial effects of the present invention are: The system successfully connects to various core financial systems of enterprises through API interface technology, fundamentally solving the "data silo" problem caused by the independent storage and inability to communicate between systems in the traditional model. After data preprocessing, the constructed financial knowledge graph connects various financial data such as vouchers, invoices, and contracts into a complete network, enabling "one-click retrieval" of financial chain data that previously required manual searching across multiple systems. This provides a solid data foundation for fast and accurate retrieval. At the same time, through professional data cleaning and standardization, duplicate data is removed, abnormal data is corrected, and data format standards are unified, effectively avoiding retrieval errors caused by chaotic data formats and inconsistent information, and improving the overall quality of financial data.

[0014] Keyword search, combined condition search, and natural language search modes cover various core scenarios such as daily financial operations and management decision analysis, significantly shortening the time spent searching for financial information. For daily business processing by financial personnel (such as retrieving tax filing data), it can greatly reduce the time spent on repetitive operations; for management to obtain decision-making data, there is no need to wait for financial personnel to prepare in advance, enabling quick access to core data and reducing the overall human resource input cost of the finance department.

[0015] The "role-permission-data scope" three-level permission control model based on the enterprise's organizational structure enables refined permission division of financial data. By clarifying the viewing permissions and data access scope of different job roles, it ensures that personnel in each position can only access financial information within their scope of responsibility, effectively avoiding the risks of vague permission control and easy leakage of departmental data in the traditional model. Attached Figure Description

[0016] Figure 1 This is a flowchart of the present invention; Figure 2 This is a system block diagram of the present invention; Figure 3 This is a flowchart of the rapid financial statement review process of the present invention; Figure 4 This is a business block diagram of the financial inquiry system of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "sleeved with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0019] This invention provides, for example Figure 1-4 The method for quickly accessing financial information, as shown, enables efficient aggregation, accurate retrieval, and correlation tracing of multi-source financial data. It includes the following functional steps: Step 1: Multi-Source Data Fusion Preprocessing: To achieve standardized integration and correlation of financial data, comprehensive data collection is conducted, including vouchers, ledgers, and reports from the accounting system; all contract data from the contract management system; tax return details and declaration data from the tax system; and the entire lifecycle data of invoices from the invoice management system, such as issuance, authentication, and deduction. This forms a multi-source heterogeneous financial data set. Quality control processing is performed on the collected data. Duplicate data from the same source is removed using deduplication algorithms. An anomaly detection model is used to identify and correct anomalies such as numerical format errors, date inconsistencies, and discrepancies in amount reconciliation. Data loss and inconsistent unit information are also addressed simultaneously. Based on pre-defined financial data standards and specifications, the heterogeneous data from different systems are unified into a standardized format, including the use of "YYYY-MM" for dates. The format is "-DD". Amounts are uniformly expressed in "yuan" and retained to two decimal places. Classified data such as departments and business types use preset enumeration values. Multi-dimensional association tags are added to the standardized data. The basic tags are used to define the core attributes of the data and should at least cover the data type, generation date, department, business category, and data source system. The association identification tags are used to establish cross-system data associations and should at least include unique identification information such as the associated contract number, invoice number, voucher number, and tax batch number. Using core financial data such as vouchers, contracts, and invoices as nodes and multi-dimensional association tags as the association edges between nodes, a financial data knowledge graph is constructed to realize the visualization and structured storage of the association relationships between different types and sources of financial data, laying the foundation for cross-source data retrieval. Step Two: Personalized Permissions and Demand-Based Configuration: To achieve refined permission control and optimized personalized operation experience, a three-tiered progressive permission control model of "role-permission-data scope" is constructed. Based on the enterprise's organizational structure and financial data confidentiality requirements, core roles such as financial operators, financial supervisors, department heads, and management are preset, and the basic access permission boundaries for each role are clearly defined. Permissions include dimensions such as data access scope, operation permissions (e.g., viewing, exporting, and annotating), and data granularity (e.g., details and summaries). By collecting users' historical access behavior data, behavioral analysis algorithms are used to mine users' frequently accessed data types, typical access periods, commonly used related data combination patterns, and search keyword preferences, generating personalized access preference profiles that include user operation preferences and data demand characteristics. The profile update cycle can be set to 3-4 months according to the enterprise's business frequency. Based on the user's role's basic permissions and personalized access preference profiles, a personalized access interface is intelligently configured. The interface needs to optimize functional partitions, with the core data display area prioritizing the presentation of data that users frequently focus on, the commonly used search entry area integrating commonly used search condition templates and keyword input boxes, and the related data quick retrieval area preset with entry points for data categories that users frequently associate, thus simplifying the operation path. Step 3: Multi-mode Intelligent Search and Precise Positioning: To meet the diverse search needs of different users and improve search accuracy, the system supports receiving search requests in various search modes, including keyword search, multi-condition combination search, and natural language search. Keyword search supports switching between fuzzy and exact matching; condition combination search supports logical combinations of multi-dimensional conditions such as time, department, data type, and amount range; and natural language search supports parsing colloquial query statements. Semantic parsing is performed on all types of search requests, extracting core search elements such as time range, data type, business entity, amount condition, and business keywords through natural language processing technology. This is combined with a pre-set thesaurus of financial terms to refine the search results. The system is expanded and optimized, with the thesaurus covering the correspondences of financial terminology, common expressions, and industry jargon to generate precise search instructions. Based on these instructions, the system queries and traverses the financial data knowledge graph. First, it uses precise matching of associated tags to initially screen and locate candidate data sets, narrowing the search scope. Then, it employs a semantic similarity algorithm to calculate the matching degree between candidate data and the search request, sorting them in descending order of matching degree to select target data that meets the matching requirements. A search result optimization mechanism is established: if the search results are empty or the matching degree is below a preset threshold (adjustable according to business needs), it automatically returns semantically similar search suggestions. Simultaneously, it provides an entry point for human assistance, connecting with professional financial personnel to provide search support. Step 4: Intelligent Display and Full-Link Traceability of Related Data: To achieve intuitive data presentation and traceability of data sources, based on the preset display priority rules in the user preference profile, target data is displayed in the core data area of ​​the personalized interface according to the priority determined by dimensions such as user job requirements and viewing frequency, ensuring that core information is presented first; based on the relationship constructed in the financial data knowledge graph, the direct and indirect related data of the target data are automatically identified and extracted, and displayed in a list form in the related data quick retrieval area, supporting quick jump to view related data by clicking on the data icon. The scope of related data includes but is not limited to corresponding contracts, invoices, vouchers, and ledger records; the complete generation link of the target data is displayed in the form of a visual flowchart, which must cover core business nodes such as "contract signing → invoice issuance → voucher entry → ledger registration". Each node is marked with key information such as data number, generation time, and operator, and supports direct jump to the data details page corresponding to any node by clicking on any node, realizing full-link traceability of data from business source to financial record; Step 5: Incremental Updates and Dynamic Optimization of Retrieval Strategies: To ensure data timeliness and continuous improvement in retrieval efficiency, an incremental update mechanism is adopted. Data changes across multiple financial systems are monitored in real-time via an interface or through scheduled scanning. Targeted processing is performed only on newly added, modified, or deleted financial data, including data cleaning, tag updates, and adjustments to knowledge graph relationships, avoiding resource consumption and time delays caused by full updates. Retrieval log data is collected and analyzed regularly. Logs must cover core indicators such as search keywords, search modes, search time, result click-through rate, frequency of related data retrieval, search time, and user feedback. Log analysis helps uncover trends in search demand, search bottlenecks, and potential user needs. Based on the analysis results, the retrieval system is dynamically optimized, including updating the thesaurus, adjusting the weight of search elements, optimizing data display priority, and adding quick templates for high-frequency search scenarios, thereby continuously improving retrieval performance and user experience.

[0020] In step one, abnormal data specifically includes numerical format errors (such as amounts containing non-numeric characters or abnormal decimal places), date logic contradictions (such as dates exceeding a reasonable range or earlier than the business occurrence date), inconsistent data reconciliation relationships (such as unequal debit and credit amounts on vouchers, inconsistencies between report data and ledger data), and duplicate data entry. Standardization processing requires the unification of data format specifications, including not only date and amount formats, but also unified enumeration values ​​for categorized data such as department name, business type, and data status, as well as unified data coding rules. Core financial data is defined as data that plays a core supporting role in financial review, including not only voucher and contract data, but also invoices and key report data. Basic tags need to comprehensively cover the core attribute dimensions of the data, including not only data type, generation date, department, and business type, but also data status, operator, and data source system. Association identification tags are used to establish unique associations across data types, including not only associated contract numbers, invoice numbers, and voucher numbers, but also associated tax batch numbers, payment record numbers, and business order numbers, ensuring that data from different systems can be accurately associated through identification tags.

[0021] In step two, the three-level access control model adopts a logic that combines access inheritance and subdivision. The basic access permissions of each role are precisely defined according to their responsibilities. The basic access permissions of the financial operator are to view detailed data such as vouchers and invoices of their own department and business-related departments, and they have basic search and related retrieval permissions. The financial supervisor has access to view detailed and summary data such as vouchers, ledgers, and reports of the entire department, and also has operation permissions such as data export, access approval, and abnormal data annotation. Department heads can only access financial data directly related to their department's business, such as contracts, invoices, sales reports, and expense reports. The scope of data is strictly limited to the boundaries of departmental responsibilities. Management has access to macro data, including company-wide summary reports, core data from each department, and trend analysis data, and supports multi-dimensional data capture and analysis. The analysis of personalized viewing preferences uses a combination of statistical analysis and machine learning. In addition to high-frequency data types and viewing periods, it also analyzes data display preferences (such as tables and charts), habits of retrieving related data, and preferences for setting search conditions. The core area and related areas of the personalized interface are precisely adapted to job positions. For example, the core area for cost accountants displays cost accounting vouchers and details by default, while the related area displays corresponding purchase invoices and warehouse receipts. The core area for R&D department heads displays R&D expense reports and project cost details by default, while the related area displays corresponding R&D contracts and expense vouchers.

[0022] In step three, the financial thesaurus is constructed using a basic thesaurus plus dynamic updates. The basic thesaurus covers the correspondence between financial professional terms and commonly used expressions. In addition to "invoice" and "note", "expense" and "spending", "receipt" and "receipt voucher", it also includes "accounts payable" and "payables", "operating revenue" and "revenue", "depreciation" and "amortization", etc. The semantic similarity algorithm uses a cosine similarity algorithm combined with a financial domain weight adjustment mechanism to assign higher weights to core financial terms and improve matching accuracy.

[0023] In step four, the priority of display is determined using a multi-dimensional weighted scoring mechanism. The weighting factors include the weight of core job requirements (e.g., finance personnel focus on detailed data, while management focuses on summary data), the weight of access frequency, and the weight of data importance. Among these, the weight of voucher data for financial accountants and the weight of report data for management are set to the highest. Related data is divided into directly related data (e.g., invoices and contracts directly corresponding to vouchers) and indirectly related data (e.g., payment records corresponding to invoices and order data corresponding to contracts) according to the degree of correlation. The quick retrieval area is sorted and displayed according to the degree of correlation. The core nodes of the data generation chain can be flexibly configured according to the enterprise's business type. In addition to the basic nodes, manufacturing enterprises can add the "inbound order → outbound order" node, and retail enterprises can add the "sales order → payment record" node. In addition to basic elements, the node information can also display data status (e.g., invoices have been certified, contracts have been fulfilled), related business remarks, and other information.

[0024] In step five, incremental updates employ a combination of real-time monitoring and scheduled verification. Real-time monitoring is achieved through the real-time data push interface of each system. For systems without a real-time interface, a scheduled scanning method is used, with a scanning cycle set to 5–30 minutes. The analysis cycle of the retrieval logs is matched with the update and optimization cycle, employing a mechanism of weekly preliminary analysis and monthly in-depth optimization. In addition to the thesaurus, retrieval element weights, and display priorities, optimizations include adjustments to retrieval algorithm parameters (such as similarity calculation weights), caching optimization for high-frequency retrieval scenarios, analysis of the causes of retrieval failures, and optimization of solutions. Through continuous optimization, both retrieval accuracy and efficiency are improved.

[0025] A financial information rapid retrieval system includes the following functional modules: a data preprocessing module, used to collect, control, standardize, tag, and construct knowledge graphs from multiple sources of financial data, providing a high-quality foundation of related data for subsequent retrieval; a user configuration module, used to build a refined permission control system, mine user viewing habits and generate preference profiles, and configure personalized operation interfaces to achieve a balance between permission security and operational convenience; a retrieval and positioning module, used to parse multi-mode retrieval requests, generate precise retrieval instructions, query the knowledge graph and sort and filter target data, and handle retrieval anomalies to achieve accurate responses to diverse retrieval needs; a display and traceability module, used to display target data according to personalized preferences, intelligently retrieve related data, and visually present the data generation chain to achieve intuitive data display and full-chain traceability; an update and optimization module, used to monitor data changes in real time and perform incremental updates, analyze retrieval logs and optimize retrieval strategies to ensure data timeliness and continuous improvement in retrieval performance; and a data storage module, used to securely store preprocessed data, knowledge graphs, user permission information, and preference profiles to provide data support for all modules of the system.

[0026] The data preprocessing module adopts a modular, collaborative architecture, including a data acquisition unit, a data cleaning unit, and a tagging and graph construction unit. The data acquisition unit connects with multiple source systems such as ERP accounting systems, contract management systems, tax systems, and invoice management systems through various adaptation methods including API interfaces, direct database connections, and file imports. It supports both real-time and batch acquisition modes, ensuring comprehensiveness and flexibility in data collection. The data cleaning unit combines a rule engine with intelligent detection, incorporating a built-in data quality detection rule library to automatically perform operations such as duplicate data removal, abnormal data correction, missing data completion, and logical inconsistency handling, outputting high-quality, standardized data. The tagging and graph construction unit automatically adds multi-dimensional tags to the data based on preset tagging rules and constructs a knowledge graph with core financial data as nodes and related tags as edges using graph database technology, supporting real-time updates and relational queries of the knowledge graph.

[0027] The user configuration module adopts a layered architecture design, including a permission management unit, a preference analysis unit, and an interface configuration unit. The permission management unit constructs a three-level control system of "role-permission-data scope" based on the RBAC (Role-Based Access Control) model, supporting role customization, fine-grained permission configuration, and dynamic permission adjustment. The built-in permission verification engine ensures real-time permission verification during user operations to prevent unauthorized access. The preference analysis unit collects user operation logs and uses behavior mining algorithms to extract user browsing habit characteristics, generating a structured personalized preference profile that supports automatic updates and manual adjustments. The interface configuration unit uses component-based interface generation technology based on the preference profile to automatically configure the layout and content of the core data display area, frequently used search entry area, and related data retrieval area, supporting users to make personalized fine-tuning adjustments to the interface layout and save them.

[0028] The display and tracing module adopts a front-end and back-end separation architecture, including a personalized display unit, an association retrieval unit, and a tracing visualization unit. The personalized display unit, based on priority rules in the user's preference profile, uses a data rendering engine to display target data in user-preferred formats such as tables and charts, supporting interactive operations such as data sorting, filtering, and annotation. The association retrieval unit generates a list of related data by parsing the relationships in the knowledge graph and provides a one-click jump interface, supporting batch export and comparative analysis of related data. The tracing visualization unit uses a flowchart drawing engine to automatically generate a generation link diagram of the entire data lifecycle, supporting the expansion / collapse of link nodes, jump viewing, and export of the link diagram. The update and optimization module... The module includes an incremental update unit and a retrieval optimization unit. The incremental update unit has a built-in data change monitoring engine that captures changes in multi-source system data in real time and performs incremental processing to ensure data timeliness. The retrieval optimization unit uses a log analysis engine to discover retrieval optimization points and automatically performs optimization operations such as thesaurus updates and retrieval weight adjustments, while supporting manual configuration of optimization strategies. The data storage module adopts a hybrid distributed storage architecture of "SSD+HDD". SSD is used to store frequently accessed data (such as vouchers, reports, and user preference profiles from the past 3 months) to improve access speed. HDD is used to store historical archived data to reduce storage costs. It has a built-in data backup and recovery mechanism to ensure the security and reliability of data storage.

[0029] Example 1 1.1 Implementation Preparation Hardware configuration: Deploy one main server (CPU: Intel Xeon Gold 6330, memory: 64GB, hard drive: 1TB SSD + 10TB HDD hybrid storage), upgrade the terminal computers of each department to Windows 10 or above; optimize the network environment to gigabit LAN to ensure network stability during multi-source data collection and retrieval.

[0030] Software tools: Python 3.9 is used as the development language, PySpark is used for data processing, Neo4j is used to build the knowledge graph, Elasticsearch is used to implement the retrieval function, and Vue.js is used to develop the front-end interface; a pre-set thesaurus is used (supplementing industry-related synonyms such as "equipment" and "machinery", "payment" and "receipt"), and data standardization rules are established (dates are uniformly formatted as "YYYY-MM-DD", amounts are retained to two decimal places and the unit is "yuan", and department names are uniformly formatted as "R&D Department, Production Department, Sales Department, Finance Department, Management Department").

[0031] II. Implementation Process of the Method for Quickly Accessing Financial Information 2.1 Step 1: Data Preprocessing Implementation 2.1.1 Multi-source data collection: Through API interfaces, we connect to the Yonyou U9 ERP accounting system, Fanwei contract management system, Golden Tax Phase IV tax system, and Baiwang invoice management system to collect all data from January 1, 2022 to June 30, 2024, including 120,000 voucher data, 80,000 ledger data, 5,000 report data, 3,000 contract data, 150,000 tax return data, and 200,000 invoice data.

[0032] 2.1.2 Data Cleaning and Standardization: PySpark was used to clean the collected data, remove duplicate data (such as 2300 duplicate entries for the same invoice), and correct abnormal data (such as changing "2024 / 08 / 32" to "2024 / 08 / 31" and the amount "10000 yuan" to "10000.00 yuan"). The data was then standardized according to preset rules to unify the data format.

[0033] 2.1.3 Adding Multi-Dimensional Tags: Add basic tags and associated identifier tags to the standardized data. Taking a sales contract (contract number HT-2024-0056) as an example, the basic tags are "Data type: contract, generation date: 2024-03-15, department: sales department, business type: machinery and equipment sales", and the associated identifier tags are "Associated invoice numbers: FP-2024-1235 to FP-2024-1238, associated voucher numbers: PZ-2024-0568 to PZ-2024-0570".

[0034] 2.1.4 Knowledge Graph Construction: Using core financial data such as vouchers, contracts, and invoices as nodes and associated tags as edges, a financial data knowledge graph is constructed using Neo4j. For example, using voucher PZ-2024-0568 as the core node, the contract HT-2024-0056 node is connected via an edge associated with the contract number, the invoice FP-2024-1235 node is connected via an edge associated with the invoice number, and the general ledger Z-2024-03 node is connected via an edge associated with the ledger, thus establishing relationships between multiple types of data.

[0035] 2.2 Step Two: Implementation of User Permissions and Requirement Configuration 2.2.1 Establishment of a Three-Tier Access Control Model: Based on the enterprise organizational structure, four roles are preset: financial operator, financial supervisor, department head, and management. Corresponding permissions are configured, as follows: Financial operators have basic access permissions to view and enter voucher data, as well as retrieve related invoices and contract data. The data they can access is limited to relevant data from their own department (Finance Department) and related business departments (such as Sales Department and Production Department). The Finance Supervisor has the authority to view all vouchers, ledgers, and reports from all departments, as well as approval and data export permissions, and can access all financial data of the entire enterprise. The Sales Department Head can only view contract, invoice, and sales report data related to their own department, and the data scope is limited to financial data related to the sales department's business. Management can view the enterprise-wide summary reports, core data from each department, and trend analysis data, covering the enterprise-wide financial summary data and core data from each department.

[0036] 2.2.2 Personalized Preference Profile Generation: User historical browsing data from January 2023 to June 2024 was collected, and browsing habits were mined using Python analysis tools. For example, the financial operator's frequently viewed data types were "sales vouchers and corresponding invoices", and the browsing time was concentrated in "the 1st to the 10th of each month (tax filing period)". The commonly used related data combination was "voucher-invoice-contract", and a personalized browsing preference profile was generated for him. Mr. Li, the general manager of the sales department, frequently viewed "monthly sales reports and large contract data", and the browsing time was concentrated in "the end of each month and the end of each quarter". This feature was marked in the preference profile.

[0037] 2.2.3 Personalized Interface Configuration: Based on roles and preference profiles, interfaces are configured for each user. The core data display area of ​​the financial operator's interface defaults to "Sales Voucher Details", the commonly used search entry is set to "Invoice Number Search, Contract Number Search", and the quick access area for related data defaults to "List of Corresponding Invoices and Contracts". The core data display area of ​​General Manager Li's interface defaults to "Monthly Sales Report", the commonly used search entry is set to "Quarterly Search, Large Contract Search (Amount ≥ 500,000 RMB)", and the quick access area for related data defaults to "List of Corresponding Contracts and Payment Data".

[0038] 2.3 Step 3: Implementation of Multi-Mode Retrieval and Data Location Using three typical user scenarios as examples, the multi-modal retrieval process is demonstrated: Scenario 1: A finance operator needs to query "vouchers and invoices corresponding to the machinery and equipment sold by the sales department to Company A in March 2024" (keyword search). After receiving the search request, the system semantically analyzes and extracts the core elements "time range: 2024-03, department: sales department, customer: Company A, business type: machinery and equipment sales, data type: vouchers, invoices". Based on the thesaurus, "machinery and equipment" is expanded to "equipment, machinery", generating a search instruction. When querying the knowledge graph, the system initially filters out the candidate data set (containing 3 vouchers and 4 invoices) through the association tags "sales department + 2024-03 + Company A". Then, the matching degree is calculated using the cosine similarity algorithm, and the data is sorted in descending order of matching degree (all matching degrees ≥ 85%). The target data is then returned. The entire process takes about 8 seconds.

[0039] Scenario 2: Finance manager Lao Wang needs to query "R&D department expense vouchers for Q1-Q2 2024, amount ≥ 100,000 yuan" (condition combination search). After the system analyzes and extracts multi-dimensional condition elements, it expands "expenses" to "expenses" to generate search instructions. It initially screens and locates 12 candidate vouchers. After sorting by semantic similarity, it returns 10 target vouchers with a matching degree ≥ 90%, which takes about 10 seconds.

[0040] Scenario 3: Management inquires about the contracts and payment records of the top 3 customers with the highest sales revenue in the first half of 2024 (natural language retrieval). The system uses natural language processing technology to analyze the core elements: "Time range: January 2024 to June 2024, Data type: Sales data, Contracts, Payments, Filter criteria: Top 3 customers in sales revenue," generates a search command, queries the knowledge graph, first locates the top 3 customers in sales revenue in the first half of the year (Company A, Company B, and Company C), then retrieves their corresponding contract and payment data, sorts them, and returns the results, taking approximately 12 seconds.

[0041] Handling Abnormal Situations: When a finance operator attempts to query "Sales Department Vouchers for December 2024", the system parsing detects an abnormal time range and returns the message "Incorrect time format, please enter the correct year and month (format: YYYY-MM)". When querying "Business data that did not occur in August 2024", the search results are empty, and the system returns a similar suggestion, "No relevant data found, we recommend similar business data from July 2024", and provides an access point for manual assistance (connecting to the finance department's customer service).

[0042] 2.4 Step Four: Implementation of Related Data Display and Traceability Taking voucher PZ-2024-0568 queried by the financial operator in Scenario 1 as an example, the system displays voucher details (voucher number, date, summary, debit amount, credit amount, preparer, etc.) in the core data area of ​​its personalized interface. The display priority is set according to the financial operator's preference as "voucher date > amount > summary". In the quick access area of ​​related data, a list of "related invoices (FP-2024-1235 to FP-2024-1238), related contracts (HT-2024-0056), and related ledger records (Z-2024-03-008)" is displayed. Clicking on the invoice number FP-2024-1235 will directly jump to the invoice details page.

[0043] Data generation chain visualization: The generation chain of this voucher is displayed in the form of a flowchart: "Contract signing (HT-2024-0056, 2024-03-15) → Invoice issuance (FP-2024-1235 to FP-2024-1238, 2024-03-20) → Voucher entry (PZ-2024-0568, 2024-03-25) → Ledger registration (Z-2024-03-008, 2024-03-31)". Each node is marked with key information (number, date). Clicking the "Contract signing" node will directly jump to the contract details page.

[0044] 2.5 Step Five: Data Update and Retrieval Optimization Implementation Incremental update implementation: An incremental update mechanism was adopted, and data changes in each source system were monitored in real time through the system interface. On July 1, 2024, the sales department added a new contract (HT-2024-0120) and issued an invoice (FP-2024-2560). The system captured the new data in real time, and only cleaned (no duplicates or anomalies), standardized and added tags to the contract and invoice data. The association relationship of the "Sales Department-2024-07" node in the knowledge graph was updated. The entire update process took about 3 seconds and did not affect the integrity of the existing data.

[0045] Search optimization implementation: Weekly analysis of search logs (including search keywords, result click-through rates, frequency of related data retrieval, etc.) revealed that users frequently used the search term "payment back," but the initial thesaurus did not include the correspondence between "payment back" and "payment records," resulting in low match rates for some search results. Searches related to "R&D expenses" had a click-through rate as high as 60%. Based on this, the following optimizations were made: the correspondence between "payment back" and "payment records" was added to the thesaurus; the weight of search elements related to "R&D expenses" was adjusted, increasing it from 0.8 to 1.2; and a dedicated search entry for "R&D expenses" was added for the head of the R&D department. After optimization, the match rate of related searches increased to 95%, and the search time was shortened to 6 seconds.

[0046] III. Deployment and Verification of the Financial Information Quick Access System 3.1 System Deployment The system described in this invention includes a data preprocessing module, a user configuration module, a retrieval and positioning module, a display and tracing module, an update and optimization module, and a data storage module. The deployment and functional implementation of each module are as follows: Data preprocessing module: Deployed on the main server, it uses PySpark to collect multi-source data (API interface to connect with various source systems), clean and standardize the data, and uses Neo4j to build a knowledge graph; it includes a data collection unit (responsible for real-time / scheduled data collection), a data cleaning unit (performs data cleaning and standardization), and a labeling and graph construction unit (adds labels and updates the knowledge graph).

[0047] User configuration module: Deployed on the main server, developed based on the Java language, it builds a three-level permission control model of "role-permission-data scope" and generates user preference profiles through Python analysis tools; it includes a permission management unit (configuring role permissions and dynamically adjusting them), a preference analysis unit (mining user browsing habits), and an interface configuration unit (generating personalized interfaces).

[0048] Search and Locating Module: Deployed on the main server, it uses Elasticsearch to implement multi-modal retrieval and parses search requests using natural language processing technology; it includes a request parsing unit (parses search requests and expands elements), an intelligent search unit (queries the knowledge graph and sorts), and a result optimization unit (returns results and provides exception handling).

[0049] The traceability module is developed using Vue.js and deployed on various user terminals. It includes a personalized display unit (displaying target data according to preferences), an associated retrieval unit (displaying associated data and supporting navigation), and a traceability visualization unit (displaying the generation chain in the form of a flowchart).

[0050] Update and optimization module: Deployed on the main server, it monitors data changes in real time and performs incremental updates, and regularly analyzes logs to optimize retrieval strategies; it includes an incremental update unit (real-time updates of newly added / modified data) and a retrieval optimization unit (optimizing the thesaurus and retrieval weights).

[0051] Data storage module: It adopts a hybrid distributed storage architecture of 1TB SSD + 10TB HDD. The SSD stores frequently accessed data (such as vouchers and reports from the past 3 months), and the HDD stores historical archived data. It is responsible for storing pre-processed data, knowledge graphs, user permission information and preference profiles, and supports data backup (automatic backup every morning).

[0052] 3.2 System Verification Verification period: July 1, 2024 to August 31, 2024, a total of 2 months.

[0053] Functional verification: Twenty test users (including 5 financial operators, 3 financial supervisors, 5 department heads, and 7 management personnel) were selected to simulate 100 different search scenarios. The results showed that the system supports three search modes: keyword, condition combination, and natural language. The accuracy rate of target data location reached 98%, the success rate of related data retrieval reached 100%, the data generation chain was clearly visualized, and the access control was precise (no unauthorized access occurred).

[0054] Performance verification: Key performance indicators were compared before and after system implementation. The results are shown in the table below: 1. Time spent retrieving a single business financial chain: Before implementation, the average time was about 40 minutes, and after implementation, it was reduced to about 10 seconds, an improvement of about 99.58%; 2. Management's preparation time for reviewing reports: Before implementation, finance staff need to prepare data 1-2 days in advance; after implementation, the review can be completed in about 15 seconds, an improvement of about 99.89%. 3. Search result accuracy: Before implementation, the accuracy rate was approximately 75%, which improved to approximately 98% after implementation, an increase of approximately 29.33%. 4. Data update time (single new data): Before implementation, manual entry and updates were required, taking about 5 minutes. After implementation, automatic incremental updates take only about 3 seconds, an improvement of about 99.90%.

[0055] User satisfaction verification: The satisfaction of 20 test users was collected through a questionnaire survey. The results showed that 16 people (80%) were very satisfied, 4 people (20%) were satisfied, and the overall satisfaction rate was 100%.

[0056] IV. Implementation Conclusion The financial information rapid retrieval method and system described in this invention can effectively solve the problem of "islands" of multi-source financial data. Through data preprocessing to construct a knowledge graph, three-level access control and personalized configuration, and multi-mode retrieval, it significantly improves the efficiency of financial information retrieval (reducing retrieval time by more than 99%) and improves retrieval accuracy (up to 98%). At the same time, it strengthens access control, ensures data security, and has good system stability and compatibility. It can adapt to the financial information retrieval needs of enterprises of different sizes and has high practical application value.

[0057] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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. A method for quickly accessing financial information, characterized in that, Includes the following steps: Step 1: Data Preprocessing: Collect multi-source financial data, including voucher data, ledger data, and report data from the accounting system, contract data from the contract management system, tax return data from the tax system, and invoice data from the invoice management system. Clean the collected financial data to remove duplicate data, correct abnormal data, and unify the data format through standardization. Step 2, User Permissions and Requirements Configuration: Establish a three-level permission control model of "role-permission-data scope", preset the roles of financial operator, financial supervisor, department head and management and configure the corresponding basic access permissions. Configure a personalized access interface based on user roles and preference profiles. The interface includes a core data display area, commonly used search entry points and a quick access area for related data. Step 3, Multi-mode retrieval and data location: Receive multi-mode retrieval requests from keyword retrieval, condition combination retrieval and natural language retrieval, and perform semantic parsing on the retrieval requests to extract core retrieval elements such as time range, data type, department and business keywords, and generate retrieval instructions based on the preset thesaurus to expand retrieval elements; Step 4: Display and trace related data: Display the target data in the core data area of ​​the personalized interface according to the display priority in the user's preference profile. Automatically identify the related data of the target data based on the financial data knowledge graph. Display the list of related data in the quick access area of ​​related data and support clicking to jump to view. Visualize the generation chain of the target data in the form of a flowchart. Step 5, Data Update and Retrieval Optimization: An incremental update mechanism is adopted to monitor data changes from multiple financial systems in real time. Only newly added or modified financial data is cleaned, tags are updated, and knowledge graph associations are adjusted. Retrieval logs are analyzed regularly.

2. The method for quickly accessing financial information according to claim 1, characterized in that: In step one, the abnormal data includes data with numerical format errors and date logic contradictions. The basic tags include data type, generation date, department, and business type. The associated identification tags include associated contract number, invoice number, and voucher number.

3. The method for quickly accessing financial information according to claim 1, characterized in that: In step two, the basic access permissions for financial operators are to view the voucher data of their own department, while the basic access permissions for management are to view the report data of the entire department. The analysis cycle for personalized access preference profiles is 3 to 4 months. The core area of ​​the personalized interface for financial accountants displays voucher details by default, and the related area displays the corresponding invoices by default. The core area of ​​the personalized interface for sales department heads displays sales reports by default, and the related area displays the corresponding contracts by default.

4. The method for quickly accessing financial information according to claim 1, characterized in that: In step three, the thesaurus contains the correspondence between "invoice" and "receipt", "fee" and "expense", and "refund" and "receipt". The semantic similarity algorithm adopts the cosine similarity algorithm, with a preset threshold of 60% to 70%.

5. The method for quickly accessing financial information according to claim 1, characterized in that: In step four, the display priority is determined based on the user's position. Financial accountants are given priority in displaying voucher details, while management is given priority in displaying summary data. The associated data includes invoices, contracts, and ledger records associated with the vouchers.

6. The method for quickly accessing financial information according to claim 1, characterized in that: In step five, the search log includes search keywords, search result click-through rate, frequency of related data retrieval, optimization of the thesaurus, search element weights, and data display priority.

7. A financial information rapid retrieval system, which can be used to implement the financial information rapid retrieval method according to any one of claims 1-6, characterized in that, It includes a data preprocessing module, a user configuration module, a retrieval and positioning module, a display and traceability module, an update and optimization module, and a data storage module. The data preprocessing module is used to collect multi-source financial data, clean and standardize it, add related tags, and construct a financial data knowledge graph. The user configuration module is used to build a three-level permission control model of "role-permission-data scope", generate personalized user viewing preference profiles and configure personalized viewing interfaces; The retrieval and positioning module is used to parse multi-mode retrieval requests and query the financial data knowledge graph to obtain target data; The display and traceability module is used to display target data and related data according to user preferences, and to visualize the data generation chain. The update and optimization module is used to maintain financial data and knowledge graphs using an incremental update mechanism, and to analyze retrieval logs to optimize retrieval strategies. The data storage module is used to store preprocessed financial data, financial data knowledge graphs, user permission information, and preference profiles.

8. A financial information rapid access system according to claim 7, characterized in that: The data preprocessing module includes a data acquisition unit, a data cleaning unit, and a label and graph construction unit; the data acquisition unit collects multi-source financial data through API interfaces and direct database connections.

9. A financial information rapid access system according to claim 7, characterized in that: The user configuration module includes a permission management unit, a preference analysis unit, and an interface configuration unit. The retrieval and positioning module includes a request parsing unit, an intelligent retrieval unit, and a result optimization unit. The request parsing unit parses multi-mode retrieval requests, expands core retrieval elements based on a preset thesaurus, and generates retrieval instructions.

10. A financial information rapid access system according to claim 7, characterized in that: The display and traceability module includes a personalized display unit, an associated retrieval unit, and a traceability visualization unit. The personalized display unit displays target data according to the user's job priority. The update and optimization module includes an incremental update unit and a retrieval optimization unit. The incremental update unit monitors data changes in real time and performs incremental updates. The data storage module adopts a hybrid distributed storage architecture of 1TB SSD + 10TB HDD.

Images