A multi-person multi-scene reimbursement filling processing system

Abstract

The application relates to the technical field of computers and discloses a multi-person and multi-scene reimbursement filing processing system, which comprises a business activity record creation module, a multi-source trace correlation module, a record completeness evaluation module and an automatic decision module. The core of the application is that the record completeness evaluation module obtains a comprehensive record completeness score by comprehensively calculating scores of five dimensions of evidence coverage, participant confirmation, semantic correlation, time consistency and space aggregation. The automatic decision module automatically executes approval, manual review or rejection and other processing paths according to the numerical interval of the score. The application also realizes real-time compliance verification through an atomization strategy engine and completes subsequent cost allocation through an automatic cost allocation engine. The application changes subjective review into objective quantitative evaluation, and significantly improves the compliance, reliability and automation efficiency of business activity processing.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a multi-user, multi-scenario expense reimbursement form processing system. Background Technology

[0002] In modern enterprise management, the reimbursement and auditing of various business expenses incurred by employees for work-related activities (such as client entertainment, team building, and business travel) are key aspects of financial management and internal control. Current methods typically rely heavily on manual review processes. Employees submit paper or simple electronic documents, along with scattered supporting materials such as invoices and receipts, which are then manually approved level by level by supervisors and finance personnel.

[0003] However, this traditional model has significant inherent flaws. The review process often lacks objective and unified quantitative evaluation standards, and the judgments of approvers rely heavily on personal experience, leading to subjective arbitrariness in assessing the authenticity and compliance of business activities, making it difficult to guarantee the consistency and fairness of the review results. Furthermore, because the evidence is scattered and its relevance to business activities requires manual verification, the review process is time-consuming and labor-intensive, resulting in low overall efficiency. In complex scenarios involving multiple participants and the need to allocate costs across different cost centers, manual calculations are not only prone to errors but also further extend the processing cycle. Simultaneously, compliance checks are often delayed; those submitting reports are often only informed of non-compliance after submission, causing unnecessary communication costs and rework.

[0004] Therefore, how to build an intelligent system that can objectively and quantitatively assess the authenticity and compliance of business activities and automate subsequent processing is a technical problem that urgently needs to be solved in the field of enterprise expense management. Summary of the Invention

[0005] This invention provides a multi-person, multi-scenario expense reimbursement form processing system, which aims to solve the technical problems of existing technologies that rely on manual judgment, lack objective quantitative evaluation basis, and have low processing efficiency when reviewing the authenticity, compliance, and cost allocation of business activities.

[0006] To achieve the above objectives, the present invention provides a multi-user, multi-scenario expense reimbursement form processing system, comprising:

[0007] The system includes a business activity record creation module, a multi-user collaborative management module, a multi-source trace association module, a record completeness assessment module, and an automated decision-making module.

[0008] The business activity record creation module is connected to an atomic policy engine. This module receives the business activity type selected by the user and retrieves the compliance policy from the atomic policy engine based on that type. Based on the retrieved policy, the module generates a structured dataset, defined as the business activity record. The business activity record includes a structured evidence list template and a corresponding set of required evidence items.

[0009] The multi-person collaborative management module operates the business activity record. This module is used to define a set of invited participants in the business activity record, and to receive and record externally input participant participation confirmation response data to generate a set of confirmed participants.

[0010] The multi-source trace association module is configured with an intelligent context association device. This module is used to associate external digital traces, along with the time and location attributes contained within the traces, with the business activity record through a data interface.

[0011] The record completeness assessment module is connected to the business activity record. This module derives a record completeness score by performing a series of calculations. The calculation process includes:

[0012] The evidence coverage is determined by comparing the evidence already provided in the business activity record with the set of required evidence items.

[0013] The participant confirmation rate is determined by calculating the ratio of the number of members in the confirmed participant set to the number of members in the invited participant set.

[0014] By analyzing the logical relationships, temporal attributes, and location attributes among the associated external digital traces, semantic relevance, temporal consistency, and spatial clustering are determined.

[0015] Finally, based on the calculation results of the five dimensions—evidence coverage, participant confirmation, semantic relevance, temporal consistency, and spatial clustering—the module performs a comprehensive calculation and outputs a transcript completeness score. Its calculation can be expressed by the following formula:

[0016] ;

[0017] in, The scores represent the five sub-dimensions, namely evidence coverage. Participant confirmation rate Semantic relevance Time consistency Spatial aggregation ; These are the preset weighting coefficients corresponding to the scores of each sub-dimension, and .

[0018] The automated decision-making module receives the record completeness score output by the record completeness assessment module. Based on the preset range to which the record completeness score falls, this module selects and executes a corresponding processing path.

[0019] Preferably, the function of the business activity record creation module in generating the evidence list template is implemented through the following steps:

[0020] The first step is to query the strategy library of the atomic strategy engine based on the business activity type and filter all compliance strategies associated with the business activity type.

[0021] The second step is to extract the evidence requirements corresponding to the compliance strategy, and to integrate and deduplicate all the extracted evidence requirements to generate the final evidence list template.

[0022] Preferably, the participation confirmation response data includes a timestamp and a digital signature for verifying the authenticity of the data.

[0023] In one specific embodiment, the intelligent context association device determines the association through the following steps:

[0024] First, the similarity between the timestamp of the external digital trace and the activity period defined in the business activity record is calculated to obtain the time similarity. ;

[0025] Secondly, the spatial similarity is obtained by calculating the proximity between the geographical location of the external digital trace and the preset location of the business activity record. ;

[0026] Next, the semantic similarity between the text content of the external digital trace and the text description of the business activity record is calculated to obtain the text similarity score. ;

[0027] Finally, a matching degree is obtained by weighting the three types of similarity mentioned above. Its calculation can be expressed by the following formula:

[0028] ,in, , , These are preset weighting coefficients. The matching degree is used to recommend the optimal placement to the user.

[0029] Preferably, the system further includes an automated cost allocation engine. The automated cost allocation engine is connected to the business activity log and is used to perform automated cost allocation calculations on associated external digital traces whose expense types are marked as shared, based on members in the confirmed participant set.

[0030] Preferably, the atomic strategy engine is also used to trigger real-time compliance verification when the data in the business activity record is updated due to the association of records involved in the confirmation response or external digital traces.

[0031] In one specific embodiment, the record integrity assessment module determines temporal consistency. This is achieved through the following steps:

[0032] The first step is to iterate through all external digital traces with time attributes that are associated with the aforementioned business activity records, obtaining a list containing... A collection of timestamps ;

[0033] The second step is to process each timestamp in the set. Through a preset time scoring function Calculate its score.

[0034] like The function outputs a perfect score within the activity cycle of the business activity record.

[0035] If it is outside the cycle, the function will calculate a penalty score based on the deviation time according to a preset rule;

[0036] The third step is to take the arithmetic mean of all scores to obtain the time consistency. Its calculation can be expressed by the following formula:

[0037] ;

[0038] In one specific embodiment, the record integrity assessment module determines spatial clustering. This is achieved through the following steps:

[0039] The first step is to acquire all external digital traces with location attributes that are associated with the aforementioned business activity record, and to calculate the average geographical distance between these location points. ;

[0040] The second step is to read a maximum acceptable distance threshold from the configuration library based on the business activity type. ;

[0041] The third step is to use a pre-defined spatial scoring function. The spatial clustering is calculated based on the ratio of the average geographical distance to the threshold. .

[0042] Preferably, the automated decision-making module selects the processing path through the following steps:

[0043] When the received record completeness score is not lower than a preset high score threshold, a fully automated approval process is triggered.

[0044] When the received record completeness score falls within a preset middle range, a simplified manual review process is triggered.

[0045] When the received record completeness score is lower than a preset low score threshold, a complete manual review process is triggered.

[0046] Preferably, the allocation strategy type required by the automated cost allocation engine when performing allocation calculations is provided by the strategy library of the atomic strategy engine.

[0047] This invention provides a multi-user, multi-scenario expense reimbursement form processing system. It has the following beneficial effects:

[0048] 1. This invention achieves an objective and quantitative assessment of business activity records by setting up a record completeness evaluation module. This module comprehensively calculates scores across five dimensions: evidence coverage, participant confirmation, semantic relevance, temporal consistency, and spatial clustering, and ultimately outputs a comprehensive record completeness score. This approach transforms the previous subjective judgment, which relied on human experience, into a calculation process based on multidimensional data that is repeatable and traceable, thereby improving the consistency and reliability of the evaluation results.

[0049] 2. This invention improves the completeness and compliance of business activity records during the creation stage by linking the business activity record creation module with the atomic strategy engine. The system automatically retrieves the corresponding compliance strategy from the strategy library based on the business activity type selected by the user, and generates a structured evidence list template and a set of required evidence items accordingly. This mechanism provides clear guidance to users at the source of information entry, effectively reducing the subsequent correction work caused by missing or non-compliant evidence, and lowering processing costs.

[0050] 3. By setting up an automated decision-making module and an automated cost-sharing engine, this invention significantly improves the automation level and efficiency of business processing. On the one hand, the automated decision-making module automatically selects different processing paths based on the numerical range of the record completeness score, realizing the diversion of records with different risk levels. On the other hand, the automated cost-sharing engine uses the generated set of confirmed participants to automatically calculate the shared costs. These two functions reduce the manual judgment and calculation steps in the process and shorten the overall processing cycle. Attached Figure Description

[0051] Figure 1 This is a functional module block diagram of a multi-person, multi-scenario expense reimbursement form processing system according to an embodiment of the present invention;

[0052] Figure 2 This is a system overall business process diagram according to an embodiment of the present invention;

[0053] Figure 3 This is a branch diagram of the automated decision-making logic of one embodiment of the present invention;

[0054] Figure 4 This is an example diagram of a user interface according to an embodiment of the present invention;

[0055] Figure 5 This is a schematic diagram of the automated cost allocation result according to an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions in 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.

[0057] Please refer to the appendix. Figure 1 , Figure 1 This is a functional block diagram of a multi-person, multi-scenario expense reimbursement form processing system according to an embodiment of the present invention.

[0058] This invention provides a multi-user, multi-scenario expense reimbursement processing system that can be deployed on one or more server devices. The server device includes a processor, a memory, and a network interface for communicating with other devices. The memory stores a computer program that, when executed by the processor, implements the various functional modules of the system. System data, such as the business activity records and the strategy library of the atomic strategy engine described below, can be stored in the server's memory or a separate database system.

[0059] Furthermore, the system can be deployed not only on physical servers but also on cloud servers, virtual machines, or containerized environments (such as Docker and Kubernetes) to achieve elastic scaling and high availability. In a preferred embodiment, the various functional modules of the system can be implemented as independent microservices, which communicate with each other through application programming interfaces (APIs) or message queues (such as RabbitMQ and Kafka), thereby improving the system's flexibility, maintainability, and scalability. These modules work logically together but can be physically distributed across different server nodes.

[0060] As attached Figure 1 As shown, the system includes: a business activity record creation module 10, a multi-person collaborative management module 20, a multi-source trace association module 30, a record completeness evaluation module 40, an automated decision-making module 50, an atomic strategy engine 60, and an automated cost allocation engine 70.

[0061] The Atomized Policy Engine 60 is the system's rule center. Its internal policy library stores the evidence requirements, compliance rules, and cost-sharing rules needed for various business activities, and provides policy query services for other modules.

[0062] The business activity record creation module 10 receives the business activity type selected by the user, queries the atomic strategy engine 60 for the corresponding strategy, and generates a structured "business activity record" as the core data carrier.

[0063] The multi-person collaboration management module 20 is responsible for managing participants in business activities. It defines invited participants in the record and updates the "confirmed participant set" after receiving a valid participation confirmation response.

[0064] The multi-source trace association module 30 uses its built-in intelligent context association tool to intelligently associate and write external digital traces (such as electronic invoices) containing time and location attributes into the corresponding business activity record.

[0065] The record completeness assessment module 40 is the core assessment component of the system. It reads various data from the record of business activities and uses algorithms to comprehensively calculate five dimensions: evidence coverage, participant confirmation, semantic relevance, temporal consistency, and spatial clustering. Finally, it outputs a quantitative record completeness score.

[0066] The automated decision-making module 50 receives the score and, based on the preset decision threshold, automatically selects and executes the subsequent processing path, such as direct approval or transfer to manual review.

[0067] After receiving the allocation instruction, the automated cost allocation engine 70 performs automated calculations on the participants and shared costs confirmed in the business activity record, based on the allocation rules obtained from the atomic strategy engine 60.

[0068] During system operation, each module achieves information exchange and functional collaboration by creating, reading, updating, and deleting the central data object, namely the business activity record. For example, when the multi-person collaborative management module 20 updates the confirmed participant set, or the multi-source trace association module 30 associates a new trace, the atomic strategy engine 60 can be triggered to perform a real-time compliance check and update the check result to the business activity record.

[0069] The Atomized Policy Engine 60 is a core component that provides rule bases to other functional modules within the system. Internally, this engine maintains a policy library that stores a series of structured policy data units. These data units are stored digitally in the system's memory or a separate database.

[0070] In one specific embodiment, each policy data unit may be organized in JSON (JavaScript Object Notation) format, which defines all the rules that should be followed for a particular type of business activity.

[0071] A typical strategy data unit consists of: a strategy identifier (strategyId), an applicable business activity type (activityType), and a set of rules. The rule set specifically includes evidence requirements, compliance rules, and cost allocation rules.

[0072] The evidence requirements are an array, where each element defines a required or optional piece of evidence. Each element contains the following fields:

[0073] An evidence code (evidenceCode) is used for machine recognition; an evidence description (description) is used to present the evidence to the user; and a Boolean flag indicating whether it is required (isMandatory).

[0074] When the business activity record creation module 10 requests a strategy from the atomic strategy engine 60, the module extracts all the description fields of the evidence requirements to generate an evidence list template, and extracts all the evidence codes marked as true to generate a set of required evidence items.

[0075] Compliance rules define the constraints that business activities must meet, such as monetary limits, time limits, or location restrictions. When the data in the business activity record is updated due to user operations, such as when the multi-user collaboration management module 20 records a new participation confirmation response, or the multi-source trace association module 30 associates a new external digital trace, the system can call the atomic policy engine 60 to use these compliance rules to perform real-time verification on the updated business activity record to determine whether it still meets the preset compliance requirements.

[0076] Cost allocation rules define how costs should be calculated and allocated in business activities that include shared expenses. The rule specifies the type of allocation, such as average allocation or weighted allocation. When the automated cost allocation engine 70 is triggered, it retrieves the cost allocation rule matching the current business activity type from the atomic strategy engine 60 and performs allocation calculations on the external digital traces of shared expense types associated with the business activity record according to that rule.

[0077] In a multi-user, multi-scenario expense reimbursement processing system, the business activity record is the core data object carrying all relevant information. It is generated by the business activity record creation module 10 and read and updated by other modules within the system. Physically, this data object can be represented as a data table record in a relational database, a document in a document-oriented database, or an instantiated object in the application's memory.

[0078] In one specific implementation, the structured dataset of business activity records includes the following fields:

[0079] Record ID (recordId): A globally unique identifier;

[0080] Activity Type (activityType): Used for query strategies;

[0081] Event organizer ID (ownerId): Creator identifier;

[0082] Activity description: the foundation for semantic relevance calculation;

[0083] Activity Period and Preset Location: These serve as the benchmarks for calculating temporal consistency and spatial clustering, respectively.

[0084] The sets of invited participants and confirmed participants are maintained by the multi-person collaboration management module 20.

[0085] The evidence checklist template and the mandatory evidence set are generated by the business activity record creation module 10 according to the strategy.

[0086] Associated Traces: Stores traces associated with the multi-source trace association module 30;

[0087] Record Completeness Score (sasScore): Stores the evaluation results, which may include scores for each sub-item;

[0088] Current Status: Indicates the process stage and is used by the automated decision-making module 50.

[0089] The workflow of the Business Activity Record Creation Module 10 is triggered when a user initiates a new business activity record request and ends with the generation of a business activity record in an initialized state.

[0090] The workflow of the Business Activity Record Creation Module 10 is triggered after the user specifies a business activity type (e.g., "Customer Banquet"). This module takes the user-selected business activity type as a parameter and sends a query request to the interface of the Atomic Strategy Engine 60.

[0091] Upon receiving a request, the atomic policy engine 60 retrieves and returns a set of all policy data units that match the type of business activity from its policy library.

[0092] The business activity record creation module 10 processes the received set of strategy data units. It first iterates through the set, extracting the "evidence requirement" array from each strategy data unit and integrating all extracted evidence requirement elements. Then, the module performs a deduplication operation based on the unique "evidence code" in each evidence requirement, generating a final set of evidence requirements that is unique and applicable to the current business activity type.

[0093] Based on this final collection, module 10 generates two key data sets:

[0094] Structured Evidence List Template: Extracts the descriptive information required for each piece of evidence, formats it, and uses it for display on the user interface.

[0095] Required evidence set: Extract the codes of all evidence requirements marked as "whether it is required" to form a set of codes for machine verification.

[0096] Finally, the business activity record creation module 10 instantiates a new business activity record data object, assigns it a unique record ID, and writes the user's ID, business activity type, and the evidence list template and required evidence item set generated in the previous step into the corresponding fields of the object. The newly created record is initially set to "draft" and persistently stored in the database, awaiting subsequent user input.

[0097] The workflow of the multi-person collaborative management module 20 is initiated by the event organizer to manage the invitation and confirmation of participants in business activities.

[0098] First, after the event organizer designates the participants, the multi-person collaborative management module 20 defines the "set of invited participants" in the business activity record and sends an electronic notification containing a unique confirmation link to each invitee.

[0099] When an invited participant confirms their participation via this link, the system generates a structured participation confirmation response. This response includes the participant's ID, a server timestamp, and a key digital signature.

[0100] Preferably, the digital signature is generated by concatenating the record ID of the business activity record, the current participant ID, and the timestamp, calculating a hash digest of the concatenated data using a preset hash function (such as SHA-256), and finally performing asymmetric encryption on the hash digest using a private key bound to the participant's identity.

[0101] Upon receiving the response data, the multi-user collaborative management module 20 executes a verification process. It obtains the public key corresponding to the participant ID and uses this key to decrypt the digital signature to reconstruct the original hash digest. Simultaneously, the module independently calculates a new hash digest for the record ID, participant ID, and timestamp in the response data using the same hash function. The authenticity and integrity of the response data are verified by comparing whether these two hash digests match.

[0102] If the verification is successful, module 20 will update the participant's identity information to the "Confirmed Participant Set" field of the business activity record, thereby providing an undeniable and accurate participant data foundation for subsequent completeness assessment and cost allocation.

[0103] In a preferred embodiment of the present invention, the multi-person collaborative management module 20 further includes an exception handling mechanism. If digital signature verification fails, the module will record a failed confirmation attempt and mark the response as invalid, while simultaneously sending a security alert to the system administrator and / or the activity manager of the business activity. The participant's status will not be updated in the 'confirmed participant set', thereby ensuring the authenticity and non-repudiation of the identities of all members in the set.

[0104] The workflow of the multi-source trace association module 30 and its internally configured intelligent context association device is as follows. The function of this module is to accurately associate the digital traces (such as electronic invoices, payment vouchers, itinerary slips, etc.) provided by the user from external systems with the existing business activity records in the system.

[0105] The multi-source trace association module 30 acquires external digital traces through one or more data receiving interfaces. These interfaces allow users to manually upload trace files or establish connections with third-party services (such as ride-hailing platforms and e-commerce websites) via application programming interfaces (APIs) to automatically acquire structured trace data. Each received trace data item contains at least text content, a time attribute (timestamp), and a location attribute (geographic coordinates).

[0106] Once an external digital trace is received, the multi-source trace association module 30 invokes its internal intelligent context association tool. This tool calculates the matching degree between the trace and one or more candidate business activity records. Its core mechanism involves parallel calculation of temporal similarity, spatial similarity, and textual similarity, followed by a weighted sum of these three similarities to arrive at a comprehensive matching degree.

[0107] For time similarity The calculation involves the correlator first extracting the timestamp from the external digital trace. And read the start timestamp of the activity cycle from the candidate business activity record. and end timestamp The calculation of time similarity is performed by a pre-defined function. When hour, .when When located outside the activity cycle, the similarity is reduced based on the duration of the deviation, calculated as follows:

[0108] ,in, This represents the time difference between the trace timestamp and the nearest activity cycle boundary; It is a preset time decay constant greater than zero, used to control the rate at which similarity decreases as the deviation time increases.

[0109] For spatial similarity The calculation, the correlator extracts the geographic coordinates from external digital traces. It also retrieves the geographic coordinates of the preset locations from the candidate business activity records. The correlator first uses the Haversine formula to calculate the great circle distance between two geographic coordinate points. Spatial similarity is calculated using an inverse proportional function, as shown in the following formula:

[0110] ,in, This is a preset reference distance parameter (e.g., 500 meters) used to normalize the distance. When two points coincide, =0, =1; with distance The increase, The value approaches 0.

[0111] For text similarity The calculation involves the correlator extracting text content (such as merchant names and product details) and business activity descriptions from external digital traces and business activity records. The calculation process includes the following steps:

[0112] First, the two text segments are preprocessed, including word segmentation, stop word removal, and stemming. Then, a pre-trained deep learning language model (such as BERT) is used to convert the preprocessed text segments into high-dimensional semantic vectors. and Finally, the cosine similarity between the two vectors is calculated to obtain the text similarity score: The result of this formula is in the range of [-1,1] or [0,1]. The closer the value is to 1, the more similar the semantics of the two texts are.

[0113] After calculating the similarity scores for the three dimensions mentioned above, the intelligent context association algorithm uses a weighted summation formula to calculate the final overall matching degree. : ,in, , , These are pre-configured weighting coefficients that sum to 1. These coefficients can be adjusted based on different types of business activities. For example, for travel-related activities, the weights for time and space are... and It can be set to a higher level. The calculated matching degree... As a quantitative basis for association, the system can recommend this external digital trace to the business activity record with the highest matching degree, so that the user can finally confirm the attribution.

[0114] The record completeness assessment module 40 calculates the first dimension of the assessment: evidence coverage. The purpose of this calculation is to quantitatively assess whether the evidence materials provided in the business activity record meet preset requirements. Before the calculation begins, the record completeness assessment module 40 reads two core sets of data from the specified business activity record data object. The first set of data is the "mandatoryEvidenceSet" stored in the record. This set is obtained and stored by the business activity record creation module 10 from the atomic strategy engine 60 during the record creation phase, and contains unique codes for all evidence required to complete this business activity. The second set of data is a list of evidence that the user has submitted and associated with the record. This list can be extracted from the "associatedTraces" list or other submitted evidence data fields, forming a set containing codes for the provided evidence.

[0115] The calculation process is achieved through the following steps:

[0116] The first step is for the completeness assessment module 40 to obtain the total number of evidence codes contained in the "Required Evidence Item Set", denoted as . This figure represents the total number of pieces of evidence that must be submitted to meet compliance requirements.

[0117] The second step involves the module comparing the set of evidence codes provided by the user with the "set of required evidence items" and determining the intersection of the two sets, that is, identifying which of the evidence submitted by the user is required.

[0118] The third step involves the module calculating the number of elements in this intersection, that is, the number of unique pieces of evidence provided by the user that are required evidence, denoted as . .

[0119] Fourth, the module calculates the final evidence coverage score by dividing the number of required evidence items provided by the total number of required evidence items. The calculation formula is as follows: ,in, The score represents the coverage of evidence. This is the count of the required evidence items that have been provided; It is the total count of the required evidence items.

[0120] The score obtained from this calculation This is a numerical value between 0 and 1. A score of 1 indicates that all required evidence has been provided; a score of 0 indicates that none of the required evidence has been provided. This score is then used as one of the five input parameters for calculating the final transcript completeness score.

[0121] The Transcript Completeness Assessment Module 40 calculates the second dimension of the assessment: Participant Confirmation Rate. The purpose of this calculation is to quantify the proportion of participants who have received valid confirmation out of all invited participants in a multi-person business activity.

[0122] Before the calculation begins, the record completeness assessment module 40 reads two key set fields from the specified business activity record data object. The first is the "invited participants" set, which is defined by the multi-person collaboration management module 20 based on the input of the activity organizer. The second is the "confirmed participants" set, which is updated by the multi-person collaboration management module 20 after receiving and verifying the participants' participation confirmation response data.

[0123] The calculation process is achieved through the following steps:

[0124] The first step is for the completeness assessment module 40 to obtain the total number of members in the "invited participant set", denoted as . This value represents the total number of participants invited to this business activity; the second step is for the module to obtain the total number of members in the "confirmed participant set", denoted as... This value represents the total number of participants who have received a valid confirmation response through the system process; in the third step, the module calculates the final participant confirmation score by dividing the number of confirmed participants by the total number of invited participants. The calculation formula is as follows:

[0125] ,in, The score represents the participants' level of confirmation. It is the member count of the "confirmed participant set"; It is the member count of the "invited participants' group".

[0126] When there are no invited participants (i.e.) =0), this business activity does not involve multi-person collaboration, this score is 0. The maximum score can be set to 1 by default. Otherwise, the calculated score is a value between 0 and 1. A score of 1 indicates that all invited participants have confirmed their participation. This score is then used as one of the five input parameters for calculating the final transcript completeness score.

[0127] The specific calculation process of the record completeness assessment module 40 in determining semantic relevance is as follows. The purpose of this dimension is to quantitatively assess the degree of semantic consistency between the text content of all external digital traces associated with the business activity record and the business activity description set by the record.

[0128] The first step of the calculation process is to construct two text files for comparison: a "Business Activity Description Document" and an "Aggregated Trace Document". The "Business Activity Description Document" is directly taken from the text content of the "Business Activity Description" field in the business activity record data object.

[0129] The construction of the "Aggregated Trace Document" is achieved through the following steps: First, traverse each trace object in the "Associated External Trace List" field in the business activity record; then, extract preset key text fields from each trace object, such as merchant name, product or service details, category, remarks, etc.; finally, concatenate all the extracted text contents to form a single document that aggregates all the evidence text information.

[0130] The second step of the calculation process is to perform text preprocessing on the above two documents. The preprocessing process includes:

[0131] First, tokenize the document content, splitting it into independent words or morphemes; second, according to a preset stopword dictionary, remove words lacking substantial semantic meaning in the document (such as "de", "yi ge"); third, perform stemming or lemmatization to unify different morphological words into their basic forms. This step aims to eliminate redundant information in the text and reduce the computational complexity of subsequent processing.

[0132] The third step of the calculation process is to convert the two preprocessed text documents into numerical semantic vectors. In this embodiment, a pre-trained deep learning language model, such as the BERT (Bidirectional Encoder Representations from Transformers) model, is used to perform this conversion. After the "Business Activity Description Document" is input into the model, a high-dimensional floating-point vector is generated, denoted as . Similarly, after the "Aggregated Trace Document" is input into the model, another high-dimensional vector of the same dimension is generated, denoted as . The positions and directions of these two vectors in the vector space represent the overall semantics of their corresponding original texts.

[0133] The last step of the calculation process is to obtain the final semantic correlation score by calculating the cosine similarity between the above two semantic vectors . Its calculation formula is as follows:

[0134] In this formula, the numerator is the dot product of two vectors, and the denominator is the product of the Euclidean norms of the two vectors. The result is a scalar value between 0 and 1. The closer the value is to 1, the more semantically relevant and logically consistent the overall textual content of the aggregated trace is with the established description of the business activity. This score is then used as an input to calculate the final transcript completeness score.

[0135] The completeness assessment module 40 determines the consistency over time. The specific calculation process is as follows. This process aims to quantitatively assess whether all associated external digital traces are consistent with the preset cycle of business activities in time.

[0136] At the start of the calculation, the record completeness assessment module 40 first reads two key data from the business activity record data object being assessed: one is the "activity period" field, from which the start timestamp of the business activity is obtained. and end timestamp The other is the "List of Associated External Tracks" field.

[0137] The module then iterates through the list of traces and extracts the time attribute, i.e., the timestamp, from each trace object. This step will generate a file containing... A collection of timestamps ,in This is the total number of associated traces with valid time attributes. If a business activity record has no associated traces with time attributes, the score for this evaluation can be set to 0 or a preset default value.

[0138] Next, the module will process each element in the timestamp set. Perform a score calculation. The score is determined by a pre-defined time-based scoring function. This function is executed to output a value between 0 and 1. The specific logic of the function is as follows: if the timestamp... It is within the effective period of the business activity, that is, it meets the conditions. If the timestamps are identical, the timestamps are considered completely identical, and their scores are set to the maximum value of 1. That is: ;

[0139] If timestamp If a trace is outside the effective period of a business activity, its time consistency will be deviated, and its score will be penalized. The severity of the penalty depends on the duration of the deviation from the period. First, the deviation duration is calculated. Its value is the time difference between the timestamp and the nearest activity cycle boundary: ;

[0140] Then, a penalized score is calculated based on the deviation duration. In this embodiment, the score is calculated using an exponential decay function:

[0141] ,in, It is a pre-configured, positive-zero time penalty coefficient. This coefficient controls the rate at which the score decreases as the deviation time increases.

[0142] For each timestamp in the set Calculate the corresponding score Then, the transcript integrity assessment module 40 calculates the arithmetic mean of all these scores to obtain the final time consistency score. The calculation formula is as follows: The final score It reflects the overall concentration and consistency of all related traces over time and serves as an input for calculating the comprehensive record completeness score.

[0143] The record integrity assessment module 40 determines spatial clustering. The specific calculation process is as follows. The purpose of this process is to quantitatively assess the spatial clustering of these traces by calculating the average deviation of the geographical locations of all associated traces from the preset locations of business activities.

[0144] At the start of the calculation, the record completeness assessment module 40 first reads two key data from the business activity record data object being assessed: one is the "preset location" field, from which the baseline geographic coordinates of the business activity are obtained. The other is the "List of Associated External Tracks" field.

[0145] The module then iterates through the list of traces and extracts the location attribute, i.e., geographic coordinates, from each trace object. This step will generate a file containing... A set of geographic coordinates ,in This is the total number of associated traces with valid location attributes. If a business activity record has no associated traces with location attributes, the score for this evaluation can be set to 0.

[0146] Next, the module calculates an average geographical distance. The calculation consists of two sub-steps. The first step is to calculate the coordinates of each location in the coordinate set. The module uses the Haversine formula to calculate its coordinates relative to the baseline geographic coordinates. Great circle distance between The second step is for the module to calculate all these independent distances. The arithmetic mean of the distances is used to obtain the average geographical distance. Its calculation formula is: This average geographical distance It quantitatively represents the degree of deviation of the overall geographical location of all associated traces from the pre-set location of business activities.

[0147] In obtaining average geographical distance Then, the module converts it into a final spatial clustering score using a preset scoring function. The function is designed with an inverse relationship, meaning the smaller the average distance, the higher the score. In this embodiment, the scoring function is as follows:

[0148] ,in, This is a preset, positive-zero distance threshold parameter, such as 500 meters or 1000 meters. This parameter is used to normalize the average distance and control the sensitivity of the score. When all tracks occur at the preset locations, =0, score The maximum value is 1. This is achieved with the average geographical distance. The increase in score It starts from 1 and smoothly decreases, approaching 0.

[0149] The final score It reflects the overall consistency and rationality of all related traces in the spatial dimension, and serves as an input item for calculating the comprehensive record completeness score.

[0150] The transcript completeness assessment module 40 calculates the scores for all individual dimensions and then generates the final comprehensive transcript completeness score as follows. This process is the final step in the assessment workflow, and its results provide a quantitative input basis for subsequent automated decision-making.

[0151] The evidence coverage score was calculated separately. Participant confirmation score Semantic relevance score Time consistency score and spatial clustering score Next, the transcript completeness assessment module 40 performs a weighted summation calculation to combine these five independent scores into a single, comprehensive score, namely the transcript completeness score, denoted as . .

[0152] Completeness score of the transcript The calculation formula is as follows:

[0153] ,in, , , , , These represent the numerical values ​​of evidence coverage, participant confirmation, semantic relevance, temporal consistency, and spatial clustering calculated in the aforementioned steps, respectively. , , , , It is a set of pre-configured weight coefficients, corresponding to the five evaluation dimensions mentioned above.

[0154] The weighting coefficients are dimensionless non-negative numbers, and their sum is set to 1, thus satisfying the constraint condition: These weighting coefficients are stored in a configuration area of ​​the system and can be configured differently according to different business activity types. For example, for a business activity of the "travel expense reimbursement" type, which has strong constraints on time and space, weighting coefficients can be configured accordingly. and Assign a higher value. For a business activity like "online contract approval," the requirements for participant confirmation and evidence completeness are even higher, so a higher value could be assigned. and Assign a higher value.

[0155] After the calculation is completed, the record completeness assessment module 40 will finally obtain the result. Numerical values, and scores for each sub-item. , , , , This data is then written to the corresponding fields of the evaluated business activity record data object for persistent storage. Subsequently, this comprehensive score... , , , , It will be passed to the automated decision-making module 50 as direct input for its decision-making process.

[0156] Please refer to the attached document. Figure 3 , Figure 3 This is an automated decision-making logic branch diagram of an embodiment of the present invention. In the system processing flow, the automated decision-making module 50 receives the output of the record integrity evaluation module 40 and starts the subsequent processing path.

[0157] The automated decision-making module 50 is driven by a preset rule set. This rule set is stored digitally in a configuration library within the system. The module's workflow begins upon receiving the comprehensive record completeness score calculated by the record completeness assessment module 40. It was subsequently triggered.

[0158] The module received The next step is to read a set of numerical decision thresholds from the configuration library. In one specific embodiment, this set of thresholds includes a threshold... And one person's review threshold ,in The value is set to be greater than The values ​​are [values]. These two thresholds can be configured differently depending on the type of business activity.

[0159] The second step in the module's execution is to receive... The values ​​are compared with these two thresholds, and a preset output operation is selected based on the comparison result. The specific decision logic branches are as follows:

[0160] Condition 1: If This condition indicates that the data integrity of the business activity record has reached a level that allows for direct approval. In this case, the module performs two operations: First, it updates the "Current Status" field in the business activity record data object, setting its value to "Approved"; second, it sends an allocation processing instruction containing the record ID to the automated cost allocation engine 70 to initiate the subsequent cost allocation process.

[0161] Condition 2: If This condition indicates that the data in the business activity record is uncertain and requires manual intervention for judgment. In this case, the module performs the following operation: updates the "Current Status" field in the business activity record data object, sets its value to "Pending Manual Review", and pushes a review task notification to the account of an operator with a preset role (e.g., department head or finance personnel) through the system's notification service.

[0162] Condition 3: If This condition indicates that the data completeness of the business activity record does not meet the minimum requirements. In this case, the module performs the following operation: updates the "Current Status" field in the business activity record data object, setting its value to "Rejected" or "Pending Supplementation". Simultaneously, the module extracts the scores for each sub-item from the record, using the evaluation dimensions with lower scores (e.g., "Insufficient Evidence Coverage" or "Low Temporal Consistency") as the basis for rejection or request for supplementation, and generates a notification message to be sent to the activity manager of that business activity.

[0163] By executing the above process, the automated decision module 50 converts a quantitative evaluation result into a clear state change or processing instruction that drives subsequent processes in the system.

[0164] The automated cost allocation engine 70 is activated after receiving the allocation processing instruction from the automated decision module 50. Its core function is to accurately calculate and allocate shared costs in business activities according to preset rules.

[0165] The execution flow of the automated cost allocation engine 70 begins with receiving an allocation processing instruction containing a specific business activity record ID. Upon receiving this instruction, the engine performs the first step, which involves retrieving and loading the complete business activity record data object from the system database using the record ID.

[0166] In the second step, the engine extracts the business activity type code from the loaded business activity records and uses this code as a parameter to send a policy query request to the atomic policy engine 60. The atomic policy engine 60 retrieves and returns the corresponding policy data unit based on this type code. From the returned policy data unit, the engine specifically extracts the "cost allocation rule" section. This rule section defines how costs should be allocated; for example, the rule may specify allocation modes such as "average allocation" or "weighted allocation."

[0167] The third step performed by the engine is to identify the shared costs that need to be allocated. It iterates through the "List of Associated External Traces" field of the business activity log and examines each trace object in the list. The engine filters out all traces of shared costs using a preset cost type field within the trace object (e.g., labeled "Shared Costs" or "Personal Costs"). Subsequently, the engine sums the amount fields of all trace objects identified as shared costs to obtain a total cost amount that needs to be allocated.

[0168] The fourth step executed by the engine is to determine the entities participating in the cost-sharing. The engine reads the "Confirmed Participant Set" field from the business activity record; each member in this set is a cost-sharing entity. The engine calculates the total number of members in this set, denoted as . .

[0169] The final step the engine performs is to calculate the cost allocation and generate the results. The engine executes the calculation based on the cost allocation rules obtained in the second step. If the rule is specified as "average allocation," the engine divides the total cost calculated in the third step by the total number of participants obtained in the fourth step. This will determine the amount of the cost to be borne by each participant. If the rule is specified as "weighted apportionment", the rule will further provide a weight coefficient for each participant or participant type, and the engine will allocate the total cost proportionally based on these weight coefficients.

[0170] After the calculation is complete, the engine generates a unique allocation result data record for each member in the "confirmed participant set". Each record contains at least the participant ID, the amount of the allocated cost, and a reference to the original shared cost trace. These allocation result data records are persistently stored in a cost detail database of the system, serving as the basis for subsequent financial accounting and reimbursement voucher generation.

[0171] The Atomized Policy Engine 60 provides real-time compliance verification during the user's submission of business activity records. This function complements the retrospective and comprehensive evaluation of the record completeness assessment module 40, aiming to move the binding effect of the policy forward to the source of data generation.

[0172] This real-time validation feature is not triggered only after the business activity record has been fully submitted, but rather activated when the user interacts with the system user interface, filling in or modifying data item by item. The front-end application captures this event when a user completes input in an input field and leaves that field, or when a user uploads a file.

[0173] Once the event is captured, the front-end application will trigger an asynchronous validation request to the atomic strategy engine 60. The data payload of this request will contain at least three key pieces of information: the record ID of the business activity being operated on, the unique identifier of the modified data field, and the current new value or new state of that field.

[0174] The real-time verification process of the Atomization Strategy Engine 60 is as follows:

[0175] Upon receiving a verification request, the engine reads the context of the business activity record (especially its "business activity type") based on the record ID in the request, and retrieves the atomic policy rules directly related to the field currently being verified from its policy library. These rules can be specifically expressed as follows:

[0176] Quantitative limit rules: For example, check whether the hospitality expenses exceed the preset upper limit (such as 500 yuan).

[0177] Formatting rules: For example, use regular expressions to validate the character composition and length of an invoice number.

[0178] Association logic rules: For example, the value of the "end time" field of a validation activity must be later than its associated "start time" field.

[0179] After the validation is performed, the engine generates a structured response containing a boolean "pass / fail" flag and specific prompt text (e.g., "Entertainment expenses exceeded the limit") and returns it asynchronously to the front-end application.

[0180] The front-end application provides users with immediate, context-sensitive feedback, such as highlighting the corresponding input field and displaying a prompt when validation fails. This real-time validation mechanism transforms compliance strategies into interactive guides, effectively preventing the generation of non-compliant data during the data entry stage, thereby significantly improving the quality of data submitted to the record integrity assessment module 40 and the automated decision-making module 50.

[0181] Example:

[0182] To more clearly reveal the collaborative working mechanism between the various technical modules of this invention, the following will take a typical "customer banquet" business activity as an example to fully demonstrate the entire process of a business activity record from creation to final processing.

[0183] Scenario: On the evening of April 25, 2025, an event coordinator (e.g., a sales manager) invites two clients from a partner company and a colleague from the company (i.e., another internal participant) to dinner to discuss project collaboration.

[0184] Step 1: Creating Business Activity Records

[0185] The event organizer initiates a new business activity record through the user interface of this invention system. They select "Client Banquet" as the "Business Activity Type." The business activity record creation module 10 is activated, immediately sending a query request to the atomic strategy engine 60. The atomic strategy engine 60 returns a strategy data unit for "Client Banquet," where the "Evidence Requirement" field specifies that two pieces of evidence must be provided: "Catering Invoice" and "Event Photos." Based on this, module 10 creates a business activity record in a "Draft" state and generates an evidence list for the organizer containing the aforementioned two requirements.

[0186] Step 2: Collaborative Confirmation by Participants

[0187] The event organizer added the aforementioned colleague as an internal participant in the event log. The multi-person collaborative management module 20 then sent a participation confirmation request to the colleague's account. The colleague clicked a link in the message, and the system displayed basic event information and requested confirmation. After the colleague clicked confirm, the system backend used the colleague's personal digital certificate's private key to sign data containing the event ID, the colleague's ID, and the current timestamp, generating a participation confirmation response. Upon receiving this response, module 20 successfully verified the validity of the signature using the colleague's public key and then updated the colleague's identity from the "invited participant set" to the "confirmed participant set."

[0188] Step 3: Intelligent Association of External Tracks

[0189] After dinner, the event organizer received an electronic invoice from the restaurant and uploaded it to the system via mobile phone. The multi-source trace association module 30 received the invoice file. Its built-in intelligent context association tool then began to function:

[0190] It extracts the timestamp "2025-04-25 20:15" from the invoice. This time falls within the activity period set by the administrator, "18:00 to 21:00," indicating a high degree of time similarity. The value is calculated to be 1.0.

[0191] It extracts the merchant's address from the invoice and compares it with the geographical coordinates of the banquet location preset by the business owner, calculating that the distance between the two points is only 150 meters. Based on the formula, spatial similarity... Calculate it to a high value, such as 0.95.

[0192] It uses the BERT model to convert text such as "XX Catering Services" in the invoice and "hosting a banquet for XX company clients to discuss cooperation" in the transcript into semantic vectors, and calculates the cosine similarity. It is 0.88.

[0193] Finally, the multi-source trace association module 30 calculates the comprehensive matching degree. The invoice is highly recommended and automatically recommended to the "Customer Banquet" record. The person in charge clicks to confirm on the interface to complete the association.

[0194] Step 4: Real-time verification of the atomization strategy

[0195] The event organizer entered an invoice amount of 1200 yuan in the record. Since the banquet involved four people (the event organizer, another internal participant, and two clients), the average cost per person was 300 yuan. At this point, the front-end application immediately initiated a verification request to the atomic strategy engine 60. The atomic strategy engine 60 retrieved an atomic rule applicable to "client banquets": "The average cost per person for entertainment must not exceed 250 yuan." Since 300 yuan exceeded this limit, the atomic strategy engine 60 immediately returned a verification failure response containing specific warning information. The system interface displayed a red warning next to the amount input box in real time: "Average cost per person exceeds the limit; the strategy limit is 250 yuan," and requested the organizer to provide a reason for exceeding the limit, such as... Figure 4 As shown.

[0196] Step 5: Comprehensive assessment of record completeness and automated decision-making

[0197] The event organizer uploaded photos of the event and provided reasons for exceeding the limits before clicking "Submit for Review." The "Completeness Assessment Module 40" was then triggered, and scores across five dimensions were calculated. Assuming the calculation result is: Evidence Coverage =1.0 (all required evidence is complete), participant confirmation rate =1.0 (confirmed by all internal participants), semantic relevance =0.92, time consistency =0.99, spatial clustering =0.95. Based on the preset weights, the final calculated comprehensive transcript completeness score, sasScore, is 0.96.

[0198] The score is then passed to the automated decision-making module 50. Module 50 reads the decision threshold for "customer banquet". =0.90. Since 0.96 ≥ 0.90, this record meets the conditions for automatic approval. The automated decision module 50 automatically updates the status of this record to "approved".

[0199] Step 6: Automated Cost Allocation

[0200] After the record is approved, the automated decision-making module 50 immediately issues an instruction to the automated cost allocation engine 70. The automated cost allocation engine 70 reads the record, identifies the total shared cost as 1200 yuan, and confirms the internal allocation entities as the event organizer and another internal participant. Based on the cost allocation rule in the strategy library for "customer banquets"—"allocated equally among internal participants"—the automated cost allocation engine 70 calculates that each person should share 600 yuan. Subsequently, it generates two records in the expense details library, respectively recording 600 yuan into the cost centers of the two internal participants, as follows: Figure 5 As shown.

[0201] Thus, a complex business activity has been fully, compliantly, and efficiently recorded and processed.

Claims

1. A multi-user, multi-scenario expense reimbursement form processing system, characterized in that, include: The Business Activity Record Creation Module is used to obtain compliance policies from the Atomic Policy Engine based on the type of business activity selected by the user, and generate a business activity record containing a structured evidence list template and a corresponding set of required evidence items. The multi-person collaboration management module is used to define the set of invited participants in the business activity record, record the participants' participation confirmation response data, and generate a set of confirmed participants. The multi-source trace association module is configured with an intelligent context association device to associate external digital traces along with the time and location attributes contained in the traces to the business activity record, wherein the external digital traces include electronic invoices, merchant names and product details; The intelligent context association is specifically used for: Extract the text content of the external digital traces and the business activity description of the business activity record, and use a deep learning language model to convert them into semantic vectors respectively. and And through the cosine similarity formula Calculate text similarity ; Calculate the time similarity between the external digital trace and the business activity record. and spatial similarity ; Using the weighted summation formula Calculate the overall matching degree ,in , , These are pre-configured weighting coefficients that sum to 1; The system recommends the external digital traces to the business activity records with the highest matching degree, so that users can finally confirm the attribution; The transcript completeness assessment module is used for: Evidence coverage is determined by comparing the evidence already provided in the business activity record with the set of required evidence items; The participant confirmation rate is determined by calculating the ratio of the confirmed participant set to the invited participant set. By analyzing the logical relationships between the associated external digital traces and the temporal and locational attributes of the external digital traces, the semantic relevance, temporal consistency, and spatial clustering are determined. Based on the evidence coverage, participant confirmation, semantic relevance, temporal consistency, and spatial clustering, a transcript completeness score is calculated. An automated decision-making module is used to receive the record completeness score and select the corresponding processing path based on the numerical range of the record completeness score.

2. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The function of the business activity record creation module to obtain compliance policies from the atomic policy engine and generate the evidence list template is implemented through the following steps: Based on the business activity type, query the strategy library of the atomic strategy engine and filter all compliance strategies associated with the business activity type; Extract the evidence requirements corresponding to the compliance strategy, integrate and deduplicate the evidence requirements, and generate the evidence list template.

3. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The participation confirmation response data includes a timestamp and digital signature used to verify authenticity.

4. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The intelligent context association device determines the association through the following steps: Calculate the proximity between the timestamp of the external digital trace and the activity cycle of the business activity record; Calculate the proximity between the geographical location of the external digital trace and the preset location of the business activity record; Calculate the semantic similarity between the text content of the external digital trace and the description of the business activity record; The matching degree is obtained by weighting the above three types of similarity and is used to recommend the optimal affiliation.

5. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, It also includes an automated cost allocation engine, which performs automated allocation calculations on external digital traces of shared cost types associated with the business activity record based on members in the confirmed participant set.

6. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The atomic strategy engine is used to trigger real-time compliance verification when the data in the business activity record is updated due to the association of records involved in the confirmation response or external digital traces.

7. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The record integrity assessment module determines temporal consistency through the following steps: Iterate through the time attributes of all external digital traces associated with the aforementioned business activity record; Determine whether each time attribute is within the activity cycle of the business activity record. If it is outside the cycle, calculate a penalty score based on the deviation duration according to preset rules. The time consistency is obtained by taking the arithmetic mean of all scores.

8. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The record integrity assessment module determines spatial clustering through the following steps: Obtain the location attributes of all external digital traces associated with the business activity record, and calculate the average geographic distance of the location attributes; Set the maximum acceptable distance threshold according to the type of business activity; The spatial clustering is calculated based on the ratio of the average geographical distance to the threshold.

9. The multi-user, multi-scenario expense reimbursement form processing system according to claim 1, characterized in that, The automated decision-making module selects the processing path through the following steps: When the completeness score of the recorded event is not lower than the preset high score threshold, a fully automated approval process is triggered. When the completeness score of the recorded event falls within the preset middle range, a simplified manual review process is initiated. When the completeness score of the recorded event is lower than the preset low score threshold, the process enters the full manual review process.

10. A multi-user, multi-scenario expense reimbursement form processing system according to claim 5, characterized in that, The allocation strategy types required by the automated cost allocation engine when performing allocation calculations are provided by the strategy library of the atomic strategy engine.

Images