An information processing method and device, electronic equipment and storage medium

Abstract

This application discloses an information processing method, apparatus, electronic device, and storage medium. The method includes: acquiring virtual resource transfer voucher information of a target object; performing a preliminary matching between the virtual resource transfer voucher information and a set of accounting records to be matched based on a preliminary matching strategy, wherein the preliminary matching strategy indicates matching based on a unique voucher identifier; if the preliminary matching fails, determining a fuzzy matching result based on the semantic similarity between the virtual resource transfer voucher information and the accounting records to be matched; if the fuzzy matching result indicates a matching failure, performing association reasoning matching based on the virtual resource transfer voucher information and the target object's knowledge graph using preset reasoning rules to obtain an association reasoning matching result; if the association reasoning matching result indicates a successful match, obtaining a reconciliation result based on the association reasoning matching result. This application improves the matching success rate for non-standard transfer vouchers and increases the reconciliation success rate in this scenario.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an information processing method, apparatus, electronic device and storage medium. Background Technology

[0002] In the process of financial information management, bank reconciliation and accounts receivable / payable reconciliation are high-frequency daily operations for enterprises. Related technologies primarily rely on standardized transfer vouchers with complete information (such as standard electronic documents with complete order and invoice numbers) for comparison and matching. This results in a low success rate for bank reconciliation in scenarios where non-standard transfer vouchers exist, increasing the difficulty of enterprise financial information management. Summary of the Invention

[0003] To address the problems of the prior art, embodiments of this application provide an information processing method, apparatus, electronic device, and storage medium. The technical solution is as follows: On the one hand, an information processing method is provided, the method comprising: Obtain the virtual resource transfer certificate information of the target object; Based on a primary matching strategy, the virtual resource transfer voucher information is matched against the set of accounting records to be matched; the primary matching strategy indicates that matching is performed based on a unique voucher identifier. If the initial matching fails, a fuzzy matching result is determined based on the semantic similarity between the virtual resource transfer certificate information and the accounting records to be matched in the set of accounting records to be matched; If the fuzzy matching result indicates that the matching fails, based on the virtual resource transfer certificate information and the knowledge graph of the target object, a pre-defined inference rule is used to perform association inference matching to obtain an association inference matching result; the knowledge graph includes multiple entities and the relationships between entities, and the multiple entities include the resource transfer object of the target object and the accounting record to be matched. If the correlation reasoning matching result indicates a successful match, the reconciliation result is obtained based on the correlation reasoning matching result.

[0004] On the other hand, an information processing apparatus is provided, the apparatus comprising: The credential information acquisition module is used to acquire the virtual resource transfer credential information of the target object; The primary matching module is used to perform a primary matching between the virtual resource transfer voucher information and the set of accounting records to be matched, based on a primary matching strategy; the primary matching strategy indicates that the matching process is based on a unique voucher identifier. The fuzzy matching module is used to determine the fuzzy matching result based on the semantic similarity between the virtual resource transfer voucher information and the accounting records to be matched in the set of accounting records to be matched when the primary matching fails. The association reasoning and matching module is used to perform association reasoning and matching based on the virtual resource transfer certificate information and the knowledge graph of the target object, using preset reasoning rules, to obtain an association reasoning and matching result when the fuzzy matching result indicates that the matching fails. The knowledge graph includes multiple entities and the relationships between entities. The multiple entities include the resource transfer object of the target object and the accounting record to be matched. The reconciliation result determination module is used to obtain the reconciliation result based on the association inference matching result if the association inference matching result indicates a successful match.

[0005] In some implementations, the association reasoning and matching module includes: The pending matching accounting record search module is used to determine the first pending matching accounting record associated with the resource transfer object in the virtual resource transfer voucher information based on the knowledge graph of the target object; The inference matching module is used to perform inference matching based on the virtual resource transfer certificate information and the first accounting record to be matched, using preset inference rules to obtain inference matching results; the inference matching results include candidate matching hypotheses and corresponding confidence levels, and the preset inference rules indicate that, under preset inference conditions, candidate matching hypotheses and corresponding confidence levels are generated according to preset inference paths; The association reasoning result determination module is used to determine a successful match and generate the association reasoning matching result based on the candidate matching hypothesis when the confidence level corresponding to the candidate matching hypothesis exceeds a preset confidence threshold.

[0006] In some implementations, the fuzzy matching module includes: The candidate accounting record determination module is used to filter out candidate accounting records that meet the entity matching conditions from the set of accounting records to be matched based on the virtual resource transfer voucher information and the entity matching degree of the accounting records to be matched in at least one key entity dimension; the at least one key entity dimension includes the quantity of virtual resources, time, and resource transfer object; The semantic similarity determination module is used to call the semantic similarity model to determine the semantic similarity between the virtual resource transfer voucher information and the candidate matching accounting record based on the summary information in the virtual resource transfer voucher information and the business description in the candidate matching accounting record. The fuzzy matching result determination module is used to determine the fuzzy matching result based on the semantic similarity corresponding to the candidate accounting records to be matched and the entity matching degree in each of the key entity dimensions.

[0007] In some implementations, the fuzzy matching result determination module is specifically used to: for each candidate accounting record to be matched, perform a weighted summation of the corresponding semantic similarity and the entity matching degree in each of the key entity dimensions to obtain the comprehensive matching degree of the candidate accounting record to be matched; if the highest comprehensive matching degree does not meet the matching degree threshold condition, then determine that the fuzzy matching result indicates a matching failure; if the highest comprehensive matching degree meets the matching degree threshold condition, then determine that the matching is successful and obtain the reconciliation result based on the candidate accounting record to be matched corresponding to the highest comprehensive matching degree.

[0008] In some implementations, the primary matching module is specifically used for: if the virtual resource transfer certificate information contains a unique certificate identifier that meets a preset identifier encoding rule, then searching for a target matching accounting record in the set of matching accounting records that matches the unique certificate identifier; if the target matching accounting record is not found, determining that the primary matching has failed; if the target matching accounting record is found, determining the reconciliation result based on the quantity difference between the number of virtual resources in the virtual resource transfer certificate information and the number of virtual resources in the target matching accounting record; if the virtual resource transfer certificate information does not contain a unique certificate identifier that meets the preset identifier encoding rule, then determining that the primary matching has failed.

[0009] In some implementations, the credential information acquisition module includes: The preprocessing module is used to determine the corresponding virtual resource transfer certificate text based on the original virtual resource transfer certificate of the target object; The entity recognition module is used to call the named entity recognition model to perform entity recognition on the virtual resource transfer certificate text to obtain at least one key entity; and to perform entity standardization on each key entity based on the knowledge graph. The intent recognition module is used to call the business intent recognition model to perform intent recognition on the virtual resource transfer certificate text to obtain the target business intent; The structured credential object generation module is used to generate a structured credential object based on at least one key entity after entity standardization and the target business intent. The structured credential object serves as virtual resource transfer credential information for the target object.

[0010] In some embodiments, the apparatus further includes: The automatic reconciliation and cancellation module is used to automatically process accounting entries and cancellations based on the reconciliation results and the target business intent.

[0011] On the other hand, an electronic device is provided, including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information processing method of any of the above aspects.

[0012] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the information processing method as described in any of the above aspects.

[0013] On the other hand, a computer program product is provided, which includes a computer program that, when executed by a processor, implements the information processing method of any of the above aspects.

[0014] This application embodiment obtains the virtual resource transfer certificate information of the target object, and performs a primary matching strategy to match the virtual resource transfer certificate information with the set of accounting records to be matched. The primary matching strategy indicates that matching is performed based on a unique certificate identifier. If the primary matching fails, a fuzzy matching result is determined based on the semantic similarity between the virtual resource transfer certificate information and the accounting records to be matched in the set of accounting records to be matched. If the fuzzy matching result indicates that the matching has failed, an association reasoning matching is performed based on the virtual resource transfer certificate information and the knowledge graph of the target object, using preset reasoning rules to obtain an association reasoning matching result. If the association reasoning matching result indicates that the matching has succeeded, the reconciliation result is obtained based on the association reasoning matching result. Thus, by coupling multiple levels of matching, including precise matching, fuzzy matching, and association reasoning matching, the matching success rate for non-standard transfer certificates is improved, thereby increasing the reconciliation success rate in this scenario and reducing the difficulty of enterprise financial information management. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating an information processing method provided in an embodiment of this application; Figure 2 This is a flowchart illustrating another information processing method provided in an embodiment of this application; Figure 3This is a flowchart illustrating another information processing method provided in an embodiment of this application; Figure 4 This is a flowchart illustrating another information processing method provided in an embodiment of this application; Figure 5 This is a schematic diagram of a system architecture for implementing the information processing method of this application embodiment; Figure 6 This is a structural block diagram of an information processing device provided in an embodiment of this application; Figure 7 This is a hardware structure block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

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

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0019] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0020] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0021] The information processing method of this application embodiment can be applied to the information processing device provided in this application embodiment. The information processing device can be configured in a computer device, which may include a terminal or server of an enterprise. The computer device can provide financial information management services for the enterprise, which may include automatic reconciliation, automatic entry and cancellation.

[0022] The terminals include, but are not limited to, mobile phones, computers, and smart voice interaction devices. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0023] Please see Figure 1 The diagram illustrates a flowchart of an information processing method provided in an embodiment of this application. It should be noted that while this specification provides the operational steps of the method as described in the embodiments or flowchart, more or fewer operational steps may be included based on conventional or non-inventive methods. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or product execution, the method can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment) as shown in the embodiments or accompanying drawings. Specifically, as shown... Figure 1 As shown, the method may include:

[0024] S101, Obtain the virtual resource transfer certificate information of the target object.

[0025] The target object can be a physical business entity, such as a company. The computer equipment can obtain the company's virtual resource transfer certificate information. This virtual resource transfer certificate information can be structured data text obtained after data structuring based on the original virtual resource transfer certificate. The virtual resource transfer certificate information is used to record the virtual resource transfer transactions of the target object. For example, this virtual resource transfer certificate information can be transfer voucher information.

[0026] In some possible implementations, obtaining the virtual resource transfer credential information of the target object may include the following steps (1) to (5): (1) Based on the original virtual resource transfer certificate of the target object, determine the corresponding virtual resource transfer certificate text.

[0027] The original virtual resource transfer certificate can be presented in the form of images or text. The source of the original virtual resource transfer certificate is unrestricted; for example, it can be from online banking systems, third-party payment platforms, or scanned copies of paper transfer certificates.

[0028] The original Virtual Resource Transfer Certificate (VRLP) of the target object can be preprocessed to obtain the corresponding VRLP text. Specifically, for the original VRLP presented as an image, image cleaning (such as noise reduction and tilt correction) can be performed first, combined with OCR recognition to obtain the corresponding electronic text. Then, deduplication and special character filtering are performed on the electronic text to obtain clean text data, which is the corresponding VRLP text. In other words, for the original VRLP presented as text, deduplication and special character filtering can be directly performed to obtain the VRLP text.

[0029] (2) Call the named entity recognition model to perform entity recognition on the virtual resource transfer certificate text to obtain at least one key entity.

[0030] The named entity recognition model can be a neural network model based on a pre-trained Transformer architecture. Specifically, the named entity recognition model can include an encoding layer, a fully connected layer, and a conditional random field layer. The encoding layer encodes the input virtual resource transfer certificate text to obtain a sequence of text-encoded vectors. The fully connected layer maps each text-encoded vector in this sequence to a low-dimensional fractional vector to obtain a sequence of fractional vectors. The length of each fractional vector is equal to the size of a predefined label set, and the fractional vector indicates the probability that a word is predicted as one of the labels in the label set. The conditional random field layer decodes the sequence based on the fractional vectors using a B (beginning of named entity) I (inside named entity) O (non-named entity) E (end of named entity) S (single entity) labeling system to obtain the output entity recognition result. This entity recognition result includes at least one key entity in the virtual resource transfer certificate text. Specifically, the label set can be {resource transfer object (e.g., recipient's name), virtual resource quantity (e.g., transaction amount), time (e.g., transaction date), summary text, unique certificate identifier (e.g., order number)}.

[0031] In some examples, the encoding layer can be a bidirectional semantic encoding network based on a self-attention mechanism, such as the BERT network; in other examples, the encoding layer can also be an LSTM network (Long Short-Term Memory network).

[0032] Training the named entity recognition model can be achieved by supervising the pre-trained neural network model using a training dataset. The training dataset includes historical voucher data, with each voucher corresponding to a reference label sequence. This reference label sequence can be obtained by financial experts using the BIOES annotation system to label the entities in the historical voucher data. The labeled entity types include, but are not limited to, resource transfer objects (e.g., counterparty's name), virtual resource quantity (e.g., transaction amount), time (e.g., transaction date), summary text, and unique voucher identifiers (e.g., order number).

[0033] The pre-trained neural network model includes an encoding layer, a fully connected layer, and a conditional random field (CRF) layer. During training, training data is input into the encoding layer of the pre-trained neural network model. The encoding layer, fully connected layer, and CRF layer process the data sequentially, outputting the predicted label sequence corresponding to the training data. Then, based on the difference between the predicted label sequence and the corresponding reference label sequence, the model parameters of the pre-trained neural network model are adjusted using a preset loss function. Training continues iteratively based on the adjusted model parameters until the training termination condition is met, resulting in a trained named entity recognition model. The preset loss function can be the cross-entropy loss function. The training termination condition can include the number of iterations reaching an iteration threshold (which can be set as needed), or the calculated loss function value being less than the preset loss threshold (which can be set as needed). In specific implementations, the optimizer used during training can be the AdamW optimizer, which decouples weight decay from gradient updates, more effectively preventing overfitting and promoting model generalization. The learning rate of the AdamW optimizer can be 2e-5.

[0034] After training the named entity recognition model, the model is called to perform entity recognition on the virtual resource transfer certificate text, thereby obtaining at least one key entity, such as the resource transfer object (e.g., the recipient's name), the quantity of virtual resources (e.g., the transaction amount), the time (e.g., the transaction date), the summary text, and the unique certificate identifier (e.g., the order number).

[0035] (3) Standardize the key entities based on the knowledge graph of the target object; Considering that the key entities extracted from the virtual resource transfer certificate text using the named entity recognition model are usually diverse in form and non-standard in expression, with a large number of abbreviations, aliases, common names, old names and input errors, in order to improve the success rate of subsequent matching and thus improve the success rate of reconciliation, this application embodiment performs entity standardization on at least one extracted key entity based on the knowledge graph of the target object, thereby mapping each extracted key entity to a unique standardized entity in the knowledge graph.

[0036] Specifically, the target object's knowledge graph includes multiple entities (which can be understood as standardized entities) and the relationships between them. This knowledge graph can be constructed by connecting to the target object's core business systems, such as ERP / CRM systems, to extract business entities and their relationships. For example, from the CRM, the official registered names, unified social credit codes, and common abbreviations of all "customers / suppliers" can be extracted. This information is then manually entered (domain experts can review and complete the extracted business entities). For instance, the constructed knowledge graph includes entity nodes such as "customers / suppliers," "products," and "projects," along with their relationships, such as "Customer A" - purchased -> "Product B." Each standardized entity node in the knowledge graph can be associated with a dynamically maintained set of aliases for entity standardization. In this embodiment, entity standardization is achieved by combining fuzzy string matching with graph entity retrieval. Specifically, for each key entity identified by entity recognition, fuzzy matching is performed with each standardized entity node in the knowledge graph. For example, the text similarity between the key entity and each standardized entity is calculated, such as by calculating the edit distance. Then, the top preset number (which can be set as needed) of standardized entities with the highest text similarity are selected as candidate standardized entities corresponding to the key entity, resulting in a set of candidate standardized entities corresponding to the key entity. Next, for each candidate standardized entity in the set of candidate standardized entities corresponding to the key entity, the associated standardized entities are queried from the knowledge graph. The overlap between the associated standardized entities and the remaining key entities is then calculated, and the highest overlap score is taken as the association score of the candidate standardized entity. After obtaining the association score of each candidate standardized entity, the candidate standardized entity with the highest association score can be taken as the result of entity standardization corresponding to the key entity. By combining fuzzy string matching with graph entity retrieval, the key entities identified by entity recognition are standardized, which improves the accuracy of entity standardization. This allows each key entity after standardization to be used in subsequent information processing with its unique knowledge graph identifier, which helps to improve the success rate of reconciliation.

[0037] (4) Call the business intent recognition model to perform intent recognition on the virtual resource transfer certificate text to obtain the target business intent; Specifically, the business intent recognition model can identify the target business intent based on the summary text in the virtual resource transfer voucher text. Business intent recognition is modeled as a text classification task. In specific implementation, the business intent recognition model can include an encoding layer and a classification layer. The encoding layer can be a pre-trained BERT model, used to perform bidirectional semantic encoding based on a self-attention mechanism on text starting with the [CLS] marker to generate a context vector representation for each input marker, resulting in an encoded vector sequence. The context vector representation corresponding to the [CLS] marker at the beginning of the sequence aggregates the semantic information of the entire sequence, which can be understood as a global semantic representation of the input sentence. This global semantic representation is then input to the classification layer, which predicts the probability distribution across all preset business intent categories and determines the preset business intent category with the highest probability as the identified target business intent. The classification layer can include a fully connected layer and a Softmax function connected to it. The preset business intent categories can be set based on actual needs, such as "purchase payment," "service payment," "expense reimbursement," "internal transfer," and "refund income." For example, the preset business intent categories should cover the main resource transfer types of the target object.

[0038] The business intent recognition module can be trained using historical business data labeled with intent categories. These intent category labels can be obtained by domain experts annotating the historical business data. The loss function used during training can be the cross-entropy loss function, and the training objective is to minimize the difference between the predicted intent category and the intent category label.

[0039] (5) Based on at least one key entity after entity standardization and the target business intent, generate a structured credential object, which serves as the virtual resource transfer credential information of the target object.

[0040] Specifically, a corresponding business element template can be pre-set for each preset business intent. This business element template contains multiple business element slots corresponding to the corresponding preset business intent. For each business element slot, if there is matching entity information in at least one key entity after entity standardization, the corresponding business element is obtained from that entity information and filled into the business element slot. If there is no matching entity information in at least one key entity after entity standardization, the corresponding key business element can be extracted from the summary text of the virtual resource transfer certificate text by keywords and filled in under the guidance of the target business intent.

[0041] The above implementation method obtains structured virtual resource transfer voucher information by performing pre-processing on the original virtual resource transfer voucher, entity recognition based on the named entity recognition model, entity standardization based on the knowledge graph, and business intent classification. This enables intelligent parsing and standardization of non-standard virtual resource transfer vouchers (such as non-standard transfer vouchers), greatly improving the accuracy of processing non-standard vouchers and covering more than 90% of enterprise non-standard financial voucher scenarios.

[0042] S103, Based on the primary matching strategy, perform primary matching on the virtual resource transfer certificate information and the set of accounting records to be matched.

[0043] The primary matching strategy indicates that matching is performed based on a unique credential identifier, thus the primary matching strategy is a precise matching method based on a unique credential identifier.

[0044] In some implementations, such as Figure 2 As shown, step S103, when performing a preliminary match between the virtual resource transfer certificate information and the set of accounting records to be matched based on the preliminary matching strategy, may include: S201, if there is a unique voucher in the virtual resource transfer voucher information that meets the preset identification coding rules, then search for the target matching accounting record in the matching accounting record set that matches the unique voucher.

[0045] Specifically, in this embodiment of the application, the existence of a unique credential identifier that meets the preset identifier encoding rule is used as the triggering condition for triggering the initial matching. If the triggering condition is met, the following steps S203 to S205 are executed; otherwise, if the triggering condition is not met, the initial matching is not executed, which can be considered as the initial matching failure, and the process proceeds to step S207.

[0046] The preset identifier encoding rules are used to indicate the encoding rules for unique credential identifiers. For example, they may include the length of the unique credential identifier and the prefix check code, and can be set based on practical experience. In specific implementation, when the unique credential identifier (such as standardized order number or standardized invoice number) field of the virtual resource transfer credential information is not empty, and the content of this field conforms to the preset identifier encoding rules, it is considered that the preliminary matching trigger condition is met, and preliminary matching can be performed.

[0047] During the initial matching phase, the system searches for target accounting records within the target object's set of matching accounting records that match the unique voucher identifier. Specifically, the unique voucher identifier can be used as the key to query the target object's set of matching accounting records, which contains the target object's unreconciled accounting records. In practice, the unreconciled accounting records of the target object can be divided into multiple subsets based on the type of the unique voucher identifier. For example, standardized order numbers correspond to the "unreconciled order / contract pool" subset, and standardized invoice numbers correspond to the "unreconciled accounts receivable / payable invoice pool" subset. During the search, the corresponding subset is searched based on the type of the unique voucher identifier to improve matching efficiency. For instance, the standardized order number can be used as the key to query the "unreconciled order / contract pool" subset, or the standardized invoice number can be used as the key to query the "unreconciled accounts receivable / payable invoice pool" subset.

[0048] S203, if the target accounting record to be matched is not found, the initial matching is determined to have failed.

[0049] S205, if the target matching accounting record is found, the reconciliation result is determined based on the quantity difference between the number of virtual resources in the virtual resource transfer voucher information and the number of virtual resources in the target matching accounting record.

[0050] Specifically, if a target matching accounting record is found, the reconciliation result is further determined based on the quantity difference between the number of virtual resources in the virtual resource transfer voucher information and the number of virtual resources in the target matching accounting record.

[0051] In practice, if the number of virtual resources in the virtual resource transfer voucher information matches the number of virtual resources in the target matching accounting record, it is considered a perfect match. A reconciliation result can be generated indicating that the virtual resource transfer voucher information and the target matching accounting record are associated and that their virtual resource quantities are consistent. If the number of virtual resources in the virtual resource transfer voucher information is less than the number of virtual resources in the target matching accounting record, it is considered a partial match. In this case, a reconciliation result can be generated indicating that the virtual resource transfer voucher information and the target matching accounting record are associated, as well as the number of virtual resources already covered and the remaining number of virtual resources. Understandably, the above reconciliation result can be used for subsequent automatic posting and reconciliation processing.

[0052] For example, the transfer voucher information is: {Order No.: "PO-20231025-001", Amount: 10000, Recipient's Name: "A Technology Co., Ltd.", Type: "Payment for Goods"...}. Searching using "PO-20231025-001" will find a purchase order with "A Technology Co., Ltd." for 10,000 yuan. Since the amount in the transfer voucher information matches the amount due in the purchase order, there is a 100% exact match. The generated reconciliation result indicates that the transfer voucher information is associated with the purchase order and that the amounts are consistent. Therefore, in subsequent automatic posting and reconciliation processing based on this reconciliation result, an accounting voucher can be automatically generated as "Debit: Accounts Payable - A Company 10000 Credit: Bank Deposit 10000", and the purchase order will be marked as "Paid". Understandably, if the amount in the transfer voucher is less than the amount due in the purchase order, a partial write-off record can be created and the remaining amount of the purchase order updated during the subsequent automatic posting and write-off process based on the generated reconciliation results.

[0053] S207, if there is no unique credential identifier in the virtual resource transfer credential information that meets the preset identifier encoding rules, then the primary matching is determined to have failed.

[0054] In the above implementation, precise matching based on unique voucher identifiers is used as the first level of matching. The standardized unique identifiers (complete order number, invoice number) are precisely matched with the accounting records to be matched, with a matching accuracy of 100%.

[0055] S105, if the initial matching fails, a fuzzy matching result is determined based on the semantic similarity between the virtual resource transfer certificate information and the accounting records to be matched in the set of accounting records to be matched.

[0056] Specifically, if the initial matching fails, the virtual resource transfer certificate information proceeds to a second-level fuzzy matching process. During fuzzy matching, the semantic similarity between the summary text in the virtual resource transfer certificate information and the business description in the accounting record to be matched is calculated to determine the fuzzy matching result. The semantic similarity can be represented by the cosine similarity between the sentence vectors of the summary text and the sentence vectors of the business description.

[0057] In some implementations, step S105, which determines the fuzzy matching result based on the semantic similarity between the virtual resource transfer certificate information and the accounting records to be matched in the set of accounting records to be matched, may include the following steps (1) to (3): (1) Based on the virtual resource transfer certificate information and the entity matching degree of the accounting record to be matched in at least one key entity dimension, candidate accounting records that meet the entity matching conditions are selected from the set of accounting records to be matched.

[0058] The at least one key entity dimension includes the quantity of virtual resources, time, and the object to which resources are transferred. Specifically, the candidate accounting records to be matched can be accounting records that meet the corresponding entity matching conditions in all key entity dimensions, or they can be accounting records that meet the corresponding entity matching conditions in one or two key entity dimensions.

[0059] For key entity dimensions that are the quantity of virtual resources (such as monetary value), a tolerance parameter can be set, for example, denoted as... This represents the upper limit of the relative deviation of the allowed number of virtual resources. Its specific value can be set to, for example, 0.5%, 1%, or 5%, etc., depending on the business scenario. Based on this tolerance parameter and the number of virtual resources in the virtual resource transfer certificate information, the virtual resource quantity tolerance range can be determined as: [A] ], where A represents the quantity of virtual resources in the virtual resource transfer voucher information. Taking the transfer voucher information as an example, assume that... =5%, then the tolerance range for the amount is [voucher amount A] 0.95, voucher amount A [1.05], then the entity matching condition corresponding to this key entity dimension is that the accounting amount in the accounting record to be matched is within the amount tolerance range, and the corresponding amount matching score (that is, the entity matching degree of the quantity dimension) can be calculated by the following formula:

[0060] Where B represents the amount of the transaction in the transaction record to be matched; This indicates the score for matching the monetary amount.

[0061] Specifically, when calculating the amount matching score based on the above formula, it is first determined whether the amount B in the accounting record to be matched is within the amount tolerance range. If the amount is outside the tolerance range, it is directly determined that the amount does not match, and the amount matching score is calculated. =0. If the accounting amount B is within this tolerance range, then the absolute deviation between the accounting amount B and the voucher amount A is determined. Then calculate the absolute deviation. With maximum permissible deviation The ratio is then used to calculate the difference between 1 and this ratio to obtain the monetary matching score. Therefore, when the accounting amount B equals the voucher amount A, the absolute deviation is... If the score is 0, then the amount matching score is... =1, as the accounting amount B deviates from the voucher amount A and gradually approaches the tolerance boundary, the amount matching score... The amount matching score decreases linearly from 1 to 0 when the accounting amount B is exactly equal to the tolerance boundary value. =0.

[0062] For key entity dimensions such as time (e.g., date), a time window can be constructed for filtering by setting a preset duration, for example, denoted as T. This preset duration T represents the maximum allowed time offset, and its specific value can be, for example, 15 days, 30 days, etc., which can be set according to the business scenario. Thus, the time window is: ,in, This refers to the time / date in the virtual resource transfer voucher information. Taking the transfer voucher information as an example, let's assume... =15 days, then the time window is [voucher date] -15 days, voucher date [+15 days], then the entity matching condition corresponding to this key entity dimension is the date (such as due date / occurrence date) in the accounting record to be matched within this time window. The corresponding time matching score (i.e., the entity matching degree of the time dimension) can be calculated by the following formula:

[0063] in, This indicates the date in the accounting record to be matched. This represents the time-match score.

[0064] Specifically, when calculating the time matching score based on the above formula, the date in the accounting record to be matched is first determined. Date in the Virtual Resource Transfer Certificate Information absolute time difference Is it less than or equal to the preset duration? If it exceeds the preset duration If the date does not match, the time matching score will be determined directly. =0. If less than or equal to the preset duration. This indicates that within the time window, it is possible to base on Get the time matching score Therefore, when the date in the accounting record to be matched... With voucher date When they are the same, the absolute time difference The score is 0, indicating a time-matching degree score of 0. As the absolute time difference increases, the time matching score... The time matching score decreases linearly from 1 to 0 when the absolute time difference is exactly equal to T. =0.

[0065] For key entity dimensions where the resource transfer object is the recipient (e.g., the recipient's name), a string similarity algorithm (e.g., edit distance algorithm) can be used to calculate the similarity between the resource transfer object in the accounting record to be matched and the resource transfer object in the voucher. The entity matching condition for this key entity dimension is that the similarity between the resource transfer objects exceeds a preset similarity threshold (e.g., 90%), and the corresponding object similarity score is determined. This refers to the entity matching degree of the resource transfer object dimension.

[0066] Understandably, in practical applications, the matching records in the set of matching records can be filtered based on the need to consider more key entity dimensions.

[0067] (2) Call the semantic similarity model to determine the semantic similarity between the virtual resource transfer certificate information and the candidate matching accounting record based on the summary information in the virtual resource transfer certificate information and the business description in the candidate matching accounting record.

[0068] The semantic similarity model can employ a symmetric Siamese network structure to ensure fair and consistent encoding of the input text at both ends. Specifically, the semantic similarity model includes an encoding layer, a mean pooling layer, and a similarity calculation layer. The encoding layer comprises two BERT encoders with shared weights, used to encode the input credential summary information and business description respectively. The mean pooling layer performs mean pooling on the encoded vectors output by the two BERT encoders to obtain a first vector representation of the corresponding credential summary information and a second vector representation of the corresponding business description. These first and second vector representations are further input into the similarity calculation layer to calculate the cosine similarity between them as the semantic similarity.

[0069] The training objective of a semantic similarity model is to ensure that semantically similar sentences have similar vector distances, while semantically dissimilar sentences have large vector distances. Therefore, contrastive learning can be used for training. In practice, the training data includes positive and negative sample pairs. Positive sample pairs are "voucher summary-business description" pairs collected from historically verified financial data, such as a reimbursed invoice and its corresponding payment application summary. Negative sample pairs are unmatched "voucher summary-business description" pairs, obtained by pairing a voucher summary with randomly selected unrelated business descriptions. The loss function used for training can be a triplet loss. Each training sample includes (voucher summary, matching business description, unmatched business description). The training samples are input into the semantic similarity model to be trained. The loss value is determined using the following triplet loss calculation formula, and the model parameters are adjusted in the direction of minimizing this loss value to obtain the trained semantic similarity model.

[0070] Where L represents the loss value; f represents the model; A, P, and N represent the voucher summary, the matching business description, and the non-matching business description, respectively; and margin represents the preset hyperparameter, which can be set based on practical experience.

[0071] (3) Determine the fuzzy matching result based on the semantic similarity of the candidate accounting records to be matched and the entity matching degree in each of the key entity dimensions.

[0072] For example, determining the fuzzy matching result based on the semantic similarity corresponding to the candidate accounting records to be matched and the entity matching degree in each of the key entity dimensions may include: For each candidate accounting record to be matched, the corresponding semantic similarity and the entity matching degree in each of the key entity dimensions are weighted and summed to obtain the comprehensive matching degree of the candidate accounting record to be matched. If the highest overall matching degree does not meet the matching degree threshold condition, then the fuzzy matching result indicates a matching failure. If the highest overall matching degree meets the matching degree threshold condition, the matching is determined to be successful and the reconciliation result is obtained based on the candidate accounting records to be matched corresponding to the highest overall matching degree.

[0073] Specifically, the overall matching degree of candidate accounting records can be calculated using the following formula:

[0074] in, , , , This indicates the corresponding weighting coefficient, which can be set based on practical experience, and ; Represents semantic similarity score, Indicates the score of monetary matching. Indicates the time matching score. This represents the object similarity score.

[0075] After obtaining the overall matching degree of each candidate accounting record to be matched, the overall matching degrees can be sorted in descending order from high to low, so that the one ranked first is determined as the highest overall matching degree. The matching degree threshold condition can be that the highest overall matching degree is greater than or equal to a preset matching degree threshold (e.g., 0.8), and the difference between the highest overall matching degree and the second-ranked overall matching degree is not less than a preset minimum value (e.g., 0.1), to improve the accuracy of fuzzy matching. If the highest overall matching degree does not meet the matching degree threshold condition, the fuzzy matching result indicates a matching failure; conversely, if the highest overall matching degree meets the matching degree threshold condition, the fuzzy matching result indicates a successful matching. In this case, a reconciliation result can be obtained based on the candidate accounting record to be matched corresponding to the highest overall matching degree. This reconciliation result indicates that the virtual resource transfer voucher information is associated with the candidate accounting record to be matched, so that automatic posting and reconciliation processing can be performed subsequently based on this reconciliation result.

[0076] For example, the transfer voucher information is as follows: {Summary text: "March technical service fee", amount: 5000, date: 2023-10-10, recipient's name: "Shenzhen Yi Co., Ltd."...}.

[0077] One accounting record to be matched is: There is an invoice issued to "Shenzhen Yi Co., Ltd.", described as "Technical Service Fee for the Third Quarter of 2023", amount: RMB 5,000, invoice date: 2023-09-25.

[0078] Fuzzy matching process: Amount (5000 vs 5000), time (October 10th vs September 25th, within ±15 days), and account name (highly similar) all passed the screening. The semantic similarity model calculated the semantic similarity between "March technical service fee" and "2023 third quarter technical service fee" to be 0.92.

[0079] The overall matching degree is calculated to be over 0.8, so the fuzzy matching result is a successful match. At this time, the invoice can be automatically associated and reimbursed.

[0080] The above implementation method combines at least one key entity dimension and the semantic similarity of the semantic similarity model for fuzzy matching, which can achieve a matching success rate of 75% at the second level, thus improving the reconciliation success rate.

[0081] S107, if the fuzzy matching result indicates that the matching has failed, based on the virtual resource transfer certificate information and the knowledge graph of the target object, a pre-set inference rule is used to perform association inference matching to obtain an association inference matching result; the knowledge graph includes multiple entities and the relationships between entities, and the multiple entities include the resource transfer object of the target object and the accounting record to be matched.

[0082] Specifically, if the second-level fuzzy matching fails, a third-level association reasoning matching is performed based on the knowledge graph of the target object and using preset reasoning rules to improve the success rate of the matching. This third level can cover the remaining 10% of non-standard scenarios.

[0083] In some implementations, such as Figure 3 As shown, step S107 may include the following when implemented: S301, based on the knowledge graph of the target object, determine the first matching accounting record associated with the resource transfer object in the virtual resource transfer certificate information.

[0084] Understandably, the entity nodes in the target object's knowledge graph include the target object's resource transfer objects and the accounting records to be matched. The accounting record nodes to be matched can also correspond to attribute information, such as dynamically maintained current status (e.g., settled or unsettled). When determining the first accounting record to be matched, we can first search for the resource transfer object node in the target object's knowledge graph based on the resource transfer object in the virtual resource transfer voucher information. After finding the resource transfer object node, we can find the associated accounting record nodes to be matched based on the edges originating from it. Then, based on the attribute information of these associated accounting record nodes, we can determine the associated accounting record nodes whose current status is unsettled, thus obtaining the first associated accounting record to be matched.

[0085] For example, entity nodes in an enterprise knowledge graph can include customers / suppliers, contracts, orders, invoices, and payments. Edges (i.e., relationships between entities) can include signing, inclusion, issuance, payment, and verification. Thus, by using the "counterparty's name" in the voucher information, the corresponding customer / supplier node can be located in the enterprise knowledge graph. Starting from that customer / supplier node, we can explore its associated contract, order, and invoice nodes that are in an incomplete (unsettled) state, and then determine the first related accounting record to be matched.

[0086] S303, based on the virtual resource transfer certificate information and the first accounting record to be matched, inference matching is performed using preset inference rules to obtain inference matching results; the inference matching results include candidate matching hypotheses and corresponding confidence levels, and the preset inference rules indicate that candidate matching hypotheses and corresponding confidence levels are generated according to preset inference paths under preset inference conditions.

[0087] In specific implementation, the pre-set inference rules may include one or more of the following rules: Preset inference rule A (aggregate payment inference): { Preset inference condition: The amount of a single voucher is approximately equal to the total amount of multiple outstanding small invoices / orders for a certain customer. For example, the difference between the amount of a single voucher and this total is within the preset tolerance range (which can be set as needed).

[0088] Preset reasoning path: This payment is intended to settle multiple scattered debts in one go.

[0089] Preset action: Prompt "It is recommended to match the following X invoices for combined reconciliation" and list them, which can be confirmed by the user or executed automatically.

[0090] } Preset reasoning rule B (progress payment / partial payment reasoning): { Preset inference conditions: The voucher summary contains keywords such as "progress payment" and "Nth installment", and the customer has an unfinished contract with a total amount much greater than the voucher amount.

[0091] Pre-defined reasoning: This payment is a phase payment under the contract.

[0092] Preset action: Automatically create a record associated with the contract as a partial payment record and update the amount paid under the contract.

[0093] } Preset reasoning rule C (historical pattern reasoning): { Preset inference conditions: The customer has a history of regular payments (such as a fixed monthly service fee), and the amount, time and pattern of this voucher are consistent, but the summary is extremely vague (such as "transfer").

[0094] Preset reasoning path: Based on historical behavior patterns, infer that this is a regular periodic payment.

[0095] Preset action: Associated with the most recent unpaid periodic service invoice.

[0096] } In specific implementation, the candidate matching hypothesis corresponding to the aforementioned preset inference rule A (aggregated payment inference) can be: generating a "merged reconciliation" hypothesis, the content of which is: "It is recommended to match this payment to the following X invoices (list invoice number, date, and amount) for merged reconciliation." This candidate matching hypothesis points to a matching set. The candidate matching hypothesis corresponding to the aforementioned preset inference rule B (progress payment / partial payment inference) can be: generating a "contract progress payment" hypothesis, the content of which is: "It is inferred that this payment is the Nth progress payment for contract [contract number], and it is recommended to match and update the contract payment progress." The candidate matching hypothesis corresponding to the aforementioned preset inference rule C (historical pattern inference) can be: generating a "periodic payment" hypothesis, the content of which is: "Based on historical patterns, it is inferred that this payment is a periodic fee (period X) paid to [supplier name] for [service name], and it is recommended to match it to the most recent unpaid corresponding invoice (invoice number: XXX)."

[0097] The aforementioned pre-defined inference rules can yield one or more candidate matching hypotheses. Each candidate matching hypothesis carries a confidence level, which can be calculated based on rule strength, data completeness, and historical pattern consistency. In specific implementation, the confidence level of a candidate matching hypothesis can be calculated using the following formula:

[0098] Where α, β, and γ are the preset weight coefficients for each factor, and α + β + γ = 1. The weight allocation varies under different rules to highlight the most critical judgment criterion of that rule. All factor scores are within the interval [0, 1].

[0099] For the aforementioned pre-defined inference rule A (aggregate payment inference), the definitions and calculations of the various factors involved in the confidence level calculation can be found in Table 1 below:

[0100] Table 1

[0101] and then, .

[0102] For example, the rule strength weight is 0.85; data completeness score: assuming all three invoices are found and there are no other outstanding invoices, the score is 1.0; historical pattern matching score: assuming the customer's historical consolidated payment ratio is 80%, the score is 0.8; amount matching accuracy: assuming the voucher is 8500 yuan and the total invoice amount is 8500 yuan, the accuracy is 1.0. Furthermore, .

[0103] For the aforementioned pre-defined inference rule B (progress payment / partial payment inference), the definitions and calculations of the various factors involved in the confidence level calculation can be found in Table 2 below:

[0104] Table 2

[0105] and then, .

[0106] For example, rule strength weight: if the summary contains "progress payment", the weight is 0.9; data completeness score: if the contract amount is clear (1.0) and there is an installment plan (1.0), the score is 1.0; historical pattern consistency score: if the customer's historical installment payment accuracy rate is 90%, the score is 1.0; reasonableness of amount ratio: if the voucher amount is exactly 30% of the total contract amount, which meets the agreed ratio, the reasonableness of the amount ratio is 1.0.

[0107] and then, .

[0108] For the aforementioned pre-defined inference rule C (historical pattern inference), the definitions and calculations of the various factors involved in the confidence calculation can be found in Table 3 below:

[0109] Table 3

[0110] and then, .

[0111] For example, the rule strength weight is set to 0.7; the data completeness score is 1.0 if the records for the past 6 months are complete (1.0) and the amount fluctuates little (1.0); the historical pattern consistency score is 1.0 if the voucher date matches the expected date (1.0) and the amount of 5000 differs from the historical average of 5050 by 1% (0.99). The time consistency score is 1.0, the amount consistency score is 0.99, and the weighted average is 0.6 × 1.0 + 0.4 × 0.99 = 0.996. Furthermore, .

[0112] S305, if the confidence level corresponding to the candidate matching hypothesis exceeds the preset confidence threshold, the matching is determined to be successful and the association inference matching result is generated based on the candidate matching hypothesis.

[0113] Specifically, when there are multiple candidate matching hypotheses, the candidate matching hypothesis with the highest confidence can be selected, and it can be determined whether the highest confidence exceeds a preset confidence threshold (such as 0.85). If it does, the third-level association inference matching is successful. The association inference matching result can be generated based on the candidate matching hypothesis corresponding to the highest confidence. It can be understood that the association inference matching result can indicate that the matching is successful, and also includes the candidate matching hypothesis and the preset actions in the corresponding preset inference rules.

[0114] The specific value of the preset reliability threshold can be set based on practical experience.

[0115] Understandably, if the confidence level corresponding to the candidate matching hypothesis does not exceed the preset confidence threshold, it can be determined that the associated matching result indicates a matching failure.

[0116] S109, if the correlation reasoning matching result indicates a successful match, obtain the reconciliation result based on the correlation reasoning matching result.

[0117] The association reasoning matching result can indicate a successful match, and also includes candidate matching hypotheses and preset actions in the corresponding preset reasoning rules. The association recommendation matching result can be used as a reconciliation result, so that the information in the reconciliation result can be used for automatic accounting and reconciliation processing in the future.

[0118] For example, the transfer voucher information is as follows: {Summary: "Transfer", Amount: 8,500 yuan, Date: 2023-10-05, Recipient's Account Name: "Individual Business Owner C"}.

[0119] The results of the first-level exact match and the second-level fuzzy match are: no unique credential identifier, semantic match failed ("transfer" has no information).

[0120] Then proceed to the third level of association reasoning and matching: find the "Self-employed Individual C" node in the enterprise knowledge graph, and explore the three outstanding invoices associated with it from this node: Invoice A (3000 yuan), Invoice B (2000 yuan), and Invoice C (3500 yuan), all of which were issued in September 2023, with a total amount of 8500 yuan.

[0121] This triggers the preset inference rule A (aggregated payment inference). Since the amount (8500) in the transfer voucher information is completely consistent with the sum of the three invoices (8500), a candidate matching hypothesis with high confidence (assumed to be 0.95) is generated: "This payment is used to settle invoices A, B, and C in one lump sum."

[0122] Based on the candidate matching assumption, an associated matching result indicating a successful match can be generated, and this associated matching result can be used as the reconciliation result. Subsequently, when performing automatic posting and reconciliation processing based on this reconciliation result, a payment can be automatically associated with three invoices and all of them can be reconciled.

[0123] In some implementations, see also Figure 3 To further improve the matching success rate, the method may also include: S307, if the confidence level of the candidate matching hypothesis does not exceed the preset confidence threshold, determine whether the confidence level is greater than or equal to the second confidence threshold. If it is greater than or equal to the second confidence threshold, then execute the following step S309; ​​otherwise, if it is less than the second confidence threshold, then execute the following step S311.

[0124] Wherein, the second confidence threshold is less than the preset confidence threshold in the aforementioned step S305. For example, if the preset confidence threshold in step S305 is 0.85, then the second confidence threshold can be 0.6.

[0125] S309, Based on the candidate matching hypothesis, reasoning path and reasoning basis, generate reconciliation information to be reviewed, and display the reconciliation information to be reviewed on the display.

[0126] Specifically, the reconciliation information to be reviewed can be sent to the review node device, which can then display the information on its monitor, allowing the staff at the review node to manually review it.

[0127] S311, if the correlation reasoning result indicates that the match failed, add a marker to the virtual resource transfer certificate information, and the marker indicates that the match cannot be made.

[0128] Understandably, this tagging information can be added to the original virtual resource transfer certificate corresponding to the virtual resource transfer certificate information.

[0129] In some implementations, such as Figure 4 As shown, the method may further include: S401, Automatic accounting and reconciliation processing is performed based on the reconciliation results and the target business intent.

[0130] The virtual resource transfer certificate information includes the target business intent.

[0131] In practice, the automated posting and reconciliation process can invoke financial transaction interfaces based on reconciliation results and target business intent to generate financial transactions and debit / credit accounting vouchers. Simultaneously, it updates the accounting status, automatically marking fully received / paid amounts as reconciled to complete the reconciliation process and recording the processing log. This achieves seamless integration of business and finance, automatically linking business data from ERP (Enterprise Resource Planning) and CRM (Customer Relationship Management) systems, enabling automatic linking of business and financial flows, reducing manual cross-system verification steps, and improving information processing efficiency by over 60%.

[0132] In some implementations, incremental training data can be generated in response to confirmation, modification, or supplementary matching of reconciliation information to be reviewed, thereby obtaining an incremental training dataset. At a preset time point (such as the end of each quarter), the named entity recognition model and semantic similarity model can be updated and trained based on the incremental training dataset (such as incremental training or full fine-tuning), thereby achieving iterative optimization of information processing capabilities, which is conducive to further improving the matching success rate and thus improving the reconciliation success rate.

[0133] As can be seen from the above technical solutions of the embodiments of this application, the embodiments of this application improve the matching success rate of non-standard transfer vouchers through multi-level matching coupling of precise matching, fuzzy matching, and associative reasoning matching, thereby improving the reconciliation success rate in this scenario and reducing the difficulty of enterprise financial information management. Furthermore, the embodiments of this application streamline the entire process of reconciliation, posting, and write-off, achieving end-to-end automated processing, and establish a closed-loop feedback learning mechanism, enabling information processing capabilities to be iteratively optimized through historical data.

[0134] To facilitate understanding of the technical solutions in the embodiments of this application, the following is combined with... Figure 5 The system architecture used to implement the information processing method of the embodiments of this application is illustrated by way of example.

[0135] like Figure 5 As shown, the system architecture may include a data access and preprocessing module, an intelligent parsing and standardization module, a multi-level intelligent matching engine module, an automatic accounting and reconciliation module, and a manual review and feedback learning module.

[0136] The data access and preprocessing module includes multi-channel data interfaces (online banking interface, payment platform interface, file upload interface), image preprocessing subunit, OCR recognition subunit, and text preprocessing subunit, which are used to complete the access of original vouchers and standardized text output.

[0137] The intelligent parsing and standardization module includes a NER (Named Entity Recognition) model subunit, a knowledge graph subunit, and a text classification subunit, which are used to extract and standardize key entities and identify business intent.

[0138] The multi-level intelligent matching engine module includes a rule engine subunit, a semantic matching model subunit, and a knowledge graph reasoning subunit, used to perform multi-dimensional intelligent matching and output matching results and confidence scores. Specifically, for example... Figure 5 As shown, the first level of matching is exact matching (the rule engine matches unique credential identifiers); the second level of matching is fuzzy matching (semantic similarity calculation using word vectors / deep learning models); and the third level of matching is association reasoning matching (knowledge graph entity association reasoning).

[0139] The automatic posting and reconciliation module includes an accounting voucher generation sub-unit, an accounting status update sub-unit, and a reconciliation marking sub-unit, which are used to complete the end-to-end business execution of automatic posting and reconciliation.

[0140] The manual review and feedback learning module includes a visual review interface subunit, a training dataset storage subunit, and a model training subunit, which are used to collect manually corrected data and optimize the system model and rules.

[0141] Corresponding to the information processing methods provided in the above embodiments, this application also provides an information processing device. Since the information processing device provided in this application corresponds to the information processing methods provided in the above embodiments, the implementation methods of the aforementioned information processing methods are also applicable to the information processing device provided in this embodiment, and will not be described in detail in this embodiment.

[0142] Please see Figure 6 The diagram shown is a structural schematic of an information processing device provided in an embodiment of this application. This device has the function of implementing the information processing method described in the above-described method embodiments. This function can be implemented in hardware or by hardware executing corresponding software. Figure 6 As shown, the information processing device 600 may include: The credential information acquisition module 610 is used to acquire the virtual resource transfer credential information of the target object; The primary matching module 620 is used to perform primary matching between the virtual resource transfer voucher information and the set of accounting records to be matched based on a primary matching strategy; the primary matching strategy indicates that matching processing is performed based on a unique voucher identifier. The fuzzy matching module 630 is used to determine the fuzzy matching result based on the semantic similarity between the virtual resource transfer voucher information and the accounting records to be matched in the set of accounting records to be matched when the primary matching fails. The association reasoning and matching module 640 is used to perform association reasoning and matching based on the virtual resource transfer certificate information and the knowledge graph of the target object, and to obtain an association reasoning and matching result when the fuzzy matching result indicates that the matching fails; the knowledge graph includes multiple entities and the relationships between entities, and the multiple entities include the resource transfer object of the target object and the accounting record to be matched. The reconciliation result determination module 650 is used to obtain the reconciliation result based on the association reasoning matching result when the association reasoning matching result indicates that the matching is successful.

[0143] In some embodiments, the association reasoning matching module 640 includes: The pending matching accounting record search module is used to determine the first pending matching accounting record associated with the resource transfer object in the virtual resource transfer voucher information based on the knowledge graph of the target object; The inference matching module is used to perform inference matching based on the virtual resource transfer certificate information and the first accounting record to be matched, using preset inference rules to obtain inference matching results; the inference matching results include candidate matching hypotheses and corresponding confidence levels, and the preset inference rules indicate that, under preset inference conditions, candidate matching hypotheses and corresponding confidence levels are generated according to preset inference paths; The association reasoning result determination module is used to determine a successful match and generate the association reasoning matching result based on the candidate matching hypothesis when the confidence level corresponding to the candidate matching hypothesis exceeds a preset confidence threshold.

[0144] In some embodiments, the fuzzy matching module 630 includes: The candidate accounting record determination module is used to filter out candidate accounting records that meet the entity matching conditions from the set of accounting records to be matched based on the virtual resource transfer voucher information and the entity matching degree of the accounting records to be matched in at least one key entity dimension; the at least one key entity dimension includes the quantity of virtual resources, time, and resource transfer object; The semantic similarity determination module is used to call the semantic similarity model to determine the semantic similarity between the virtual resource transfer voucher information and the candidate matching accounting record based on the summary information in the virtual resource transfer voucher information and the business description in the candidate matching accounting record. The fuzzy matching result determination module is used to determine the fuzzy matching result based on the semantic similarity corresponding to the candidate accounting records to be matched and the entity matching degree in each of the key entity dimensions.

[0145] In some implementations, the fuzzy matching result determination module is specifically used to: for each candidate accounting record to be matched, perform a weighted summation of the corresponding semantic similarity and the entity matching degree in each of the key entity dimensions to obtain the comprehensive matching degree of the candidate accounting record to be matched; if the highest comprehensive matching degree does not meet the matching degree threshold condition, then determine that the fuzzy matching result indicates a matching failure; if the highest comprehensive matching degree meets the matching degree threshold condition, then determine that the matching is successful and obtain the reconciliation result based on the candidate accounting record to be matched corresponding to the highest comprehensive matching degree.

[0146] In some implementations, the primary matching module 620 is specifically configured to: if the virtual resource transfer certificate information contains a unique certificate identifier that meets a preset identifier encoding rule, then search for a target matching accounting record in the set of matching accounting records that matches the unique certificate identifier; if the target matching accounting record is not found, determine that the primary matching has failed; if the target matching accounting record is found, determine the reconciliation result based on the quantity difference between the number of virtual resources in the virtual resource transfer certificate information and the number of virtual resources in the target matching accounting record; if the virtual resource transfer certificate information does not contain a unique certificate identifier that meets the preset identifier encoding rule, then determine that the primary matching has failed.

[0147] In some embodiments, the credential information acquisition module 610 includes: The preprocessing module is used to determine the corresponding virtual resource transfer certificate text based on the original virtual resource transfer certificate of the target object; The entity recognition module is used to call the named entity recognition model to perform entity recognition on the virtual resource transfer certificate text to obtain at least one key entity; and to perform entity standardization on each key entity based on the knowledge graph. The intent recognition module is used to call the business intent recognition model to perform intent recognition on the virtual resource transfer certificate text to obtain the target business intent; The structured credential object generation module is used to generate a structured credential object based on at least one key entity after entity standardization and the target business intent. The structured credential object serves as virtual resource transfer credential information for the target object.

[0148] In some embodiments, the device 600 further includes: The automatic reconciliation and cancellation module is used to automatically process accounting entries and cancellations based on the reconciliation results and the target business intent.

[0149] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0150] This application provides an electronic device, which includes a processor and a memory. The memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement any of the information processing methods provided in the above method embodiments.

[0151] The method embodiments provided in this application can be executed in a computer terminal, server or similar computing device, that is, the above-mentioned electronic device may include a computer terminal, server or similar computing device. Figure 7 This is a hardware structure block diagram of an electronic device that runs an information processing method, as provided in the embodiments of this application. Figure 7 As shown, the internal structure of this electronic device may include, but is not limited to, a processor, a network interface, and a memory. The processor, network interface, and memory within the electronic device can be connected via a bus or other means, as illustrated in the embodiments of this specification. Figure 7 Taking the example of a connection between China and Israel via a bus.

[0152] The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device. The network interface may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.). Memory is the storage device in the computer device used to store programs and data. It is understood that the memory here can be a high-speed RAM storage device, or a non-volatile storage device, such as at least one disk storage device; optionally, it can also be at least one storage device located remotely from the aforementioned processor. The memory provides storage space, which stores the operating system of the electronic device, including but not limited to: Windows (an operating system), Linux (an operating system), Android (a mobile operating system), iOS (a mobile operating system), etc., which are not limited in this invention; and the storage space also stores one or more instructions suitable for loading and execution by the processor, which can be one or more computer programs (including program code). In the embodiments of this specification, the processor loads and executes one or more instructions stored in the memory to implement the information processing method provided in the above method embodiments.

[0153] Embodiments of this application also provide a computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one program related to implementing an information processing method. The at least one instruction or the at least one program is loaded and executed by the processor to implement any of the information processing methods provided in the above-described method embodiments.

[0154] Embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the information processing methods provided in the above-described method embodiments.

[0155] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0156] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0157] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0158] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0159] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An information processing method, characterized in that, The method includes: Obtain the virtual resource transfer certificate information of the target object; Based on a primary matching strategy, the virtual resource transfer voucher information is matched against the set of accounting records to be matched; the primary matching strategy indicates that matching is performed based on a unique voucher identifier. If the initial matching fails, a fuzzy matching result is determined based on the semantic similarity between the virtual resource transfer certificate information and the accounting records to be matched in the set of accounting records to be matched; If the fuzzy matching result indicates that the matching fails, based on the virtual resource transfer certificate information and the knowledge graph of the target object, a pre-defined inference rule is used to perform association inference matching to obtain an association inference matching result; the knowledge graph includes multiple entities and the relationships between entities, and the multiple entities include the resource transfer object of the target object and the accounting record to be matched. If the correlation reasoning matching result indicates a successful match, the reconciliation result is obtained based on the correlation reasoning matching result.

2. The method according to claim 1, characterized in that, The process of performing association reasoning and matching based on the virtual resource transfer certificate information and the knowledge graph of the target object, using preset reasoning rules, to obtain association reasoning and matching results includes: Based on the knowledge graph of the target object, determine the first accounting record to be matched associated with the resource transfer object in the virtual resource transfer certificate information; Based on the virtual resource transfer certificate information and the first accounting record to be matched, inference matching is performed using preset inference rules to obtain inference matching results; the inference matching results include candidate matching hypotheses and corresponding confidence levels, and the preset inference rules indicate that, under preset inference conditions, candidate matching hypotheses and corresponding confidence levels are generated according to preset inference paths; If the confidence level corresponding to the candidate matching hypothesis exceeds a preset confidence threshold, the matching is determined to be successful and the association inference matching result is generated based on the candidate matching hypothesis.

3. The method according to claim 1, characterized in that, The step of determining the fuzzy matching result based on the semantic similarity between the virtual resource transfer certificate information and the accounting records to be matched in the set of accounting records to be matched includes: Based on the virtual resource transfer certificate information and the entity matching degree of the accounting records to be matched in at least one key entity dimension, candidate accounting records that meet the entity matching conditions are selected from the set of accounting records to be matched; the at least one key entity dimension includes the quantity of virtual resources, time, and resource transfer object; The semantic similarity model is invoked to determine the semantic similarity between the virtual resource transfer certificate information and the candidate matching accounting record based on the summary information in the virtual resource transfer certificate information and the business description in the candidate matching accounting record. The fuzzy matching result is determined based on the semantic similarity of the candidate accounting records to be matched and the entity matching degree in each of the key entity dimensions.

4. The method according to claim 3, characterized in that, The determination of the fuzzy matching result based on the semantic similarity corresponding to the candidate accounting records to be matched and the entity matching degree in each of the key entity dimensions includes: For each candidate accounting record to be matched, the corresponding semantic similarity and the entity matching degree in each of the key entity dimensions are weighted and summed to obtain the comprehensive matching degree of the candidate accounting record to be matched. If the highest overall matching degree does not meet the matching degree threshold condition, then the fuzzy matching result indicates a matching failure. If the highest overall matching degree meets the matching degree threshold condition, the matching is determined to be successful and the reconciliation result is obtained based on the candidate accounting records to be matched corresponding to the highest overall matching degree.

5. The method according to claim 1, characterized in that, The preliminary matching of the virtual resource transfer certificate information with the set of accounting records to be matched based on the preliminary matching strategy includes: If the virtual resource transfer certificate information contains a unique certificate identifier that meets the preset identifier encoding rules, then search for the target matching accounting record in the set of matching accounting records that matches the unique certificate identifier; If the target accounting record to be matched is not found, the initial match is determined to have failed. If the target matching accounting record is found, the reconciliation result is determined based on the quantity difference between the quantity of virtual resources in the virtual resource transfer voucher information and the quantity of virtual resources in the target matching accounting record; If the virtual resource transfer certificate information does not contain a unique certificate identifier that meets the preset identifier encoding rules, then the initial matching is determined to have failed.

6. The method according to any one of claims 1 to 5, characterized in that, The process of obtaining the virtual resource transfer certificate information of the target object includes: Based on the original virtual resource transfer certificate of the target object, determine the corresponding virtual resource transfer certificate text; The named entity recognition model is invoked to perform entity recognition on the virtual resource transfer certificate text to obtain at least one key entity; based on the knowledge graph, each key entity is standardized. The business intent recognition model is invoked to perform intent recognition on the virtual resource transfer certificate text to obtain the target business intent; Based on at least one key entity after entity standardization and the target business intent, a structured credential object is generated, which serves as the virtual resource transfer credential information of the target object.

7. The method according to claim 6, characterized in that, The method further includes: Automatic accounting and reconciliation processes are performed based on the reconciliation results and the target business intent.

8. An information processing device, characterized in that, The device includes: The credential information acquisition module is used to acquire the virtual resource transfer credential information of the target object; The primary matching module is used to perform a primary matching between the virtual resource transfer voucher information and the set of accounting records to be matched, based on a primary matching strategy; the primary matching strategy indicates that the matching process is based on a unique voucher identifier. The fuzzy matching module is used to determine the fuzzy matching result based on the semantic similarity between the virtual resource transfer voucher information and the accounting records to be matched in the set of accounting records to be matched when the primary matching fails. The association reasoning and matching module is used to perform association reasoning and matching based on the virtual resource transfer certificate information and the knowledge graph of the target object, using preset reasoning rules, to obtain an association reasoning and matching result when the fuzzy matching result indicates that the matching fails. The knowledge graph includes multiple entities and the relationships between entities. The multiple entities include the resource transfer object of the target object and the accounting record to be matched. The reconciliation result determination module is used to obtain the reconciliation result based on the association reasoning matching result if the association reasoning matching result indicates a successful match.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the information processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor to implement the information processing method as described in any one of claims 1 to 7.

Images