Automatic financial transaction query, reply and response system and method
By constructing an automated financial transaction query and response system, and utilizing deep learning and natural language processing technologies, the system solves the problems of low efficiency and information interpretation bias in the process of financial transaction query and response, and achieves efficient and accurate transaction status monitoring and process automation.
Patent Information
- Application Number
- CN202510994867.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-10-31
AI Technical Summary
In the current process of querying and responding to financial transactions, manual operation is inefficient, making it difficult to achieve timely and comprehensive monitoring and processing. It is also susceptible to subjective factors, leading to information interpretation bias and potential risks.
An automated financial transaction query and response system is constructed, including an automatic transaction status identification module, an automatic query message generation and sending module, an NLP query and response message parsing module, an automatic response message generation module, and an automatic business status advancement module. Through deep learning and natural language processing technologies, transaction status identification, message generation, and automatic business status advancement are achieved.
It improved transaction processing efficiency, reduced labor costs, enhanced the accuracy and stability of the transaction process, and reduced the risk of information interpretation errors and transaction disputes.
Smart Images
Figure CN120873141A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of transaction inquiry technology, and in particular to an automated financial transaction inquiry and response system and method. Background Technology
[0002] In the daily operations of the financial industry, with the continuous expansion of transaction volume and the increasing complexity of transaction types, every financial transaction must go through a clearing process after the payment instruction is completed. Taking common large and small value clearing instructions as an example, after the instruction is issued, business personnel need to monitor the transaction status in real time and manually initiate large and small value query messages to understand the transaction processing progress; after receiving the other party's query reply message, they also need to manually judge to advance the subsequent transaction process. As the party being queried, business personnel also need to manually complete the response to the query message based on the actual status of the transaction.
[0003] This traditional operating model has significant drawbacks: on the one hand, manual operation is inefficient. When faced with massive transactions, business personnel find it difficult to monitor and process them in a timely and comprehensive manner, which not only prolongs the transaction processing cycle but also slows down the turnover of funds. On the other hand, manual operation is greatly affected by subjective factors and is prone to errors in filling out reports and misinterpreting information due to negligence, which may lead to transaction disputes and pose potential risks to financial institutions.
[0004] In the context of increasingly fierce competition in the financial market, financial institutions urgently need to leverage innovative technologies to optimize existing inquiry and response processes and comprehensively improve the efficiency and quality of transaction processing. Summary of the Invention
[0005] This invention provides an automated financial transaction query and response system and method. By constructing an efficient, intelligent and accurate automated system, it improves transaction processing efficiency, reduces labor costs, and enhances the accuracy and stability of the transaction process.
[0006] According to one aspect of the present invention, an automated financial transaction query and response system is provided. The system includes: an automatic transaction status identification module, an automatic query message generation and sending module, an NLP query and response message parsing module, an automatic response message generation module, and an automatic business status progression module; wherein,
[0007] The automatic transaction status identification module is connected to the automatic query message generation and sending module, the automatic reply message generation module, and the automatic business status advancement module, and is used to obtain the transaction status; wherein, the transaction status refers to the state of the transaction at the current stage;
[0008] The automatic query message generation and sending module is used to generate a query message based on the transaction status.
[0009] The NLP query and reply message parsing module is connected to the automatic reply message generation module and the automatic business status advancement module. It is used to obtain the query and reply message corresponding to the query message and extract the structured key information in the query and reply message. The structured key information is used to represent the core information in the query and reply message.
[0010] The automatic response message generation module is connected to the automatic business status advancement module and is used to generate a response message based on the structured key information and the transaction status.
[0011] The automatic business status advancement module is used to generate a target business status based on the structured key information, the reply message, and the transaction status; wherein, the target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.
[0012] According to another aspect of the present invention, an automated financial transaction query and response method is provided, the method comprising:
[0013] Obtain the transaction status; wherein, the transaction status refers to the state of the transaction at the current stage;
[0014] Generate a query message based on the transaction status;
[0015] Obtain the reply message corresponding to the query message, and extract the structured key information from the reply message; wherein the structured key information is used to characterize the core information in the reply message;
[0016] Based on the structured key information and the transaction status, a response message is generated;
[0017] A target business status is generated based on the structured key information, the response message, and the transaction status; wherein, the target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.
[0018] The technical solution of this invention, through an automatic transaction status identification module connected to an automatic query message generation and sending module, an automatic reply message generation module, and an automatic business status advancement module, is used to obtain the transaction status; the automatic query message generation and sending module is used to generate query messages based on the transaction status; the NLP query-reply message parsing module, connected to the automatic reply message generation module and the automatic business status advancement module, is used to obtain structured key information corresponding to the query message; the automatic reply message generation module, connected to the automatic business status advancement module, is used to generate reply messages based on the structured key information and the transaction status; and the automatic business status advancement module is used to generate the target business status based on the structured key information, the reply message, and the transaction status. This technical solution, by constructing an efficient, intelligent, and accurate automated system, improves transaction processing efficiency, reduces labor costs, and enhances the accuracy and stability of the transaction process.
[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a schematic diagram of an automated financial transaction query and response system provided in Embodiment 1 of the present invention;
[0022] Figure 2 The flowchart for automatic transaction status recognition provided in Embodiment 1 of this application;
[0023] Figure 3 This is a flowchart of the automatic generation and sending of query messages provided in Embodiment 1 of this application;
[0024] Figure 4 A flowchart for parsing a query message provided in Embodiment 1 of this application;
[0025] Figure 5 The flowchart of the automatic reply message provided in Embodiment 1 of this application;
[0026] Figure 6 The flowchart for the automatic progress of business status provided in Embodiment 1 of this application;
[0027] Figure 7This is a flowchart of an automated financial transaction query and response method provided in Embodiment 2 of the present invention. Detailed Implementation
[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0029] It should be noted that the terms "history," "target," etc., used in the specification, claims, and accompanying drawings of this invention 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 the invention 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 apparatus 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 apparatus.
[0030] Example 1
[0031] Figure 1 This is a schematic diagram of an automated financial transaction inquiry and response system according to Embodiment 1 of the present invention, as shown below. Figure 1 As shown, the system includes: an automatic transaction status identification module 110, an automatic query message generation and sending module 120, an NLP query and reply message parsing module 130, an automatic reply message generation module 140, and an automatic business status advancement module 150; among which,
[0032] The automatic transaction status identification module 110 is connected to the automatic query message generation and sending module 120, the automatic reply message generation module 140, and the automatic business status advancement module 150, and is used to obtain the transaction status; wherein, the transaction status refers to the state of the transaction at the current stage;
[0033] The query message automatic generation and sending module 120 is used to generate a query message based on the transaction status.
[0034] The NLP query and reply message parsing module 130 is connected to the automatic reply message generation module 140 and the automatic business status advancement module 150. It is used to obtain the query and reply message corresponding to the query message and extract the structured key information in the query and reply message. The structured key information is used to represent the core information in the query and reply message.
[0035] The automatic response message generation module 140 is connected to the automatic business status advancement module 150 and is used to generate a response message based on the structured key information and the transaction status.
[0036] The automatic business status advancement module 150 is used to generate a target business status based on the structured key information, the reply message, and the transaction status; wherein, the target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.
[0037] The transaction status refers to the current state of a transaction. For example, the transaction status could be submitted for review, approved and awaiting settlement, settlement in progress, or settlement completed and awaiting final settlement.
[0038] In this solution, the transaction status automatic identification module 110 determines the transaction status of the financial transaction and sends the transaction status to the query message automatic generation and sending module 120, the reply message automatic generation module 140, and the business status automatic advancement module 150.
[0039] In this embodiment, a query message is an inquiry information carrier sent by the business party or system to the counterparty during a financial transaction in order to obtain information such as the processing progress and status of a certain transaction. Its content includes key identifiers related to the transaction and the query intent.
[0040] In this solution, the query message automatic generation and sending module 120 generates query messages based on the transaction status. Specifically, it can extract information related to the transaction status and fill this information into the corresponding query message template to automatically generate the query message.
[0041] Furthermore, the query message automatic generation and sending module 120 sends the generated query message to the counterparty.
[0042] Among them, the query response message refers to the response made by the counterparty to the received query message, which details the transaction processing results, reasons and other information, and is an important basis for promoting the subsequent transaction process.
[0043] In this embodiment, structured key information is used to represent the core information in the query message. The structured key information is presented in the form of key-value pairs. For example, the structured key information can be {"Transaction Result": "Failed", "Amount": "500 yuan", "Reason for Failure": "Incorrect Respondent Account Information"}, etc.
[0044] Specifically, the NLP query message parsing module 130 receives query messages sent by the counterparty, parses the content of the query message using natural language processing (NLP) technology, and accurately extracts structured key information from it.
[0045] Furthermore, the NLP query and reply message parsing module 130 sends the structured key information to the reply message automatic generation module 140 and the business status automatic advancement module 150.
[0046] In this solution, the automatic response message generation module 140 receives structured key information output by the NLP response message parsing module 130 and transaction status provided by the automatic transaction status identification module 110. These two pieces of information are integrated to gain a comprehensive understanding of the transaction's background and current status. For a specific transaction, the integrated information is filled into the corresponding position in the matching response template to generate a complete response message. A response message is a structured information carrier that responds to received requests, inquiries, notifications, and other messages during information exchange.
[0047] Furthermore, the automatic response message generation module 140 sends the response message to the automatic business status advancement module 150.
[0048] In this embodiment, the automatic business status advancement module 150 integrates structured key information, response messages, and automatic transaction status query content, and generates a target business status based on the query content. The target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.
[0049] Furthermore, based on pre-defined business progression rules, it is determined whether the query and reply content meets the business status transition conditions. When the conditions are met, the business status is automatically updated to the corresponding target business status.
[0050] Optionally, the automatic transaction status identification module 110 is specifically used for:
[0051] Obtain target transaction data;
[0052] The target transaction data is input into the target transaction state prediction model, and the transaction state is output; wherein, the target transaction state prediction model is a long short-term memory network model.
[0053] Among them, the target transaction status prediction model is the Long Short-Term Memory (LSTM) network model. The LSTM network can effectively process time series data and capture the characteristic changes and long-term dependencies of transaction data at different time points.
[0054] In this plan, Figure 2 The flowchart for automatic transaction status recognition provided in Embodiment 1 of this application is as follows: Figure 2 As shown, the automatic transaction status identification module 110 acquires target transaction data in real time and inputs it into the target transaction status prediction model. The model predicts the current status of the transaction based on the learned patterns. The output transaction status may include various states such as submitted for review, approved and awaiting settlement, settlement in progress, and settlement completed and awaiting settlement, providing key basis for the automatic generation of subsequent query messages.
[0055] By using a target transaction status prediction model, the transaction status corresponding to the target transaction data can be automatically identified, thereby improving transaction processing efficiency, reducing labor costs, and enhancing the accuracy and stability of the transaction process.
[0056] Optionally, the transaction status automatic identification module 110 is further configured to:
[0057] Acquire historical transaction data; wherein, the historical transaction data is data generated at each stage of financial transactions;
[0058] Determine the initial parameters of the transaction state prediction model to be trained;
[0059] The historical transaction data is input into the transaction state prediction model to be trained, and the predicted transaction state is output.
[0060] Based on the predicted transaction status and the preset actual transaction status, the loss function value is determined;
[0061] The initial parameters of the transaction state prediction model to be trained are adjusted based on the loss function value to obtain the target transaction state prediction model.
[0062] In this embodiment, as Figure 2 As shown, the automatic transaction status identification module 110 extensively collects multi-source data generated at various stages of financial transactions, including but not limited to transaction order information (such as transaction amount, transaction time, and accounts of both parties), clearing system feedback data (such as clearing instruction status and clearing timestamp), and bank account balance change information, and treats this multi-source data as historical transaction data. By connecting to different data sources through data interfaces, this data is aggregated into a unified data storage platform, where it is cleaned and preprocessed to ensure data accuracy and consistency.
[0063] Furthermore, a transaction status prediction model is constructed using Recurrent Neural Networks (RNNs) and their variant, Long Short-Term Memory (LSTM), from deep learning. The model is trained using historical transaction data. During training, data features from different transaction stages and their corresponding actual transaction states are used as training samples. By continuously adjusting the model parameters, the model learns the mapping relationship between data features and transaction states. For example, the data feature of no feedback from the clearing system for a period after a transaction is submitted is associated with the transaction status awaiting confirmation during clearing.
[0064] The initial parameters of the trading state prediction model to be trained include network structure parameters, regularization parameters, and other parameters. These initial parameters can be set according to the training requirements of the trading state prediction model.
[0065] In this scheme, historical transaction data is input into the transaction state prediction model to be trained, and the model processes the historical transaction data and outputs the predicted transaction state.
[0066] The loss function is a function used to measure the difference between the model's predictions and the actual results. Loss functions can include mean squared error, mean absolute error, mean absolute percentage error, log loss function, etc.
[0067] Specifically, during model training, the loss function value is calculated using the predicted trading state and the actual trading state. By calculating this loss function value, the difference between the model's predicted trading state and the actual trading state can be scientifically quantified, thereby comprehensively and accurately evaluating the model's predictive performance and generalization ability under the current training state.
[0068] In this scheme, the gradients of each initial parameter in the model are calculated using the loss function value and the backpropagation algorithm. The gradient represents the rate of change of the loss function under the current parameters, indicating the direction and magnitude of parameter adjustment. Then, based on the calculated gradients, an appropriate optimization algorithm, such as stochastic gradient descent (SGD) or adaptive moment estimation (Adam), is selected to update the initial parameters.
[0069] Furthermore, during the parameter update process, the entire target encoding needs to be trained iteratively multiple times. Each iteration repeats the above process. Through continuous iterative training, the model's parameters are gradually adjusted, making the model's predicted trading state increasingly closer to the actual trading state, and the loss function value gradually decreases. When the loss function value reaches a pre-set small value, or when the change in the loss function value is no longer significant after multiple iterations, the model is considered to have converged, and the resulting model is the target trading state prediction model.
[0070] Deep learning models have powerful learning capabilities, enabling them to learn complex trading patterns and rules from massive amounts of historical trading data. The automatic trading status recognition module can adapt to status judgment in various complex financial trading scenarios.
[0071] Optionally, the query message automatic generation and sending module 120 is specifically used for:
[0072] Construct a query rule base; wherein the query rule base stores the matching relationship between transaction status and query message rules; the query message rules are specifications and criteria used to extract target information from transaction status.
[0073] In this solution, the query message automatic generation and sending module 120 establishes a comprehensive and flexible query rule base. This rule base pre-sets the key information to be queried and the corresponding query message format under different circumstances based on various factors such as transaction type (e.g., remittance, securities trading, bill settlement), transaction status, and risk level. For example, for cross-border remittance transactions, when the transaction status is in the clearing process for a long time, the rule base stipulates that information such as the reason for the clearing delay and the estimated completion time should be queried, and the corresponding cross-border remittance query message template should be matched.
[0074] The flexible configuration of the query rule base enables the system to generate personalized query messages based on different transaction types, transaction conditions, and business needs, easily responding to the ever-changing transaction demands and new business models in the financial market.
[0075] Optionally, the query message automatic generation and sending module 120 is further configured to:
[0076] Determine whether the transaction status meets the query trigger condition;
[0077] If the conditions are met, then based on the matching relationship between transaction status and query message rules in the query rule base, a query message rule matching the transaction status is determined;
[0078] Based on the query message rules, target information is extracted from the transaction status;
[0079] The target information is filled into a preset query message template to generate a query message.
[0080] In this embodiment, Figure 3 This is a flowchart of the automatic generation and sending of query messages provided in Embodiment 1 of this application, such as... Figure 3As shown, when the transaction status output by the automatic transaction status identification module 110 meets the query trigger conditions, the system retrieves matching query message rules from the query rule base. Based on the query message rules, it extracts the target information of the current transaction, such as the transaction number, amount, and participating account information, and fills this target information into the corresponding query message template, automatically generating a complete query message. For example, if a transfer transaction is in a state of "submitted for settlement for more than the prescribed time," the system extracts information such as the transaction number, transfer amount, and the transferring and receiving accounts according to the rules, and generates a query message according to the transfer transaction query message template.
[0081] Furthermore, through a standard interface that connects to the financial communication network, the generated query messages are accurately sent to the counterparty. Simultaneously, the system tracks and records the sent query messages, noting the sending time, message number, recipient, and other information for subsequent querying and management.
[0082] By building an efficient, intelligent, and precise automated system, we can improve transaction processing efficiency, reduce labor costs, and enhance the accuracy and stability of the transaction process.
[0083] Optionally, the NLP query message parsing module 130 is specifically used for:
[0084] Receive reply messages;
[0085] The query message is segmented into words to obtain the target query message;
[0086] The target reply message is input into the reply message parsing model, which outputs structured key information corresponding to the reply message; wherein, the reply message parsing model is the BERT model.
[0087] In this plan, Figure 4 The flowchart for parsing the query message provided in Embodiment 1 of this application is as follows: Figure 4 As shown, BERT (Bidirectional Encoder Representations from Transformers), a pre-trained language model based on the Transformer architecture, is used as the core model for message parsing. The BERT model possesses powerful bidirectional semantic understanding capabilities, simultaneously considering contextual information and effectively handling complex semantic relationships in natural language text. Its architecture consists of multiple stacked Transformer blocks, each containing a multi-head attention mechanism and a feedforward neural network. Through self-attention, it weights each word in the input text, thereby better capturing key information within the text.
[0088] Further, preprocess the received reply message. First, use a word segmentation tool to split the message text into individual words or sub-word units, remove stop words (e.g., words like "of", "in", "is" that have no actual semantic contribution), and then perform词性标注 on each word to determine its grammatical role in the message. For example, segment "The transaction has been successfully completed, and the amount is 1000 yuan" into "transaction", "has", "successfully", "completed", ",", "amount", "is", "1000", "yuan", and mark the词性. After preprocessing, convert the message into a target reply message that the model can handle, where the target reply message is represented by word vectors.
[0089] In this embodiment, fine-tune the pre-trained BERT model using a large amount of reply message data in the financial field. During the fine-tuning process, use the reply message and its corresponding structured annotation data (such as transaction result, amount, processing time, reason, etc.) as training samples, and adjust the model parameters through the backpropagation algorithm to enable the model to better adapt to the characteristics and semantic understanding requirements of financial reply messages. For example, for a reply message "Due to incorrect information in the other party's account, the transaction failed, the amount is 500 yuan, and the time is 2024-10-01 10:00:00", the model can accurately extract key information such as the transaction failure result, the amount of 500 yuan, and the failure time after fine-tuning training.
[0090] Further, input the preprocessed target reply message into the fine-tuned BERT model. The model performs in-depth semantic analysis on the target reply message, identifies the key information in the message through its internal multi-layer neural network structure, and outputs this information in a pre-defined structured format. The output result is presented in the form of key-value pairs, such as {"transaction result": "failed", "amount": "500 yuan", "reason for failure": "incorrect information in the other party's account", "processing time": "2024-10-01 10:00:00"}, which is convenient for subsequent modules to further process.
[0091] NLP technology analyzes reply messages based on deep learning models, can accurately understand complex and diverse expressions in messages, and the accuracy of extracting key information is much higher than that of manual processing. The model trained with a large amount of financial reply message data has a more accurate grasp of various professional terms and semantic contexts, effectively avoiding problems such as information omission and misunderstanding that may occur in manual parsing. It ensures the accuracy of subsequent reply message generation and the correctness of business status advancement, and reduces transaction risks caused by information errors. From a set of preset response message templates, a target response message template is matched based on the structured key information and the transaction status.
[0095] The structured key information and the transaction status are mapped to the corresponding variable positions in the target response message template to generate a response message.
[0096] Specifically, Figure 5 The flowchart of the automatic reply message provided in Embodiment 1 of this application is as follows: Figure 5 As shown, the system receives structured key information output by the NLP query message parsing module 130 and transaction status information provided by the automatic transaction status identification module 110. These two pieces of information are integrated to gain a comprehensive understanding of the transaction's background and current status. For example, if the query result indicates a successful transaction and the current transaction status is "clearing," combining these two pieces of information clarifies that the transaction has completed clearing and is about to enter the settlement stage.
[0097] Furthermore, such as Figure 5 As shown, based on a pre-set response template library, the system matches corresponding target response message templates according to different transaction situations and structured key information. The template library covers response content frameworks for various common financial transaction scenarios, such as transaction success response templates, transaction failure response templates, and transaction delay response templates. For a specific transaction, the integrated transaction information and structured key information are filled into the corresponding positions of the matched target response message template to generate a complete response message. For example, for a reply to a successful transaction inquiry, a successful transaction response template is selected from the template library, and the transaction number, successful amount, completion time, and other information are filled into the template to generate a response message that reads, "The transaction you inquired about with [Transaction Number] has been successfully completed, the transaction amount is [Successful Amount], and the completion time is [Completion Time]."
[0098] In this solution, after generating the response message, the system performs syntactic and semantic checks on the message to ensure that the message is accurate, clear, and unambiguous. Simultaneously, the message is appropriately polished according to preset optimization rules, such as standardizing format and terminology, to enhance the professionalism and readability of the response message before it is sent to the querying party.
[0099] The automated generation of response messages significantly reduces manual processing time, improving the efficiency of the entire transaction query and response process several times over. This greatly enhances the speed of fund transfers and the business processing capabilities of financial institutions. The flexible configuration of the response template library allows the system to generate personalized response messages based on different transaction types, transaction conditions, and business needs, easily responding to the ever-changing transaction demands and new business models in the financial market.
[0100] Optionally, the automatic business status advancement module 150 is specifically used for:
[0101] Receive the structured key information, the response message, and the transaction status;
[0102] Based on the structured key information, the reply message, and the transaction status, generate the query response content;
[0103] Determine whether the content of the query and reply meets the conditions for business status transition;
[0104] If the conditions are met, the target business status will be generated based on the query response content.
[0105] In this plan, Figure 6 The flowchart for the automatic progress of business status provided in Embodiment 1 of this application is as follows: Figure 6 As shown, the system integrates the structured key information extracted by the NLP query and reply message parsing module 130, the transaction status recorded by the transaction status automatic identification module 110, and the reply message generated by the reply message automatic generation module 140. Based on pre-set business progression rules, it determines whether the integrated query and reply content meets the business status transition conditions. For example, if the query and reply content shows that the transaction was successful and the current business status is "clearing," then according to the business progression rules, the condition for updating the business status to "transaction completed and awaiting settlement" is met.
[0106] Furthermore, if the content of the query and reply meets the conditions for business status transition, the target business status is generated based on the content of the query and reply.
[0107] The automation of business status updates has reduced a significant amount of manual processing time, and the efficiency of the entire transaction query and response process has been improved several times over.
[0108] Optionally, the automatic business status advancement module 150 is further configured to:
[0109] The target business status is sent to the transaction status automatic identification module to update the transaction status based on the target business status.
[0110] Furthermore, such as Figure 6 As shown, when the conditions for a business status transition are met, the system automatically updates the business status to the corresponding target business status and synchronizes the updated target business status to all relevant modules of the financial transaction system, such as the accounting module, risk management module, and customer information module. Simultaneously, it records detailed historical information about the business status change, including the change time, reason for change, and the status before and after, for subsequent querying and auditing. For example, after updating the business status from "in clearing" to "transaction completed and awaiting settlement," the accounting module is simultaneously notified to prepare for fund settlement operations, and the risk management module is notified to adjust the risk status of the transaction.
[0111] In this solution, if any anomalies occur during the business process, such as discrepancies between the requested and received responses or conflicts with business rules, the system will activate an anomaly handling mechanism. Detailed records and analyses of these anomalies will be kept, and alerts will be sent to relevant personnel to prompt manual intervention. For example, if a response indicates a transaction failure, but this should not occur according to business rules, the system will record the reason for the failure, relevant transaction information, and send an SMS or email alert to the business supervisor to facilitate timely problem investigation and ensure the smooth operation of the transaction process.
[0112] The automation of business status updates has reduced a significant amount of manual processing time, and the efficiency of the entire transaction query and response process has been improved several times over.
[0113] The technical solution of this invention, through an automatic transaction status identification module connected to an automatic query message generation and sending module, an automatic reply message generation module, and an automatic business status advancement module, is used to obtain the transaction status. The automatic query message generation and sending module generates query messages based on the transaction status. An NLP query-reply message parsing module, connected to the automatic reply message generation module and the automatic business status advancement module, is used to obtain structured key information corresponding to the query message. The automatic reply message generation module, connected to the automatic business status advancement module, generates reply messages based on the structured key information and the transaction status. The automatic business status advancement module generates the target business status based on the structured key information, the reply message, and the transaction status. By executing this technical solution, the query, reply, and business status advancement stages are organically integrated to form an intelligent closed-loop system. Information is shared and collaborative among the modules, and intelligent processing is performed based on the entire transaction process information, ensuring the continuity and integrity of the transaction process. It avoids problems such as information gaps and business delays caused by the disconnect between various links in the traditional model, improves the overall intelligence and stability of financial transaction processing, and provides financial institutions with more efficient and reliable transaction support.
[0114] Example 2
[0115] Figure 7 This is a flowchart of an automated financial transaction query and response method according to Embodiment 2 of the present invention. This method can be executed by an automated financial transaction query and response system. Figure 7 As shown, the method includes:
[0116] S710. Obtain the transaction status; wherein, the transaction status refers to the state of the transaction at the current stage.
[0117] The transaction status refers to the current state of a transaction. For example, the transaction status could be submitted for review, approved and awaiting settlement, settlement in progress, or settlement completed and awaiting final settlement.
[0118] In this solution, the automatic transaction status identification module acquires target transaction data in real time and inputs it into the target transaction status prediction model. The model predicts the current status of the transaction based on learned patterns. The output transaction status may include various states such as submitted and awaiting review, approved and awaiting settlement, settlement in progress, and settlement completed and awaiting final settlement, providing crucial information for the automatic generation of subsequent query messages. The target transaction status prediction model is a Long Short-Term Memory (LSTM) network model. LSTM networks can effectively process time-series data, capturing the characteristic changes and long-term dependencies of transaction data at different time points.
[0119] S720. Generate a query message based on the transaction status.
[0120] In this embodiment, a query message is an inquiry information carrier sent by the business party or system to the counterparty during a financial transaction in order to obtain information such as the processing progress and status of a certain transaction. Its content includes key identifiers related to the transaction and the query intent.
[0121] In this embodiment, when the transaction status output by the automatic transaction status identification module meets the query triggering conditions, the system retrieves matching query message rules from the query rule base. Based on the query message rules, the system extracts the target information of the current transaction, such as the transaction number, amount, and participating account information, and fills this target information into the corresponding query message template, automatically generating a complete query message. For example, if a transfer transaction is in a submitted pending settlement state that has exceeded the specified time, the system extracts information such as the transaction number, transfer amount, and the sending and receiving accounts according to the rules, and generates a query message according to the transfer transaction query message template.
[0122] S730. Obtain the reply message corresponding to the query message, and extract the structured key information from the reply message; wherein the structured key information is used to characterize the core information in the reply message.
[0123] Among them, the query response message refers to the response made by the counterparty to the received query message, which details the transaction processing results, reasons and other information, and is an important basis for promoting the subsequent transaction process.
[0124] In this embodiment, structured key information is used to represent the core information in the query message. The structured key information is presented in the form of key-value pairs. For example, the structured key information can be {"Transaction Result": "Failed", "Amount": "500 yuan", "Reason for Failure": "Incorrect Respondent Account Information"}, etc.
[0125] Specifically, a pre-trained language model based on the Transformer architecture - BERT (Bidirectional Encoder Representations from Transformers) is adopted as the core model for parsing the reply message. The BERT model has powerful bidirectional semantic understanding capabilities, can consider the context information of the text simultaneously, and effectively handle the complex semantic relationships in natural language text. Its architecture is composed of multiple stacked Transformer blocks, each Transformer block contains a multi-head attention mechanism and a feed-forward neural network, and the self-attention mechanism is used to weight each word in the input text, so as to better capture the key information in the text.
[0126] Furthermore, preprocess the received reply message. First, use a tokenization tool to split the message text into individual words or sub-word units, remove stop words (such as words with no actual semantic contribution like "的", "在", "是", etc.), and then perform词性标注 on each word to determine its grammatical role in the message. For example, split the sentence "交易已成功完成,金额为1000元" into "交易", "已", "成功", "完成", ",", "金额", "为", "1000", "元", and mark their词性. After preprocessing, convert the message into a target reply message that the model can handle, where the target reply message is represented by word vectors.
[0127] In this embodiment, fine-tune the pre-trained BERT model using a large amount of reply message data in the financial field. During the fine-tuning process, use the reply message and its corresponding structured annotation data (such as transaction result, amount, processing time, reason, etc.) as training samples, and adjust the model parameters through the backpropagation algorithm to make the model better adapt to the characteristics and semantic understanding requirements of financial reply messages. For example, for a reply message "因对方账户信息有误,交易失败,金额500元,时间2024-10-01 10:00:00", the model can accurately extract key information such as the transaction failure result, amount of 500 yuan, and failure time after fine-tuning training.
[0128] Furthermore, input the preprocessed target reply message into the fine-tuned BERT model. The model performs in-depth semantic analysis on the target reply message, identifies the key information in the message through its internal multi-layer neural network structure, and outputs this information in a pre-defined structured format. The output result is presented in the form of key-value pairs, such as {"交易结果": "失败", "金额": "500元", "失败原因": "对方账户信息有误", "处理时间": "2024-10-01 10:00:00"}, which is convenient for subsequent modules to further process.
[0129] It should be noted that "词性标注" in the Chinese text is a specific term in linguistics for the process of marking the grammatical category of words. Since there is no exact English equivalent provided in the context, it is left in Chinese for now. If there is a more specific instruction on how to handle this term, it can be adjusted accordingly.S740. Based on the structured key information and the transaction status, generate a reply message.
[0130] A response message is a structured information carrier that responds to received requests, inquiries, notifications, and other messages during information exchange.
[0131] Specifically, it receives structured key information output by the NLP query message parsing module and transaction status information provided by the automatic transaction status identification module. Integrating these two pieces of information provides a comprehensive understanding of the transaction's origins and current status. For example, if the query result indicates a successful transaction and the current transaction status is "clearing," combining these two pieces of information clarifies that the transaction has completed clearing and is about to enter the settlement phase.
[0132] Furthermore, based on a pre-set response template library, corresponding target response message templates are matched according to different transaction situations and structured key information. The template library covers response content frameworks for various common financial transaction scenarios, such as transaction success response templates, transaction failure response templates, and transaction delay response templates. For a specific transaction, the integrated transaction information and structured key information are filled into the corresponding positions of the matched target response message template to generate a complete response message. For example, for a reply to a successful transaction inquiry, a successful transaction response template is selected from the template library, and information such as the transaction number, successful amount, and completion time are filled into the template to generate a response message such as "The transaction you inquired about [Transaction Number] has been successfully completed, the transaction amount is [Successful Amount], and the completion time is [Completion Time]".
[0133] In this solution, after generating the response message, the system performs syntactic and semantic checks on the message to ensure that the message is accurate, clear, and unambiguous. Simultaneously, the message is appropriately polished according to preset optimization rules, such as standardizing format and terminology, to enhance the professionalism and readability of the response message before it is sent to the querying party.
[0134] S750. Generate a target business status based on the structured key information, the reply message, and the transaction status; wherein, the target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.
[0135] This solution integrates structured key information extracted by the NLP query and reply message parsing module, transaction status recorded by the automatic transaction status identification module, and reply messages generated by the automatic reply message generation module. Based on pre-defined business progression rules, it determines whether the integrated query and reply content meets the business status transition conditions. For example, if the query and reply content shows a successful transaction and the current business status is "clearing," then according to the business progression rules, the condition for updating the business status to "transaction completed and awaiting settlement" is met.
[0136] Furthermore, if the content of the query and reply meets the conditions for business status transition, the target business status is generated based on the content of the query and reply.
[0137] The technical solution of this invention involves: acquiring the transaction status; generating a query message based on the transaction status; acquiring a corresponding reply message and extracting structured key information from the reply message; generating a response message based on the structured key information and the transaction status; and generating a target business status based on the structured key information, the response message, and the transaction status. By executing this technical solution, the query, reply, and business status processes are organically integrated into an intelligent closed-loop system. Information is shared and collaborative among the modules, and intelligent processing is performed based on the entire transaction process information, ensuring the continuity and integrity of the transaction process. This avoids the problems of information bottlenecks and business delays caused by disconnections in the traditional model, improving the overall intelligence and stability of financial transaction processing, and providing financial institutions with more efficient and reliable transaction support.
[0138] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.
[0139] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. An automated financial transaction inquiry and response system, characterized in that, The system includes: an automatic transaction status identification module, an automatic query message generation and sending module, an NLP query and reply message parsing module, an automatic reply message generation module, and an automatic business status progression module; wherein... The automatic transaction status identification module is connected to the automatic query message generation and sending module, the automatic reply message generation module, and the automatic business status advancement module, and is used to obtain the transaction status; wherein, the transaction status refers to the state of the transaction at the current stage; The automatic query message generation and sending module is used to generate a query message based on the transaction status. The NLP query and reply message parsing module is connected to the automatic reply message generation module and the automatic business status advancement module. It is used to obtain the query and reply message corresponding to the query message and extract the structured key information in the query and reply message. The structured key information is used to represent the core information in the query and reply message. The automatic response message generation module is connected to the automatic business status advancement module and is used to generate a response message based on the structured key information and the transaction status. The automatic business status advancement module is used to generate a target business status based on the structured key information, the reply message, and the transaction status; wherein, the target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.
2. The system according to claim 1, characterized in that, The automatic transaction status identification module is specifically used for: Obtain target transaction data; The target transaction data is input into the target transaction state prediction model, and the transaction state is output; wherein, the target transaction state prediction model is a long short-term memory network model.
3. The system according to claim 2, characterized in that, The automatic transaction status identification module is also used for: Acquire historical transaction data; wherein, the historical transaction data is data generated at each stage of financial transactions; Determine the initial parameters of the transaction state prediction model to be trained; The historical transaction data is input into the transaction state prediction model to be trained, and the predicted transaction state is output. Based on the predicted transaction status and the preset actual transaction status, the loss function value is determined; The initial parameters of the transaction state prediction model to be trained are adjusted based on the loss function value to obtain the target transaction state prediction model.
4. The system according to claim 1, characterized in that, The automatic query message generation and sending module is specifically used for: Construct a query rule base; wherein the query rule base stores the matching relationship between transaction status and query message rules; the query message rules are specifications and criteria used to extract target information from transaction status.
5. The system according to claim 4, characterized in that, The automatic generation and sending module for query messages is also used for: Determine whether the transaction status meets the query trigger condition; If the conditions are met, then based on the matching relationship between transaction status and query message rules in the query rule base, a query message rule matching the transaction status is determined; Based on the query message rules, target information is extracted from the transaction status; The target information is filled into a preset query message template to generate a query message.
6. The system according to claim 1, characterized in that, The NLP reply message parsing module is specifically used for: Receive reply messages; The query message is segmented into words to obtain the target query message; The target reply message is input into the reply message parsing model, which outputs structured key information corresponding to the reply message; wherein, the reply message parsing model is the BERT model.
7. The system according to claim 1, characterized in that, The automatic response message generation module is specifically used for: Receive the structured key information and the transaction status; From a set of preset response message templates, a target response message template is matched based on the structured key information and the transaction status. The structured key information and the transaction status are mapped to the corresponding variable positions in the target response message template to generate a response message.
8. The system according to claim 1, characterized in that, The automatic business status advancement module is specifically used for: Receive the structured key information, the response message, and the transaction status; Based on the structured key information, the reply message, and the transaction status, generate the query response content; Determine whether the content of the query and reply meets the conditions for business status transition; If the conditions are met, the target business status will be generated based on the query response content.
9. The system according to claim 1, characterized in that, The automatic business status advancement module is also used for: The target business status is sent to the transaction status automatic identification module to update the transaction status based on the target business status.
10. An automated financial transaction query and response method, characterized in that, The method includes: Obtain the transaction status; wherein, the transaction status refers to the state of the transaction at the current stage; Generate a query message based on the transaction status; Obtain the reply message corresponding to the query message, and extract the structured key information from the reply message; wherein the structured key information is used to characterize the core information in the reply message; Based on the structured key information and the transaction status, a response message is generated; A target business status is generated based on the structured key information, the response message, and the transaction status; wherein, the target business status refers to the transaction status updated based on the actual completion status of the transaction after querying the transaction status.