Client event work order processing method and device, and program product

Abstract

The invention discloses a client event work order processing method and device and a program product, and relates to the field of artificial intelligence, and the method comprises the steps: extracting feature data of a plurality of data types from an event work order, and fusing the feature data of the plurality of data types through a cross-modal attention mechanism to obtain multi-modal data; inputting the multi-modal data and the target knowledge graph into a graph neural network, and generating candidate solutions of the event work order through the graph neural network; under the condition that the event work order does not involve the financial risk information, reply information is generated according to the candidate solutions, and the event work order is responded according to the reply information; and otherwise, sending the event work order and the candidate solution to the target object so as to process the event work order. Through the method and the device, the problem of high work order assignment error rate caused by manual assignment of the event work order and processing of the event work order when a financial institution processes the event work order proposed by a client in the related technology is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a method and device for processing an event order of a customer and a program product. BACKGROUND

[0002] In a traditional event order processing system, order dispatching relies on a rule engine and a simple machine learning model. The rule engine performs preliminary classification and dispatching of orders through preset rules, such as keyword matching. However, this dispatching method is prone to classification errors when faced with complex and ambiguous customer descriptions, affecting processing efficiency and customer satisfaction. In addition, when new business or policies are put online, the rule engine has a long update cycle and is difficult to quickly adapt to changes, especially in emergency scenarios where traditional rules are difficult to respond flexibly, thereby limiting the service range and response speed of the system.

[0003] In the related art, when a financial institution processes an event order raised by a customer, the event order is manually dispatched and processed, resulting in a high error rate in order dispatching. There is currently no effective solution to this problem. SUMMARY

[0004] The main purpose of the present application is to provide a method and device for processing an event order of a customer and a program product to solve the problem of a high error rate in order dispatching when a financial institution processes an event order raised by a customer in the related art.

[0005] To achieve the above purpose, according to one aspect of the present application, a method for processing an event order of a customer is provided, which comprises: in the case of detecting an event order submitted by a customer, extracting feature data of multiple data types from the event order, and fusing the feature data of multiple data types through a cross-modal attention mechanism to obtain multi-modal data; inputting the multi-modal data and a target knowledge graph into a graph neural network to generate a candidate solution for the event order through the graph neural network; wherein the target knowledge graph is generated according to historical event orders and a business knowledge base of a financial institution; in the case that the event order involves financial risk information, sending the event order and the candidate solution to a target object to process the event order; in the case that the event order does not involve the financial risk information, generating a reply information according to the candidate solution, and responding to the event order according to the reply information.

[0006] Further, the multimodal data and the target knowledge graph are input into a graph neural network, and candidate solutions for the event work order are generated through the graph neural network. This includes: encoding the multimodal data using a bidirectional transformer encoder to obtain a first embedding vector; encoding the nodes and edges in the target knowledge graph using a relational graph convolutional network to obtain a second embedding vector; concatenating the first embedding vector and the second embedding vector to obtain a fused vector; and inputting the fused vector into the graph neural network, which then generates candidate solutions for the event work order based on the solutions for the historical event work orders.

[0007] Furthermore, before inputting the multimodal data and the target knowledge graph into the graph neural network, the method further includes: collecting document information from the financial institution every preset time interval, and collecting document information related to the financial field from various external channels to obtain a document information set; updating the business knowledge base based on the document information set; performing entity recognition on the updated business knowledge base based on the financial field, and extracting triple information from the updated business knowledge base; and generating the target knowledge graph based on the entity recognition results and the triple information.

[0008] Further, generating the target knowledge graph based on the entity recognition results and the triple information includes: constructing nodes corresponding to the entities in the entity recognition results; wherein the nodes include at least: event ticket information, customer information, product information, financial rule information, and solution information; determining the association relationships between the entities based on the triple information, and constructing edges between the nodes based on the association relationships between the entities, wherein the association relationships include at least: the association relationship between the event ticket information and the product information, the association relationship between the solution information and the financial rule information, and the association relationship between the event ticket information; and determining the target knowledge graph based on the nodes and the edges.

[0009] Further, when the event ticket involves financial risk information, sending the event ticket and the candidate solution to the target object includes: determining whether the event ticket contains target keywords, wherein the target keywords are words extracted from the financial risk information, and the target keywords include: a first keyword and a second keyword, wherein the second keyword is obtained by representing the first keyword using a regular expression; the target keywords belong to at least one of the following categories: a first risk level and a second risk level, wherein the threat level of the first risk level is higher than that of the second risk level; when the event ticket contains the target keywords and the target keywords meet preset conditions, sending the event ticket and the candidate solution to the target object, wherein the preset conditions include at least one of the following: the number of target keywords belonging to the first risk level is greater than a first preset value, and the number of target keywords belonging to the second risk level is greater than a second preset value.

[0010] Furthermore, before generating candidate solutions for the event work order through the graph neural network, the method further includes: training a preset model using the historical event work orders, historical response information corresponding to the historical event work orders, and historical feedback information for the historical response information to obtain a first model; collecting the first event work orders received by the financial institution within a preset time period; calculating the reward score of the first model based on the reward calculation method of the reinforcement learning model, according to the first response information output by the first model for the first event work order and the first feedback information for the first response information; wherein, the reinforcement learning model includes at least: state information, action information, and the reward calculation method; if the reward score is less than a preset threshold, retraining the first model until the reward score is greater than or equal to the preset threshold to obtain the graph neural network.

[0011] Further, generating response information based on the candidate solutions includes: determining whether the event ticket involves multiple services in the financial institution; if the event ticket involves multiple services, generating multiple solutions corresponding to the multiple services using the graph neural network based on the multimodal data and the target knowledge graph, and generating the response information based on the multiple solutions; if the event ticket does not involve the multiple services, generating the response information based on the candidate solutions.

[0012] Furthermore, feature data of multiple data types are extracted from the event work order, and the feature data of the multiple data types are fused through a cross-modal attention mechanism to obtain multimodal data, including: extracting first feature data from image data in the event work order using optical character recognition technology, wherein the image data includes text information in multiple language types and handwritten text information; extracting second feature data from audio and video data in the event work order using speech-to-text technology, wherein the audio and video data is data after noise suppression and timestamp annotation; performing entity recognition on text data in the event work order using natural language processing technology to obtain third feature data, wherein the text data is data after text cleaning and regular expression filtering; and fusing the first feature data, the second feature data, and the third feature data through a cross-modal attention mechanism to obtain the multimodal data.

[0013] To achieve the above objectives, according to another aspect of this application, a processing apparatus for customer event work orders is provided. The apparatus includes: an extraction unit, configured to extract feature data of multiple data types from the event work order upon detecting a customer-submitted event work order, and to fuse the feature data of the multiple data types through a cross-modal attention mechanism to obtain multimodal data; a first generation unit, configured to input the multimodal data and a target knowledge graph into a graph neural network, and to generate candidate solutions for the event work order through the graph neural network; wherein the target knowledge graph is generated based on historical event work orders and a business knowledge base of a financial institution; a processing unit, configured to send the event work order and the candidate solutions to a target object for processing the event work order if the event work order involves financial risk information; and a response unit, configured to generate response information based on the candidate solutions and respond to the event work order based on the response information if the event work order does not involve the financial risk information.

[0014] Further, the first generation unit includes: a first encoding subunit, used to encode the multimodal data through a bidirectional transformer encoder to obtain a first embedding vector; a second encoding subunit, used to encode the nodes and edges in the target knowledge graph based on a relational graph convolutional network to obtain a second embedding vector; a concatenation subunit, used to concatenate the first embedding vector and the second embedding vector to obtain a fused vector; and a first generation subunit, used to input the fused vector into the graph neural network, and generate candidate solutions for the event work order based on the solutions of the historical event work orders through the graph neural network.

[0015] Furthermore, the device further includes: a first acquisition unit, configured to acquire document information of the financial institution and document information related to the financial field from various external channels every preset time interval before inputting the multimodal data and the target knowledge graph into the graph neural network, to obtain a document information set; an update unit, configured to update the business knowledge base according to the document information set; an identification unit, configured to perform entity identification on the updated business knowledge base based on the financial field, and extract triple information from the updated business knowledge base; and a second generation unit, configured to generate the target knowledge graph based on the entity identification result and the triple information.

[0016] Further, the second generation unit includes: a first construction subunit, configured to construct nodes corresponding to the entities based on the entities in the entity recognition results; wherein the nodes include at least: event work order information, customer information, product information, financial rule information, and solution information; a second construction subunit, configured to determine the association relationships between the entities based on the triple information, and construct edges between the nodes based on the association relationships between the entities, wherein the association relationships include at least: the association relationship between the event work order information and the product information, the association relationship between the solution information and the financial rule information, and the association relationship between the event work order information; and a determination subunit, configured to determine the target knowledge graph based on the nodes and the edges.

[0017] Further, the processing unit includes: a judgment subunit, used to judge whether the event ticket contains target keywords, wherein the target keywords are words extracted from the financial risk information, and the target keywords include: a first keyword and a second keyword, wherein the second keyword is obtained by representing the first keyword using a regular expression; the target keywords belong to at least one of the following categories: a first risk level and a second risk level, wherein the threat level of the first risk level is higher than that of the second risk level; and a sending subunit, used to send the event ticket and the candidate solution to the target object when the event ticket contains the target keywords and the target keywords meet preset conditions, wherein the preset conditions include at least one of the following: the number of target keywords belonging to the first risk level is greater than a first preset value, and the number of target keywords belonging to the second risk level is greater than a second preset value.

[0018] Furthermore, the device further includes: a first training unit, used to train a preset model using the historical event work orders, historical response information corresponding to the historical event work orders, and historical feedback information for the historical response information before generating candidate solutions for the event work orders through the graph neural network, to obtain a first model; a second acquisition unit, used to acquire the first event work orders received by the financial institution within a preset time period; a calculation unit, used to calculate the reward score of the first model based on the reward calculation method of the reinforcement learning model, according to the first response information output by the first model for the first event work order and the first feedback information for the first response information; wherein, the reinforcement learning model includes at least: state information, action information, and the reward calculation method; and a second training unit, used to retrain the first model when the reward score is less than a preset threshold, until the reward score is greater than or equal to the preset threshold, to obtain the graph neural network.

[0019] Furthermore, the response unit includes: a judgment subunit, used to judge whether the event ticket involves multiple services in the financial institution; a second generation subunit, used to generate multiple solutions corresponding to the multiple services through the graph neural network based on the multimodal data and the target knowledge graph when the event ticket involves the multiple services, and generate the response information based on the multiple solutions; and a third generation subunit, used to generate the response information based on the candidate solutions when the event ticket does not involve the multiple services.

[0020] Further, the extraction unit includes: a first extraction subunit, used to extract first feature data from the image data in the event work order using optical character recognition technology, wherein the image data includes: text information in multiple language types and handwritten text information; a second extraction subunit, used to extract second feature data from the audio and video data in the event work order using speech-to-text technology, wherein the audio and video data is data after noise suppression processing and timestamp annotation processing; a recognition subunit, used to perform entity recognition on the text data in the event work order using natural language processing technology to obtain third feature data, wherein the text data is data after text cleaning processing and regular expression filtering processing; and a fusion subunit, used to fuse the first feature data, the second feature data, and the third feature data using a cross-modal attention mechanism to obtain the multimodal data.

[0021] To achieve the above objectives, according to one aspect of this application, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the customer event ticket processing method described in any of the above-mentioned embodiments, and the computer program, when executed by a processor, implements the steps of the customer event ticket processing method described in various embodiments of this application.

[0022] To achieve the above objectives, according to one aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including stored computer instructions, wherein, when the computer instructions are executed by a processor, the method for processing a customer's event ticket as described above is implemented.

[0023] To achieve the above objectives, according to one aspect of this application, an electronic device is provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the customer event ticket processing method described in any of the above.

[0024] In this embodiment, upon detecting a customer-submitted event ticket, feature data of multiple data types is extracted from the event ticket, and the feature data of these multiple data types is fused using a cross-modal attention mechanism to obtain multimodal data. The multimodal data and a target knowledge graph are input into a graph neural network, which generates candidate solutions for the event ticket. The target knowledge graph is generated based on the financial institution's historical event tickets and business knowledge base. If the event ticket involves financial risk information, the event ticket and the candidate solutions are sent to a target object for processing. If the event ticket does not involve financial risk information, response information is generated based on the candidate solutions, and the event ticket is responded to based on the response information. This solves the technical problem of high error rates in event ticket assignment caused by manual assignment and processing of event tickets submitted by financial institutions.

[0025] By implementing a cross-modal attention mechanism to integrate features from multiple data types, and processing the fused multimodal data with the target knowledge graph through graph neural networks, event ticket information can be comprehensively analyzed. Based on historical data and business knowledge, candidate solutions are generated, achieving intelligent decision-making and recommendation, thus improving the efficiency and accuracy of ticket processing. Simultaneously, in scenarios involving financial risk information, sending event tickets and recommended solutions to a professional team ensures prudent risk assessment and compliant handling, strengthening risk management. In non-risk tickets, responses are automatically generated based on candidate solutions, resulting in rapid and effective customer service responses, automated processing, and guaranteed customer satisfaction, further enhancing the service efficiency of financial institutions. Attached Figure Description

[0026] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0027] Figure 1 This is a hardware structure block diagram of a computer terminal (or mobile device) for implementing a customer event work order processing method according to Embodiment 1 of this application;

[0028] Figure 2 This is a flowchart of an optional customer event ticket processing method provided according to Embodiment 1 of this application;

[0029] Figure 3 This is a flowchart illustrating an optional solution for generating event work orders provided in Embodiment 1 of this application;

[0030] Figure 4 This is a schematic diagram of the data processing flow according to the optional graph neural network model provided in Embodiment 1 of this application;

[0031] Figure 5 This is a flowchart illustrating an optional response event work order provided in Embodiment 1 of this application;

[0032] Figure 6 This is a schematic diagram of a customer event work order processing device according to Embodiment 2 of this application;

[0033] Figure 7 This is a schematic diagram of an electronic device for processing customer event work orders according to Embodiment 3 of this application. Detailed Implementation

[0034] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the processing method, apparatus, storage medium, and electronic device specified in this application can be used in the process of work order dispatch in the financial technology field, improving the accuracy and efficiency of work order dispatch. They can also be used in any field other than financial technology. The application fields of the processing method, apparatus, storage medium, and electronic device specified in this application are not limited.

[0036] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, collected data, used data, generated data, processed data, etc.) and the data (including but not limited to data used for analysis, stored data, displayed data, collected information, used information, generated information, processed information, etc.) are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, and necessary confidentiality measures have been taken. These measures do not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations, providing users with corresponding operation entry points for users to choose to agree to or refuse automated decision results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0037] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0038] OCR (Optical Character Recognition): Optical character recognition uses computer vision technology to automatically identify printed or handwritten text in images and convert it into an editable text format.

[0039] ASR (Automatic Speech Recognition): Automatic speech recognition is a cross-modal technology that converts human speech signals into text sequences in real time. Its core components include acoustic modeling and language modeling.

[0040] NLP (Natural Language Processing): Natural Language Processing is a branch of artificial intelligence that enables computers to understand, generate, and reason about human language, covering semantic understanding and text generation tasks.

[0041] RL (Reinforcement Learning): A machine learning paradigm that learns optimal policies through dynamic interactions between an agent and its environment, using reward signals to drive policy optimization.

[0042] GNN (Graph Neural Network): A deep learning model specifically designed to process graph-structured data, which aggregates the relationship information between nodes and edges through a message passing mechanism.

[0043] Example 1

[0044] According to an embodiment of this application, a method embodiment for processing customer event work orders is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0045] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 This is a hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for processing customer event work orders, according to Embodiment 1 of this application. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102 (client event ticket processing), ..., 102n) (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may include: a display, an input / output interface (I / O interface), a universal serial bus (US client event ticket processing) port (which may be included as one of the ports of the US bus for client event ticket processing), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0046] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0047] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the customer event ticket processing method in this embodiment of the application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned customer event ticket processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0048] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0049] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0050] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for handling customer event tickets is shown. Figure 2 This is a flowchart of an optional customer event ticket processing method provided according to Embodiment 1 of this application.

[0051] Step S201: Upon detecting an event ticket submitted by a customer, feature data of multiple data types is extracted from the event ticket, and feature data of multiple data types is fused through a cross-modal attention mechanism to obtain multimodal data.

[0052] In this embodiment 1, upon detecting an event ticket submitted by a customer, feature data of various data types, including text, images, and speech, can be extracted from the ticket using Optical Character Recognition (OCR), Automatic Speech Recognition (ASR), and Natural Language Processing (NLP) technologies. Subsequently, a cross-modal attention mechanism is employed to align and fuse these heterogeneous feature data, generating a unified multimodal data representation to support subsequent intelligent processing and decision-making, ensuring accurate classification and efficient dispatch of the ticket.

[0053] An incident ticket is a document used by financial institutions to record, track, and resolve various issues, inquiries, or service requests from customers. It typically contains basic information about the incident, such as an event description, time of occurrence, scope of impact, priority, and customer information. Incident tickets are used to manage the entire process from initial reporting to problem resolution, ensuring that each issue or request is properly recorded, categorized, processed, and closed. For example, a financial institution's agent receives a call from a customer reporting a transaction problem. The agent records the details of the problem, including the time of occurrence, the relevant business area, and the customer's information, generating an incident ticket. This ticket is then submitted to the financial institution's incident ticket processing system, allowing financial institution staff to resolve the incident (through the agent).

[0054] Step S202: Input the multimodal data and target knowledge graph into the graph neural network, and generate candidate solutions for event work orders through the graph neural network; wherein, the target knowledge graph is generated based on the historical event work orders and business knowledge base of financial institutions.

[0055] In this embodiment 1, multimodal data is a key feature of the event ticket, containing text, image, and voice information. The target knowledge graph is constructed based on the historical work ticket data and business knowledge base accumulated by financial institutions. It encodes the entities and relationships related to work ticket processing, as well as business processes and policy information.

[0056] In a Graph Neural Network (GNN), each node is embedded with its multimodal features and related entities and relationships in the graph. Through a message-passing mechanism, GNNs allow nodes in the graph to iteratively update based on information from neighboring nodes, thereby integrating the context and historical processing knowledge of work orders. The output graph neural network embedding vector not only reflects the multimodal information of the work order itself but also contains implicit knowledge related to historically similar work orders and business rules.

[0057] Ultimately, the embedding vectors generated by the GNN are used to recommend candidate solutions. This process leverages the rich relational information stored in the graph, enabling cross-modal contextual reasoning and providing more accurate work order processing strategies. In short, the graph neural network, by fusing multimodal data and historical business knowledge, outputs the optimal processing solution suggestion for the current event work order.

[0058] Step S203: If the event ticket involves financial risk information, send the event ticket and candidate solutions to the target object for processing.

[0059] In this embodiment 1, after preliminary analysis of the event ticket, if the ticket content is found to contain financial risk information, a specific forwarding process will be executed. Once it is confirmed that the ticket is associated with financial risk, such as suspected illegal fund transfers or high-value abnormal transactions, the event ticket will not be processed automatically. Instead, the event ticket and its recommended candidate solutions will be sent to a preset target, namely the department or personnel specifically responsible for risk management. After receiving the ticket, the target will conduct a detailed review of the ticket content and recommended solutions based on their professional knowledge and experience.

[0060] In one optional embodiment, after completing the review, the target entity will take appropriate measures to address financial risks, which may include, but are not limited to, communicating with clients to confirm details, initiating internal investigation processes, or reporting to regulatory agencies. This process ensures the timely identification and professional handling of financial risks, effectively safeguarding the compliance of financial institutions and the security of client assets. Simultaneously, human intervention improves the accuracy of risk identification and the rigor of handling. Through the sensitivity detection of risk keywords (i.e., the aforementioned financial risk information) in the above steps, the aim is to respond prudently and appropriately to potential financial risks through the intervention of professionals, avoiding the uncertainties or risk oversights that may arise from automated processing.

[0061] Step S204: If the event work order does not involve financial risk information, generate response information based on the candidate solutions, and respond to the event work order based on the response information.

[0062] In this embodiment 1, if it is determined that the event ticket does not involve financial risk information, a response message is automatically constructed and sent to the customer based on the candidate solutions generated by the GNN. This process begins with the fusion analysis of the multimodal data and knowledge graph of the ticket, ensuring that the generation of the solution comprehensively considers the text description, image and voice information of the ticket, as well as business knowledge and historical data related to ticket processing.

[0063] For example, using predefined rules or models, the most suitable response is selected from candidate solutions. This response directly addresses the issue raised by the customer, providing a solution or explanation. The response is then sent to the customer and simultaneously recorded in the ticket system as part of the historical record and subsequent processing. This automated response mechanism aims to resolve customer issues efficiently and accurately, reducing human intervention and improving service quality and response speed. If the customer accepts the automated response solution, the system will follow the instructions in the response, such as calling an API to perform specific operations, providing self-service guidance, or updating the ticket status, to execute the subsequent ticket processing flow. The entire process emphasizes rapid response and resolution of non-risk tickets while retaining a channel for human intervention to handle complex or exceptional situations, ensuring comprehensive service and customer satisfaction.

[0064] Optionally, in the customer event work order processing method provided in Embodiment 1 of this application, multimodal data and target knowledge graph are input into a graph neural network, and candidate solutions for the event work order are generated through the graph neural network. This includes: encoding the multimodal data through a bidirectional transformer encoder to obtain a first embedding vector; encoding the nodes and edges in the target knowledge graph based on a relational graph convolutional network to obtain a second embedding vector; concatenating the first embedding vector and the second embedding vector to obtain a fused vector; and inputting the fused vector into a graph neural network to generate candidate solutions for the event work order based on the solutions of historical event work orders.

[0065] In this embodiment 1, to generate candidate solutions suitable for customer-submitted event tickets, preprocessed multimodal data can be received. This data includes customer text, image, voice, and video information. To transform this unstructured multimodal information into a form that the model can understand, a bidirectional transformer encoder is used to encode the dataset. This process converts information from different modes into a series of numerical vectors, each representing a different feature of the original data, thereby generating a first embedding vector that encapsulates the comprehensive information of the customer event ticket.

[0066] In one optional embodiment, multimodal data may include the following data represented in any form: event unique identifier (e.g., date, business type, random code), association identifier (e.g., unique identifier of parent event ticket, associated change ticket, unique identifier of issue ticket), reporting channel (e.g., monitoring alarms, user feedback, automated inspection), ticket category (e.g., cloud computing, network, load balancing, DNS failure, certificate expiration, API error), user impact, business system impact, affected business services (e.g., unique identifier of service, SLA level), business metrics (e.g., ... The data includes percentage decrease in transaction volume, number of affected customers, technical diagnostic data (e.g., host IP information, container information, cloud resource information), application layer information (e.g., error stack hash, TraceID, API endpoint), network information (e.g., packet loss rate, DNS resolution records, firewall rule changes), log information (e.g., key error log summaries, including timestamps and error codes), status (e.g., new status, assigned status, processing status, pending verification status, closed status), status change information (e.g., operator information, timestamp information), and action records (e.g., automatically triggered disaster recovery switchover).

[0067] Then, a relational graph convolutional network is used to encode the nodes and edges in the graph to capture the connections and complex relationships between entities. This encoding step generates a second embedding vector, which effectively represents the information structure and dynamic changes in the knowledge graph. The target knowledge graph is a complex network containing historical event work order solutions, business rules, and their interrelationships within a financial institution.

[0068] Secondly, to achieve the synergistic effect of multimodal data and knowledge graph information, the first and second embedding vectors can be concatenated. This fusion process ensures information complementarity between the two vectors through algorithms, generating a fused vector that simultaneously contains the details of the customer's work order and an overview of relevant business knowledge.

[0069] Finally, the fused vector is input into a graph neural network. This network uses previously trained weights and solutions from historical event work orders to predict the processing strategy for the current event work order. In this process, the graph neural network can utilize cross-modal fused information and historical data from the graph to perform deep reasoning and learning, ultimately generating candidate solutions for the current work order and providing intelligent decision-making support for subsequent work order processing.

[0070] Through the above steps, the technical effect of automated and intelligent work order processing is achieved. This not only improves the accuracy of work order classification, but also enhances the relevance and effectiveness of solution generation by integrating historical data analysis with current work order information, thereby significantly improving the processing efficiency of event work orders and the quality of customer service.

[0071] Optionally, in the customer event work order processing method provided in Embodiment 1 of this application, before inputting the multimodal data and target knowledge graph into the graph neural network, the method further includes: collecting document information from financial institutions every preset time interval, and collecting document information related to the financial field from various external channels to obtain a document information set; updating the business knowledge base based on the document information set; performing entity recognition on the updated business knowledge base based on the financial field, and extracting triple information from the updated business knowledge base; and generating a target knowledge graph based on the entity recognition results and triple information.

[0072] In this embodiment 1, to construct and update the target knowledge graph in real time, a periodic data collection program can be initiated. This program is set to automatically retrieve the latest document information from the financial institution's internal network every preset time interval. Simultaneously, it can also collect document information related to the financial sector from a range of external sources, such as regulatory agency websites, industry reports, and financial news, forming a document information set. This set not only covers the latest policy changes in the financial sector but also includes the impact of external environmental factors such as financial market trends and regulatory updates, providing comprehensive data support for subsequent knowledge base updates.

[0073] Then, based on the collected document information, the business knowledge base is updated. This process involves parsing, summarizing, and storing new information to ensure that the knowledge base content is up-to-date and reflects the latest practices and standards in the financial field. The updated business knowledge base becomes an important basis for subsequent knowledge graph construction.

[0074] Secondly, entity recognition technology in the financial field is used to conduct in-depth analysis of the updated business knowledge base. Through precise entity recognition, key concepts are accurately captured, namely the aforementioned entities such as financial products, regulatory policies, and business processes, laying a solid foundation for the construction of the knowledge graph. Next, triplet information is extracted from the updated business knowledge base. These triples describe the relationships between entities, such as the association between a financial service and regulatory policies, providing the necessary structured relationships for enriching and refining the knowledge graph.

[0075] Finally, based on the entity recognition results and the extracted triple information, a target knowledge graph is constructed. This graph, with entities as nodes and relationships as edges, intuitively displays the knowledge network in the financial field. It not only encompasses basic business information but also reflects the complex logical connections between entities, providing a data foundation for intelligent work order processing.

[0076] Through the above steps, the technical effects of dynamically updating and maintaining the business knowledge base and constructing a refined knowledge graph are achieved, ensuring that the decision-making basis of the intelligent processing system is always up-to-date and most relevant, thereby enhancing the accuracy and efficiency of work order processing. At the same time, the flexibility and adaptability of the knowledge graph are also improved, enabling it to better cope with the continuous changes in the financial field.

[0077] Optionally, in the customer event ticket processing method provided in Embodiment 1 of this application, generating a target knowledge graph based on entity recognition results and triple information includes: constructing nodes corresponding to entities based on entities in the entity recognition results; wherein, the nodes include at least: event ticket information, customer information, product information, financial rule information, and solution information; determining the association relationship between entities based on triple information, and constructing edges between nodes based on the association relationship between entities, wherein the association relationship includes at least: the association relationship between event ticket information and product information, the association relationship between solution information and financial rule information, and the association relationship between event ticket information; and determining the target knowledge graph based on nodes and edges.

[0078] In this embodiment 1, in order to construct a comprehensive and accurate knowledge graph to support the decision-making of the intelligent work order processing system, the results of entity recognition are first analyzed to identify the basic elements constituting the knowledge graph, namely various entities, including but not limited to specific event work order details, customer background information, financial product descriptions, current financial rules, and past solution examples. These entities are regarded as nodes in the graph, and each node carries a specific type of information.

[0079] Then, based on the extracted triple information, the relationships between entities are determined. This triple information details the logical connections between entities, such as the relationships between event tickets, between event tickets and financial products, or between solutions and financial rules. Based on these relationships, edges are established between nodes, forming the connection structure in the graph. This process not only covers the direct association between event tickets and financial products but also the indirect association between solutions and financial rules, as well as the mutual references between event tickets, constructing a graph model that reflects the complex relationships between entities.

[0080] Secondly, through comprehensive analysis of nodes and edges, the structure of the knowledge graph is further refined and optimized. This stage includes identifying redundant information, eliminating isolated nodes, and supplementing missing relationships to ensure the integrity and logical coherence of the knowledge graph, providing accurate and comprehensive support for subsequent work order processing and solution generation.

[0081] Finally, by integrating all nodes and edges, the final form of the target knowledge graph is determined. This knowledge graph not only reflects various entities and their relationships in the current business environment, but also can self-update based on new entity recognition and triple information, maintaining its timeliness and applicability.

[0082] Through the above steps, a dynamic, detailed, and scalable knowledge graph is created. This graph effectively supports intelligent work order processing systems in performing deep logical reasoning and intelligent decision-making based on multimodal data analysis, significantly improving the accuracy and efficiency of work order processing. It also enhances the flexibility in resolving event work orders and the adaptability of financial institutions to emerging business scenarios.

[0083] Optionally, in the customer event ticket processing method provided in Embodiment 1 of this application, when the event ticket involves financial risk information, sending the event ticket and candidate solutions to the target object includes: determining whether the event ticket contains target keywords, wherein the target keywords are words extracted from financial risk information, and the target keywords include: a first keyword and a second keyword, wherein the second keyword is obtained by representing the first keyword using a regular expression; the target keywords belong to at least one of the following categories: a first risk level and a second risk level, wherein the threat level of the first risk level is higher than that of the second risk level; when the event ticket contains target keywords and the target keywords meet preset conditions, sending the event ticket and candidate solutions to the target object, wherein the preset conditions include at least one of the following: the number of target keywords belonging to the first risk level is greater than a first preset value, and the number of target keywords belonging to the second risk level is greater than a second preset value.

[0084] In this embodiment 1, for intelligent assignment of event work orders, event work orders can be assigned based on risk keywords involving financial risk information (i.e., the aforementioned target keywords) and risk level assessment. Specifically, the content of each event work order is analyzed in depth to identify whether specific financial risk information exists within it. This information is transformed into target keywords, including directly extracted first keywords and second keywords represented by variations of the first keywords using regular expressions. This step can capture various forms of risk information expression, improving the comprehensiveness of risk detection.

[0085] Then, the extracted target keywords are classified according to their risk level, categorized into a predefined first risk level or second risk level, where the threat level of the first risk level is considered higher than that of the second risk level. It should be noted that the risk level of the target keywords may also include a third risk level or other risk levels, which can be adjusted according to the actual situation; no specific limitation is made in this embodiment 1. This classification process, based on the potential harm of the keywords, provides an important reference for subsequent decision-making regarding risk levels.

[0086] Secondly, based on preset conditions, the target keywords and their quantity contained in the event work order are assessed. If either of the following conditions is met: the number of first keywords belonging to the high threat level (i.e., the first risk level mentioned above) exceeds the first preset value, or the number of second keywords belonging to the low threat level (i.e., the second risk level mentioned above) exceeds the second preset value, the work order will be deemed to have significant financial risk.

[0087] Finally, if a work order meets the aforementioned risk conditions, it will be automatically sent along with recommended candidate solutions to the target entity within the financial institution—the risk control team or senior customer service representative with the appropriate authority and capabilities—for further evaluation and processing. This mechanism ensures that in high-risk situations, work orders can be quickly transferred to the most suitable professionals, preventing potential financial risks from escalating.

[0088] Through the above steps, the technical capabilities of accurately identifying and rapidly responding to financial risk events are achieved, effectively improving the timeliness and effectiveness of risk management. Simultaneously, the dual screening process of keyword detection and risk level assessment allows for more rational allocation of internal resources, ensuring that high-risk work orders are prioritized, while low-risk or routine work orders can be efficiently resolved through automated processes, thus improving both the quality of customer service and the efficiency of work order processing for financial institutions.

[0089] Optionally, in the customer event ticket processing method provided in Embodiment 1 of this application, before generating candidate solutions for the event ticket through a graph neural network, the method further includes: training a preset model using historical event tickets, historical response information corresponding to the historical event tickets, and historical feedback information for the historical response information to obtain a first model; collecting first event tickets received by financial institutions within a preset time period; calculating the reward score of the first model based on the reward calculation method of the reinforcement learning model, according to the first response information output by the first model for the first event ticket and the first feedback information for the first response information; wherein, the reinforcement learning model includes at least: state information, action information, and reward calculation method; if the reward score is less than a preset threshold, retraining the first model until the reward score is greater than or equal to the preset threshold to obtain a graph neural network.

[0090] In this first embodiment, to continuously improve customer satisfaction and work order processing efficiency, the event work order processing strategy of the model can be optimized through reinforcement learning. Specifically, a deep learning framework is used to analyze and process the financial institution's historical event work orders, the corresponding historical response information, and customer feedback data on these responses. Through this process, effective work order processing patterns and customer preferences can be learned, thereby training a preliminary model, namely the first model mentioned above.

[0091] Then, within a preset time period (e.g., 1 day, 3 days, or 30 days from the current time), new event tickets received by financial institutions are monitored and collected as input data for real-time testing and optimization. These tickets at this stage will be used to evaluate the current performance of the first model and to verify whether the model can adapt to the latest changes in market and customer needs.

[0092] Secondly, a reward calculation mechanism based on a reinforcement learning model is used to quantitatively evaluate the performance of the first model in handling new event tickets. This calculation is based on the response information generated by the model and customer feedback on these responses, which may include indicators such as resolution time and customer satisfaction. The reward score directly reflects the overall effectiveness of the model in handling ticket issues, providing clear guidance for subsequent model optimization.

[0093] Finally, if the reward score of the first model falls below a preset threshold, a retraining process is automatically triggered. Through continuous learning and adjustment, the model's reward score is adjusted until it reaches or exceeds the preset threshold. This closed-loop optimization mechanism ensures that the model can consistently provide high-quality work order processing and responses, maintaining stable and efficient responsiveness even in complex and ever-changing financial scenarios.

[0094] Through the above steps, the technical effect of dynamically optimizing model strategies and improving work order processing satisfaction and efficiency is achieved through reinforcement learning. This mechanism not only enhances the model's intelligent decision-making capabilities but also continuously improves its performance through real-time feedback and learning, effectively addressing the challenges of rapid changes and diverse customer needs in the financial sector.

[0095] Optionally, in the customer event ticket processing method provided in Embodiment 1 of this application, generating response information based on candidate solutions includes: determining whether the event ticket involves multiple services in a financial institution; if the event ticket involves multiple services, generating multiple solutions corresponding to multiple services through a graph neural network based on multimodal data and a target knowledge graph, and generating response information based on the multiple solutions; if the event ticket does not involve multiple services, generating response information based on candidate solutions.

[0096] In this embodiment 1, to effectively improve the accuracy of work order processing and customer satisfaction, multiple solutions corresponding to multiple services can be flexibly generated when faced with event work orders corresponding to complex events. Specifically, the received event work orders are deeply analyzed to determine whether they simultaneously involve multiple service areas (or business areas) of the financial institution. For example, this determination can be based on comparing the content of the work order with information in the target knowledge graph, aiming to identify the complexity and service scope of the work order requirements.

[0097] Then, after confirming that the event ticket indeed involves multiple services, a graph neural network is activated. Based on multimodal data and the correlation information of each service node in the target knowledge graph, a solution specific to each service is generated. The graph neural network can identify and understand the inherent relationships between services, thereby providing an integrated, multifaceted solution to address the complex needs of the ticket. Subsequently, these solutions are integrated to form the final response information, providing the customer with a more comprehensive solution for reference.

[0098] Finally, when an incident ticket involves only a single service, a response is generated directly based on pre-selected candidate solutions. This approach simplifies the decision-making process, speeds up response times, and is particularly suitable for the immediate processing of routine and simple tickets.

[0099] Through the above steps, intelligent identification and customized solution generation for multi-service related event work orders are achieved, while maintaining the ability to process single-service work orders efficiently and quickly. This mechanism not only improves the accuracy and speed of work order processing, but also enhances overall customer service satisfaction by comprehensively considering the needs of multiple services, thereby improving the efficiency and ability of financial institutions to solve problems in complex financial scenarios.

[0100] Optionally, in the customer event ticket processing method provided in Embodiment 1 of this application, feature data of multiple data types are extracted from the event ticket, and the feature data of multiple data types are fused through a cross-modal attention mechanism to obtain multimodal data. This includes: extracting first feature data from image data in the event ticket using optical character recognition technology, wherein the image data includes text information in multiple languages ​​and handwritten text information; extracting second feature data from audio and video data in the event ticket using speech-to-text technology, wherein the audio and video data is data after noise suppression and timestamp annotation; performing entity recognition on text data in the event ticket using natural language processing technology to obtain third feature data, wherein the text data is data after text cleaning and regular expression filtering; and fusing the first feature data, second feature data, and third feature data through a cross-modal attention mechanism to obtain multimodal data.

[0101] In this embodiment 1, to extract key information from the unstructured data of event work orders for subsequent processing, optical character recognition (OCR) technology is first introduced. This technology is applied to the image data contained in the work orders to parse the text content. Whether it is printed or handwritten text in multiple languages, it can be accurately identified and converted into processable feature data, i.e., the first feature data. This conversion process greatly broadens the types of data that can be analyzed and processed, no longer limited to plain text information.

[0102] Then, speech-to-text technology was used to process the audio and video data in the work order. This technology, based on noise suppression and timestamp annotation, converts the speech information into text format, thereby obtaining the second feature data. This conversion ensures that key information in the audio can be effectively captured by the system, while the timestamp annotation provides a time clue for subsequent data processing and analysis.

[0103] Secondly, natural language processing (NLP) technology is used to perform in-depth analysis of the text data in the work orders, including entity recognition, to identify and extract key entities such as customer information, product type, and business requests, forming the third feature data. The entity recognition process also includes text cleaning and regular expression filtering to remove irrelevant information and retain high-value entity data, ensuring the accuracy of subsequent decisions.

[0104] Finally, a cross-modal attention mechanism is used to fuse the feature data from the three sources mentioned above. This mechanism assigns weights to different feature data based on the semantic relevance of each modality, comprehensively generating multimodal data that fully reflects various aspects of the work order. This fusion process not only integrates text, image, and audio data but also ensures the priority of key information by dynamically adjusting weights.

[0105] Through the above steps, key features can be extracted and integrated from multi-source, unstructured data, enabling a deep understanding and intelligent parsing of event work order content. This provides a solid data foundation for subsequent work order classification, solution generation, and compliance risk assessment, significantly improving the accuracy and efficiency of the intelligent work order processing system.

[0106] Optionally, in this embodiment 1, Figure 3 This is a flowchart illustrating an optional solution for generating event work orders provided in Embodiment 1 of this application. First, the multimodal data in the event work order, including images, audio, video, and text, is collected uniformly by the multimodal data acquisition module, and a type detector performs preliminary format and type determination to ensure the accuracy and efficiency of subsequent processing.

[0107] Then, different types of data in the multimodal data are directed to their respective specialized processing engines: the OCR engine is responsible for extracting text information from images and converting it into readable text, supporting multilingual and handwriting recognition; the ASR engine processes speech data, transcribing it into text, and applying noise suppression and timestamp annotation techniques to ensure the accuracy and integrity of the speech content; the NLP pipeline cleans the text data, performs entity recognition and regular expression filtering, improving the structuring level and parsing depth of the text information.

[0108] Secondly, the different modal feature data processed by the aforementioned engine are fed into a cross-modal attention mechanism spatiotemporal aligner for fusion processing. This mechanism identifies the inherent relationships between the data from different modalities, aligns the temporal and semantic features of the data, and generates multimodal feature representations, providing comprehensive work order context information for subsequent decision-making and solution generation.

[0109] Finally, the fused multimodal feature data is integrated with a knowledge graph for fusion analysis. Through this fusion process, solutions for work orders can be generated based on the multimodal features of the work orders and information from the knowledge base. The confidence level of the model is also considered to ensure the reliability and applicability of the proposed solutions.

[0110] Optionally, in this embodiment 1, Figure 4 This is a schematic diagram of the data processing flow according to the optional graph neural network model provided in Embodiment 1 of this application. First, multi-source data streams, including internal documents, regulatory documents, and work orders, flow to the real-time extracted data, the dynamic graph engine, and work order knowledge, respectively. This step ensures that the system can collect comprehensive work order-related data from different channels.

[0111] Then, the key information extracted in real time from the multi-source data stream is absorbed by the dynamic graph engine to build and update the nodes and relationships in the graph. At the same time, the dynamic graph engine can also update the key information extracted in real time based on the existing graph information, guiding it to capture data more accurately. This enhances the system's ability to dynamically adjust the work order processing strategy.

[0112] Secondly, the graph information generated by the dynamic graph engine is passed to the GNN inference engine. Through the cross-modal attention mechanism, the GNN inference engine can comprehensively analyze the multimodal data of the work order, generate the most relevant historical solutions to the work order, and use manually labeled feedback information for reinforcement learning through the verification feedback processing operation to continuously optimize the model's decision-making strategy and ensure the accuracy and timeliness of the solution.

[0113] Finally, the spatiotemporal features encoded from the work order knowledge are utilized by the solution generator to generate multi-path candidate solutions for the work order. These candidate solutions are then tested by the execution effect tracking module, and the implementation effect of the solutions is judged based on key performance indicators. This feedback information is also cyclically fed into the verification feedback processing operation for subsequent training and optimization of the model, thereby further improving the system's intelligent decision-making level and customer satisfaction in the process of handling event work orders.

[0114] Optionally, in this embodiment 1, Figure 5 This is a flowchart illustrating an optional response event work order provided in Embodiment 1 of this application. First, the financial institution's work order access system performs protocol conversion and flow control through a gateway to ensure that all incoming work orders can be correctly parsed and processed in an orderly manner. Then, the intelligent hierarchical engine assesses the complexity of the work order and determines the processing path based on the assessment results. Simultaneously, the RL strategy decision tree is activated, and the Q-Learning algorithm is used to pre-plan the response strategy, providing intelligent guidance for subsequent work order processing.

[0115] Then, based on the complexity of the work order, the system routes the work order to different response channels. The first-level response channel is suitable for simple work orders, providing a rapid response through the solution output by a graph neural network, with a response time threshold set below 500ms to ensure the timeliness of basic services. The second-level response channel handles more complex work orders, using knowledge graph-assisted decision-making to generate multiple candidate solutions and present them to the customer. The final action is determined based on the customer's selection feedback, improving the intelligence and humanization of the service.

[0116] Finally, in the three-tiered response channel, complex work orders involving high risks or exceeding the scope of automated processing are handled directly. These orders are transferred to human agents, who are also equipped with AI-assisted tools to ensure timely risk identification and mitigation. Human agents then take appropriate measures based on the risk assessment and handling suggestions provided by the AI, achieving a balance between risk management and customer service.

[0117] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0118] In summary, the customer event ticket processing method provided in this application, upon detecting a customer-submitted event ticket, extracts feature data of multiple data types from the event ticket and fuses these feature data through a cross-modal attention mechanism to obtain multimodal data. The multimodal data and a target knowledge graph are then input into a graph neural network to generate candidate solutions for the event ticket. The target knowledge graph is generated based on the financial institution's historical event tickets and business knowledge base. If the event ticket involves financial risk information, the event ticket and candidate solutions are sent to the target object for processing. If the event ticket does not involve financial risk information, response information is generated based on the candidate solutions, and the event ticket is responded to based on the response information. This method solves the problem in related technologies where financial institutions manually assign and process customer event tickets, resulting in a high error rate in ticket assignment.

[0119] By implementing a cross-modal attention mechanism to integrate features from multiple data types, and processing the fused multimodal data with the target knowledge graph through graph neural networks, event ticket information can be comprehensively analyzed. Based on historical data and business knowledge, candidate solutions are generated, achieving intelligent decision-making and recommendation, thus improving the efficiency and accuracy of ticket processing. Simultaneously, in scenarios involving financial risk information, sending event tickets and recommended solutions to a professional team ensures prudent risk assessment and compliant handling, strengthening risk management. In non-risk tickets, responses are automatically generated based on candidate solutions, resulting in rapid and effective customer service responses, automated processing, and guaranteed customer satisfaction, further enhancing the service efficiency of financial institutions.

[0120] Example 2

[0121] This application also provides a device for processing customer event work orders. It should be noted that this device can be used to execute the method for processing customer event work orders provided in this application. The following describes the device for processing customer event work orders provided in this application.

[0122] According to an embodiment of this application, an apparatus for implementing the above-described method for processing customer event work orders is also provided. Figure 6 This is a schematic diagram of a customer event ticket processing device according to Embodiment 2 of this application. Figure 6 As shown, the device includes:

[0123] Specifically, the extraction unit 601 is used to extract feature data of multiple data types from the event work order when a customer-submitted event work order is detected, and to fuse the feature data of multiple data types through a cross-modal attention mechanism to obtain multimodal data.

[0124] The first generation unit 602 is used to input multimodal data and target knowledge graph into a graph neural network and generate candidate solutions for event work orders through the graph neural network; wherein, the target knowledge graph is generated based on the historical event work orders and business knowledge base of financial institutions.

[0125] The processing unit 603 is used to send the event ticket and candidate solutions to the target object for processing when the event ticket involves financial risk information.

[0126] The response unit 604 is used to generate response information based on candidate solutions and respond to the event work order based on the response information when the event work order does not involve financial risk information.

[0127] The customer event work order processing device provided in this application embodiment, through extraction unit 601, extracts feature data of multiple data types from the event work order when it detects the event work order submitted by the customer, and fuses the feature data of multiple data types through a cross-modal attention mechanism to obtain multimodal data; first generation unit 602 inputs the multimodal data and target knowledge graph into graph neural network, and generates candidate solutions for the event work order through graph neural network; wherein, the target knowledge graph is generated based on the historical event work orders and business knowledge base of financial institutions; processing unit 603, when the event work order involves financial risk information, sends the event work order and candidate solutions to the target object for processing the event work order; response unit 604, when the event work order does not involve financial risk information, generates reply information based on the candidate solutions, and responds to the event work order based on the reply information, thus solving the problem in related technologies where financial institutions manually assign and process event work orders submitted by customers, resulting in a high error rate in work order assignment.

[0128] By implementing a cross-modal attention mechanism to integrate features from multiple data types, and processing the fused multimodal data with the target knowledge graph through graph neural networks, event ticket information can be comprehensively analyzed. Based on historical data and business knowledge, candidate solutions are generated, achieving intelligent decision-making and recommendation, thus improving the efficiency and accuracy of ticket processing. Simultaneously, in scenarios involving financial risk information, sending event tickets and recommended solutions to a professional team ensures prudent risk assessment and compliant handling, strengthening risk management. In non-risk tickets, responses are automatically generated based on candidate solutions, resulting in rapid and effective customer service responses, automated processing, and guaranteed customer satisfaction, further enhancing the service efficiency of financial institutions.

[0129] Optionally, in the customer event work order processing device provided in Embodiment 2 of this application, the first generation unit 602 includes: a first encoding subunit, used to encode multimodal data through a bidirectional converter encoder to obtain a first embedding vector; a second encoding subunit, used to encode nodes and edges in the target knowledge graph based on a relational graph convolutional network to obtain a second embedding vector; a concatenation subunit, used to concatenate the first embedding vector and the second embedding vector to obtain a fused vector; and a first generation subunit, used to input the fused vector into a graph neural network, and generate candidate solutions for event work orders based on the solutions of historical event work orders through the graph neural network.

[0130] Optionally, in the customer event work order processing device provided in Embodiment 2 of this application, the device further includes: a first acquisition unit, used to acquire document information of financial institutions and document information related to the financial field from various external channels every preset time interval before inputting multimodal data and target knowledge graph into the graph neural network, to obtain a document information set; an update unit, used to update the business knowledge base according to the document information set; an identification unit, used to perform entity identification on the updated business knowledge base based on the financial field, and extract triple information from the updated business knowledge base; and a second generation unit, used to generate a target knowledge graph based on the entity identification result and triple information.

[0131] Optionally, in the customer event ticket processing device provided in Embodiment 2 of this application, the second generation unit mentioned above includes: a first construction subunit, used to construct nodes corresponding to entities based on entities in the entity recognition results; wherein, the nodes include at least: event ticket information, customer information, product information, financial rule information, and solution information; a second construction subunit, used to determine the association relationship between entities based on triple information, and construct edges between nodes based on the association relationship between entities, wherein the association relationship includes at least: the association relationship between event ticket information and product information, the association relationship between solution information and financial rule information, and the association relationship between event ticket information; and a determination subunit, used to determine the target knowledge graph based on nodes and edges.

[0132] Optionally, in the customer event ticket processing device provided in Embodiment 2 of this application, the processing unit 603 includes: a judgment subunit, used to judge whether the event ticket contains target keywords, wherein the target keywords are words extracted from financial risk information, and the target keywords include: a first keyword and a second keyword, wherein the second keyword is obtained by representing the first keyword using a regular expression; the target keywords belong to at least one of the following categories: a first risk level and a second risk level, wherein the threat level of the first risk level is higher than that of the second risk level; and a sending subunit, used to send the event ticket and candidate solutions to the target object when the event ticket contains target keywords and the target keywords meet preset conditions, wherein the preset conditions include at least one of the following: the number of target keywords belonging to the first risk level is greater than a first preset value, and the number of target keywords belonging to the second risk level is greater than a second preset value.

[0133] Optionally, in the customer event work order processing device provided in Embodiment 2 of this application, the device further includes: a first training unit, used to train a preset model using historical event work orders, historical response information corresponding to historical event work orders, and historical feedback information for historical response information before generating candidate solutions for event work orders through a graph neural network, to obtain a first model; a second acquisition unit, used to acquire first event work orders received by financial institutions within a preset time period; a calculation unit, used to calculate the reward score of the first model based on the reward calculation method of the reinforcement learning model, according to the first response information output by the first model for the first event work order and the first feedback information for the first response information; wherein, the reinforcement learning model includes at least: state information, action information, and reward calculation method; and a second training unit, used to retrain the first model when the reward score is less than a preset threshold, until the reward score is greater than or equal to the preset threshold, to obtain a graph neural network.

[0134] Optionally, in the customer event ticket processing device provided in Embodiment 2 of this application, the response unit 604 includes: a judgment subunit, used to judge whether the event ticket involves multiple services in a financial institution; a second generation subunit, used to generate multiple solutions corresponding to multiple services through a graph neural network based on multimodal data and a target knowledge graph when the event ticket involves multiple services, and generate response information based on the multiple solutions; and a third generation subunit, used to generate response information based on candidate solutions when the event ticket does not involve multiple services.

[0135] Optionally, in the customer event work order processing device provided in Embodiment 2 of this application, the extraction unit 601 includes: a first extraction subunit, used to extract first feature data from image data in the event work order using optical character recognition technology, wherein the image data includes: text information in multiple language types and handwritten text information; a second extraction subunit, used to extract second feature data from audio and video data in the event work order using speech-to-text technology, wherein the audio and video data is data after noise suppression processing and timestamp annotation processing; a recognition subunit, used to perform entity recognition on text data in the event work order using natural language processing technology to obtain third feature data, wherein the text data is data after text cleaning processing and regular expression filtering processing; and a fusion subunit, used to fuse the first feature data, second feature data and third feature data through a cross-modal attention mechanism to obtain multimodal data.

[0136] It should be noted that the extraction unit 601, the first generation unit 602, the processing unit 603, and the response unit 604 mentioned above correspond to steps S201 to S204 in Embodiment 1. The two modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102 (processing of client event work orders), ..., 102n). The above modules can also be part of the device and run in the computer terminal 10 provided in Embodiment 1.

[0137] Example 3

[0138] Embodiments of this application may provide an electronic device. Figure 7 This is a schematic diagram of an electronic device for processing customer event work orders according to Embodiment 3 of this application. Figure 7 As shown, the electronic device may include: one or more ( Figure 7Only one of the following is shown: processor 702, memory 704, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module, and display.

[0139] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0140] The processor can invoke information and applications stored in memory via a transmission device to perform the following steps: Upon detecting an event ticket submitted by a customer, extract feature data of multiple data types from the event ticket, and fuse the feature data of the multiple data types through a cross-modal attention mechanism to obtain multimodal data; input the multimodal data and a target knowledge graph into a graph neural network, and generate candidate solutions for the event ticket through the graph neural network; wherein, the target knowledge graph is generated based on the historical event tickets and business knowledge base of the financial institution; if the event ticket involves financial risk information, send the event ticket and the candidate solutions to the target object for processing; if the event ticket does not involve the financial risk information, generate response information based on the candidate solutions, and respond to the event ticket based on the response information.

[0141] The processor can access information and applications stored in memory via a transmission device to perform the following steps: inputting multimodal data and a target knowledge graph into a graph neural network, and generating candidate solutions for event work orders through the graph neural network, including: encoding the multimodal data through a bidirectional transformer encoder to obtain a first embedding vector; encoding nodes and edges in the target knowledge graph based on a relational graph convolutional network to obtain a second embedding vector; concatenating the first and second embedding vectors to obtain a fused vector; and inputting the fused vector into the graph neural network, and generating candidate solutions for event work orders based on solutions for historical event work orders through the graph neural network.

[0142] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: Before inputting the multimodal data and target knowledge graph into the graph neural network, the above method further includes: collecting document information from financial institutions every preset time interval, as well as collecting document information related to the financial field from various external channels, to obtain a document information set; updating the business knowledge base based on the document information set; performing entity recognition on the updated business knowledge base based on the financial field, and extracting triple information from the updated business knowledge base; and generating the target knowledge graph based on the entity recognition results and triple information.

[0143] The processor can invoke information and applications stored in the memory via a transmission device to perform the following steps: generating a target knowledge graph based on entity recognition results and triple information, including: constructing nodes corresponding to entities based on entities in the entity recognition results; wherein, the nodes include at least: event work order information, customer information, product information, financial rule information, and solution information; determining the relationships between entities based on triple information, and constructing edges between nodes based on the relationships between entities, wherein the relationships include at least: the relationship between event work order information and product information, the relationship between solution information and financial rule information, and the relationship between event work order information; and determining the target knowledge graph based on the nodes and edges.

[0144] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: When an event ticket involves financial risk information, the event ticket and candidate solutions are sent to the target object, including: determining whether the event ticket contains target keywords, wherein the target keywords are words extracted from the financial risk information, and the target keywords include: a first keyword and a second keyword, the second keyword being obtained by representing the first keyword using a regular expression; the target keywords belong to at least one of the following categories: first risk level, second risk level, with the first risk level having a higher threat level than the second risk level; if the event ticket contains target keywords and the target keywords meet preset conditions, the event ticket and candidate solutions are sent to the target object, wherein the preset conditions include at least one of the following: the number of target keywords belonging to the first risk level is greater than a first preset value, or the number of target keywords belonging to the second risk level is greater than a second preset value.

[0145] The processor can access information and applications stored in the memory via a transmission device to execute the following steps: Before generating candidate solutions for event work orders through a graph neural network, the above method further includes: training a preset model using historical event work orders, historical response information corresponding to historical event work orders, and historical feedback information for the historical response information to obtain a first model; collecting first event work orders received by financial institutions within a preset time period; calculating the reward score of the first model based on the reward calculation method of the reinforcement learning model, according to the first response information output by the first model for the first event work order and the first feedback information for the first response information; wherein, the reinforcement learning model includes at least: state information, action information, and reward calculation method; if the reward score is less than a preset threshold, retraining the first model until the reward score is greater than or equal to the preset threshold to obtain a graph neural network.

[0146] The processor can access information and applications stored in memory via a transmission device to perform the following steps: generating response information based on candidate solutions, including: determining whether the event ticket involves multiple services within a financial institution; if the event ticket involves multiple services, generating multiple solutions corresponding to the multiple services using a graph neural network based on multimodal data and a target knowledge graph, and generating response information based on the multiple solutions; if the event ticket does not involve multiple services, generating response information based on candidate solutions.

[0147] The processor can access information and applications stored in memory via a transmission device to execute the following steps: extract feature data of multiple data types from the event work order, and fuse the feature data of multiple data types through a cross-modal attention mechanism to obtain multimodal data, including: extracting first feature data from image data in the event work order using optical character recognition technology, wherein the image data includes: text information in multiple languages ​​and handwritten text information; extracting second feature data from audio and video data in the event work order using speech-to-text technology, wherein the audio and video data is data after noise suppression and timestamp annotation; performing entity recognition on text data in the event work order using natural language processing technology to obtain third feature data, wherein the text data is data after text cleaning and regular expression filtering; and fusing the first feature data, second feature data, and third feature data through a cross-modal attention mechanism to obtain multimodal data.

[0148] Those skilled in the art will understand that Figure 7The structure shown is for illustrative purposes only. Electronic devices can also be smartphones (such as Android phones, iOS phones, etc.), tablets, PDAs, mobile Internet devices (MIDs), PADs, and other terminal devices. Figure 7 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 7 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 7 The different configurations shown.

[0149] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0150] Example 4

[0151] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the customer event ticket processing method provided in Embodiment 1.

[0152] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0153] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing the processing method steps of a customer's event work order.

[0154] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0155] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0156] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0157] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0158] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0159] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0160] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for processing customer event work orders, characterized in that, include: Upon detecting an event ticket submitted by a customer, feature data of multiple data types is extracted from the event ticket, and the feature data of multiple data types is fused through a cross-modal attention mechanism to obtain multimodal data. The multimodal data and the target knowledge graph are input into a graph neural network, which generates candidate solutions for the event work order; wherein, the target knowledge graph is generated based on the financial institution's historical event work orders and business knowledge base; If the event ticket involves financial risk information, the event ticket and the candidate solution are sent to the target object for processing. If the event ticket does not involve the financial risk information, a response message is generated based on the candidate solution, and the event ticket is responded to based on the response message.

2. The method according to claim 1, characterized in that, The multimodal data and target knowledge graph are input into a graph neural network, and candidate solutions for the event work order are generated through the graph neural network, including: The multimodal data is encoded by a bidirectional converter encoder to obtain a first embedding vector; The nodes and edges in the target knowledge graph are encoded using a relational graph convolutional network to obtain a second embedding vector; The first embedding vector and the second embedding vector are concatenated to obtain the fused vector; The fused vector is input into the graph neural network, which then generates candidate solutions for the event work order based on the solutions of the historical event work orders.

3. The method according to claim 1, characterized in that, Before inputting the multimodal data and the target knowledge graph into the graph neural network, the method further includes: Every preset time interval, document information from the financial institutions is collected, as well as document information related to the financial field from various external channels, to obtain a document information set; Update the business knowledge base based on the document information set; Based on the financial field, entity recognition is performed on the updated business knowledge base, and triple information is extracted from the updated business knowledge base; The target knowledge graph is generated based on the entity recognition results and the triple information.

4. The method according to claim 3, characterized in that, The target knowledge graph is generated based on the entity recognition results and the triple information, including: Based on the entities in the entity recognition results, construct the nodes corresponding to the entities; wherein, the nodes include at least: event work order information, customer information, product information, financial rule information, and solution information; The association between the entities is determined based on the triple information, and the edges between the nodes are constructed based on the association between the entities. The association includes at least the following: the association between the event work order information and the product information, the association between the solution information and the financial rule information, and the association between the event work order information. The target knowledge graph is determined based on the nodes and edges.

5. The method according to claim 1, characterized in that, In cases where the event ticket involves financial risk information, sending the event ticket and the candidate solution to the target object includes: Determine whether the event work order contains target keywords, wherein the target keywords are words extracted from the financial risk information, and the target keywords include: a first keyword and a second keyword, wherein the second keyword is obtained by representing the first keyword using a regular expression; the target keywords belong to at least one of the following categories: a first risk level and a second risk level, wherein the threat level of the first risk level is higher than that of the second risk level; If the event ticket contains the target keyword and the target keyword meets the preset conditions, the event ticket and the candidate solution are sent to the target object. The preset conditions include at least one of the following: the number of target keywords belonging to the first risk level is greater than a first preset value, and the number of target keywords belonging to the second risk level is greater than a second preset value.

6. The method according to claim 1, characterized in that, Before generating candidate solutions for the event work order using the graph neural network, the method further includes: The preset model is trained using the historical event work orders, the historical response information corresponding to the historical event work orders, and the historical feedback information for the historical response information to obtain the first model; Collect the first event work order received by the financial institution within a preset time period; The reward calculation method based on the reinforcement learning model calculates the reward score of the first model based on the first response information output by the first model in response to the first event work order and the first feedback information in response to the first response information; wherein, the reinforcement learning model includes at least: state information, action information, and the reward calculation method; If the reward score is less than a preset threshold, the first model is retrained until the reward score is greater than or equal to the preset threshold, thus obtaining the graph neural network.

7. The method according to claim 1, characterized in that, Based on the candidate solutions, response information is generated, including: Determine whether the event ticket involves multiple services within the financial institution; When the event ticket involves multiple services, the graph neural network generates multiple solutions corresponding to the multiple services based on the multimodal data and the target knowledge graph, and generates the response information based on the multiple solutions; If the event ticket does not involve the multiple services, the response information is generated based on the candidate solutions.

8. The method according to claim 1, characterized in that, Feature data of multiple data types is extracted from the event work order, and the feature data of multiple data types are fused through a cross-modal attention mechanism to obtain multimodal data, including: The first feature data is extracted from the image data in the event work order using optical character recognition technology, wherein the image data includes: text information in multiple language types and handwritten text information; The second feature data is extracted from the audio and video data in the event work order using speech-to-text technology, wherein the audio and video data is data that has undergone noise suppression processing and timestamp annotation processing; The text data in the event work order is subjected to entity recognition using natural language processing technology to obtain third feature data, wherein the text data is the result of text cleaning and regular expression filtering. The first feature data, the second feature data, and the third feature data are fused using a cross-modal attention mechanism to obtain the multimodal data.

9. A device for processing customer event work orders, characterized in that, include: The extraction unit is used to extract feature data of multiple data types from the event work order when a customer-submitted event work order is detected, and to fuse the feature data of multiple data types through a cross-modal attention mechanism to obtain multimodal data. The first generation unit is used to input the multimodal data and the target knowledge graph into a graph neural network, and generate candidate solutions for the event work order through the graph neural network; wherein, the target knowledge graph is generated based on the historical event work orders and business knowledge base of the financial institution; The processing unit is configured to send the event work order and the candidate solution to the target object for processing the event work order when the event work order involves financial risk information. A response unit is configured to generate response information based on the candidate solutions and respond to the event work order based on the response information when the event work order does not involve the financial risk information.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the customer event ticket processing method according to any one of claims 1 to 8.

Images