Financial artificial intelligence management service platform and financial business processing method
Through the design of the financial artificial intelligence management service platform, using technologies such as large language models and financial knowledge graphs, the problem that large models in existing financial business cannot understand user needs is solved, and efficient financial business processing and resource conservation are achieved, adaptability is achieved. Strong, solves the problems of low business success rate and high resource requirements.
Patent Information
- Application Number
- CN202411287974.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-13
AI Technical Summary
The large models in existing financial business cannot effectively understand user needs, resulting in low success rate and long time, complex model and high resource requirements, unable to adapt to specific application scenarios, and inaccurate understanding when processing Chinese, Issues of information errors or lags.
A financial artificial intelligence management service platform was designed, including interactive modules, semantic analysis modules, model scheduling modules and model collections. It processes the service processing text input by users through a large language model, and generates alternative services, their description information and reference data links. , Users can select the target business and retrieve the corresponding model for processing, adopt the model training method of linear residual adjustment and bypass matrix training, and combine the financial knowledge graph and vector database for knowledge search.
It has achieved efficient guidance and assistance to users in handling financial services, improved the success rate and efficiency of business processing, reduced resource requirements, strong adaptability, and was able to accurately understand and handle financial business needs, avoiding the difficulties in model interpretation and increased computational volume. .
Smart Images

Figure CN118822705B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of financial data processing, and specifically, to a financial artificial intelligence management service platform and a financial business processing method. Background Art
[0002] The large models in related technologies are mostly general large models. In the financial business processing scenario, they cannot fully understand the needs of users when handling financial business, resulting in the inability to guide and assist users in handling the required business.
[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention
[0004] The embodiments of the present application provide a financial artificial intelligence management service platform and a financial business processing method to at least solve the technical problems of low business processing success rate and excessive time consumption caused by the inability of related technologies to guide and assist users in handling the required financial business.
[0005] According to one aspect of an embodiment of the present application, a financial artificial intelligence management service platform is provided, including: an interaction module, a semantic analysis module, a retrieval module, a model scheduling module and a model set, wherein the interaction module is used to receive a business processing text input by a user and forward the business processing text to the semantic analysis module, wherein the business processing text is used to indicate the financial business that the user expects to handle; the semantic analysis module is used to use a large language model to process the business processing text, and obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used by the large language model when processing the business processing text; the interaction module is also used to display to the user each alternative business and its description information obtained by the semantic analysis module, as well as the reference data link, and determine a target business from each alternative business based on the user's selection operation; the model scheduling module is used to retrieve a target model corresponding to the target business from the model set based on the target business to process the target business, wherein the model set includes multiple neural network models with different structures.
[0006] Optionally, the semantic analysis module is also used to: update the reference data set corresponding to the reference data link based on the user's modification operation on the reference data link through the interactive module, wherein the modification operation includes: adding a reference data link, deleting a reference data link; instructing the large language model to process the business handling text again based on the updated reference data set to obtain updated alternative businesses and their description information.
[0007] Optionally, each model in the model set is trained in the following manner: first-category model parameters and second-category model parameters are determined in the model to be trained, wherein the number of first-category model parameters is greater than the number of second-category model parameters; the second-category model parameters are trained by linear residual adjustment, and the first-category model parameters are frozen during the training process; the first-category model parameters and the bypass matrix parameters obtained after training the second-category model parameters are merged to obtain a trained model.
[0008] Optionally, the semantic analysis module uses a large language model to process the business processing text, including: converting the business processing text into a business processing text vector; using a large language model to determine a knowledge vector that matches the business processing text vector in a preset vector library; and using a large language model to process the business processing text vector and the knowledge vector that matches the business processing text vector.
[0009] Optionally, the model scheduling module retrieves the target model corresponding to the target business from the model set to process the target business, including: obtaining a first screening model set based on the correlation between the models in the model set and the target business, wherein the models in the first screening model set are models whose correlation indicators with the target business meet preset requirements; obtaining the business processing results obtained after each model in the first screening model set processes the target business, and again filtering the target business processing results from each business processing result and displaying them to the user.
[0010] Optionally, the model scheduling module retrieves the target model corresponding to the target business from the model collection to process the target business, including: determining the task processing flow corresponding to the target business, wherein the task processing flow is used to indicate the business system required to be called to process the target business; and calling the business system according to the task processing flow through the target model to process the target business.
[0011] Optionally, the model scheduling module retrieves the target model corresponding to the target business from the model set to process the target business, including: displaying the alternative model set and the model description information of each model in the alternative model set to the user; determining the target model from the alternative model set based on the user's selection operation, and instructing the target model to process the data input by the user.
[0012] Optionally, the financial artificial intelligence management service platform also includes a retrieval database, wherein the retrieval database includes a financial knowledge graph, a vector database, a relational database, and an Internet knowledge base.
[0013] Optionally, the semantic analysis module is also used to: extract keywords from business processing texts; determine document identification information corresponding to the keywords in the retrieval database using an inverted index based on the keywords; determine the correlation between the documents corresponding to each document identification information and the keywords; determine the target document from the documents corresponding to the document identification as a reference document for the large language model based on the correlation, and generate a document summary of the document with the greatest correlation and display it to the user through the interactive module.
[0014] According to another aspect of an embodiment of the present application, a financial business processing method is also provided, including: receiving a business processing text input by a user, wherein the business processing text is used to indicate a financial business that the user expects to handle; using a large language model to process the business processing text, to obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used when the large language model processes the business processing text; displaying each alternative business and its description information, as well as a reference data link to the user; determining a target business from each alternative business based on the user's selection operation, and retrieving a target model corresponding to the target business from a model set to process the target business.
[0015] According to another aspect of an embodiment of the present application, a financial business processing device is also provided, including: a first processing module, used to receive a business processing text input by a user, wherein the business processing text is used to indicate the financial business that the user expects to handle; a second processing module, used to use a large language model to process the business processing text, and obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used by the large language model when processing the business processing text; a third processing module, used to display each alternative business and its description information, as well as a reference data link to the user; a fourth processing module, used to determine a target business from each alternative business based on the user's selection operation, and to retrieve a target model corresponding to the target business from a model set to process the target business.
[0016] According to another aspect of an embodiment of the present application, a non-volatile storage medium is provided, in which a program is stored, wherein when the program is running, a device where the non-volatile storage medium is located is controlled to execute a financial business processing method.
[0017] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the financial business processing method is executed when the program is run.
[0018] According to another aspect of an embodiment of the present application, a computer program product is also provided, including a computer program, and the computer program implements a financial business processing method when executed by a processor.
[0019] In an embodiment of the present application, a financial artificial intelligence management service platform is provided, which receives a business processing text input by a user, wherein the business processing text is used to indicate the financial business that the user expects to handle; uses a large language model to process the business processing text to obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used by the large language model when processing the business processing text; displays each alternative business and its description information, as well as a reference data link to the user; determines a target business from each alternative business based on the user's selection operation, and retrieves a target model corresponding to the target business from a model set to process the target business. By using a large language model to determine the alternative business based on the business processing text, and after the user selects the target business from the alternative business, the target model corresponding to the target business is retrieved to process the target business, the purpose of accurately determining the financial business that the user expects to handle is achieved, thereby achieving the technical effect of efficiently guiding and assisting the user in handling the required financial business, thereby solving the technical problem of low business handling success rate and excessive time consumption caused by the inability to guide and assist the user in handling the required financial business in the related technology. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0021] Figure 1 It is a structural schematic diagram of a computer terminal (mobile terminal) provided according to an embodiment of the present application;
[0022] Figure 2 It is a structural diagram of a financial artificial intelligence management service platform provided according to an embodiment of the present application;
[0023] Figure 3 It is a flowchart of a financial service processing method provided according to an embodiment of the present application;
[0024] Figure 4 It is a flowchart of a financial business handling process provided according to an embodiment of the present application;
[0025] Figure 5 It is a structural diagram of a financial business processing device provided according to an embodiment of the present application. DETAILED DESCRIPTION
[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0028] The large models in related technologies are mostly general-purpose large models. In the scenario of financial business handling, they cannot fully understand the needs of users when handling financial business, resulting in the inability to guide and assist users in handling the required business. In addition, the large models in related technologies have problems such as complex execution processes and high equipment requirements; they are not suitable for specific application scenarios and cannot meet specific user needs; the model parameters are too large and require strict computing power; the application scenarios are scattered and the application functions are relatively simple.
[0029] Specifically, users of the models in the related technologies need to perform multiple steps when operating, which not only increases the difficulty of operation, but also may lead to reduced efficiency. Especially in application scenarios that require quick response, complex processes will greatly affect the use effect. Technical implementation may have very high performance requirements for the execution equipment, which is particularly prominent in some resource-constrained environments. For example, in areas lacking high-performance computing equipment, this has become an important factor limiting the popularization and application of large model technology in the financial business field.
[0030] Moreover, the large models in related technologies have poor adaptability in actual application scenarios and cannot fully meet the specific needs of users. For example, in certain specific application scenarios, large models may not be able to effectively solve the problem, or their output results may not meet the actual needs. In addition, large models also lack domain knowledge and real-time knowledge when processing factual knowledge, resulting in them providing wrong or outdated information in some cases.
[0031] In addition, the structure of the general large model used in related technologies is very complex, which makes the model difficult to interpret, resulting in an increase in the amount of calculation of the model and affecting the efficiency of the model. If the model parameters used by the technology are too large, higher requirements will be placed on computing power. This not only increases the demand for high-performance GPUs, but may also make it difficult to obtain the required hardware resources due to the tense international environment. If the technology is built on English models such as LLAMA2 or LLAMA3, its parameter optimization is mainly based on the English data corpus, which may lead to inaccurate understanding, incorrect information, missing information or delayed information when dealing with Chinese problems. This is particularly evident in applications that require a high degree of localization or cultural specificity.
[0032] Moreover, the large model applications currently used in banking and other financial-related industries are mostly distributed in different systems and fields within each bank. In actual use, users often need to switch between different business systems to fully utilize the functions of the large model. This reflects the challenges of integration and user experience in the application of large models in the banking industry. Although the independent deployment of large models in each system is conducive to solving specific business problems in a targeted manner, it may also lead to data silos, duplicated functional construction, and complicated user operations.
[0033] In order to solve the above problems, relevant solutions are provided in the embodiments of the present application, which are described in detail below.
[0034] According to an embodiment of the present application, an apparatus embodiment of a financial artificial intelligence management service platform and a method embodiment of a financial business processing method are provided. It should be noted that the steps shown in the flowchart of 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 can be executed in an order different from that shown here.
[0035] The method embodiments and device embodiments provided in the embodiments of the present application may be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a financial artificial intelligence management service platform and a financial business processing method is shown. Figure 1As shown, the computer terminal 10 (or mobile device 10) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.
[0036] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).
[0037] 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 financial business processing method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned financial artificial intelligence management service platform and financial business processing method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.
[0038] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0039] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).
[0040] In the above operating environment, the embodiment of the present application provides a financial artificial intelligence management service platform, such as Figure 2 As shown, the platform includes: an interaction module 20, a semantic analysis module 22, a retrieval module 24, a model scheduling module 26 and a model set 28, wherein the interaction module 20 is used to receive a business processing text input by a user and forward the business processing text to the semantic analysis module 22, wherein the business processing text is used to indicate the financial business that the user expects to handle; the semantic analysis module 22 is used to process the business processing text using a large language model, and obtain the alternative business output by the large language model based on the business processing text, the description information of each alternative business, and the reference data link used by the large language model when processing the business processing text; the interaction module 20 is also used to display to the user each alternative business and its description information, as well as the reference data link obtained by the semantic analysis module 22, and determine the target business from each alternative business according to the user's selection operation; the model scheduling module 26 is used to retrieve the target model corresponding to the target business from the model set 28 according to the target business to process the target business, wherein the model set 28 includes multiple neural network models with different structures.
[0041] In some embodiments of the present application, the semantic analysis module 22 is also used to: update the reference data set corresponding to the reference data link based on the user's modification operation on the reference data link through the interaction module 20, wherein the modification operation includes: adding a reference data link, deleting a reference data link; instructing the large language model to process the business handling text again based on the updated reference data set to obtain updated alternative businesses and their description information.
[0042] As an optional implementation, each model in the model set 28 is trained in the following manner: first-category model parameters and second-category model parameters are determined in the model to be trained, wherein the number of first-category model parameters is greater than the number of second-category model parameters; the second-category model parameters are trained by linear residual adjustment, and the first-category model parameters are frozen during the training process; the first-category model parameters and the bypass matrix parameters obtained after training the second-category model parameters are merged to obtain a trained model.
[0043] Specifically, linear residual adjustment refers to adding one or more linear layers (fully connected layers) before the output layer of the model. These layers can learn how to convert the internal representation of the model into the final output. The following is the specific training process:
[0044] The first step is to freeze the underlying parameters: During the fine-tuning phase, the underlying parameters of the pre-trained model are frozen to prevent over-training on new tasks and destroying the knowledge learned by the model on the original task.
[0045] The second step is to add linear layers: before the last layer or layers of the model, add one or more linear layers (also called fully connected layers), which map the internal representation of the model to the desired output space.
[0046] The third step is to train the linear layers: on the new task data, only these newly added linear layers are trained, while the underlying parameters of the pre-trained model remain unchanged. In this way, the model can be quickly adapted to new tasks without having to train the entire model from scratch.
[0047] Step 4, residual connection: In some cases, linear layers may be used in conjunction with residual connections. Residual connections allow the internal representation of the original model to directly affect the output, which helps alleviate the gradient vanishing problem and allows deeper network training.
[0048] Step 5: Loss function optimization: Use an appropriate loss function (such as cross entropy loss, mean square error, etc.) to measure the performance of the model on the new task, and optimize the parameters of the linear layer through the back propagation algorithm.
[0049] In addition, the bypass matrix mentioned above refers to an additional weight matrix added in a deep learning network to increase the network's expressive power or to achieve certain specific functions. The bypass matrix is a weight matrix that can be learned or designed. It can create additional paths in the network so that information can flow differently in the network.
[0050] In some embodiments of the present application, the semantic analysis module 22 uses a large language model to process business processing texts, including: converting the business processing texts into business processing text vectors; using a large language model to determine a knowledge vector that matches the business processing text vector in a preset vector library; and using a large language model to process the business processing text vectors and the knowledge vector that matches the business processing text vectors.
[0051] As an optional implementation, the model scheduling module 26 retrieves the target model corresponding to the target business from the model set 28 to process the target business, including: obtaining a first screening model set 28 based on the correlation between the models in the model set 28 and the target business, wherein the models in the first screening model set 28 are models whose correlation indicators with the target business meet preset requirements; obtaining the business processing results obtained after each model in the first screening model set 28 processes the target business, and again filtering the target business processing results from each business processing result and displaying them to the user.
[0052] In some embodiments of the present application, when screening the models in the model set 28, in addition to screening the models first and then screening the model results, a two-stage ranking optimization method can also be used. For example, after the user enters a query instruction, in the sorting process of the first stage, a simple model and a small number of features can be used to quickly screen out a certain number of results with top relevance. Later, in the sorting process of the second stage, complex models and a large number of features can be used for detailed sorting, and several results at the top of the sorting can be displayed to the user. Through this two-stage sorting strategy, the accuracy of sorting and user satisfaction can be improved while ensuring sorting efficiency. It is suitable for application scenarios that need to process a large number of candidate results and require sorting quality at the same time.
[0053] In some embodiments of the present application, the model scheduling module 26 retrieves the target model corresponding to the target business from the model set 28 to process the target business, including: determining the task processing flow corresponding to the target business, wherein the task processing flow is used to indicate the business system required to be called to process the target business; and calling the business system according to the task processing flow through the target model to process the target business.
[0054] As an optional implementation, the model scheduling module 26 retrieves the target model corresponding to the target business from the model set 28 to process the target business, including: displaying the alternative model set 28 and the model description information of each model in the alternative model set 28 to the user; determining the target model from the alternative model set 28 based on the user's selection operation, and instructing the target model to process the data input by the user.
[0055] In some embodiments of the present application, the financial artificial intelligence management service platform also includes a retrieval database, wherein the retrieval database includes a financial knowledge graph, a vector database, a relational database, and an Internet knowledge base.
[0056] As an optional implementation, the semantic analysis module 22 is also used to: extract keywords from the business processing text; determine the document identification information corresponding to the keywords in the retrieval database using an inverted index based on the keywords; determine the correlation between the documents corresponding to each document identification information and the keywords; determine the target document from the documents corresponding to the document identification as the reference document of the large language model based on the correlation, and generate a document summary of the document with the highest correlation and display it to the user through the interactive module 20.
[0057] Specifically, in the platform provided by the embodiment of the present application, a forward index and an inverted index can be combined to build an efficient knowledge search engine. Combining the forward index and the inverted index can not only quickly locate the document containing the query vocabulary, but also quickly obtain the detailed content of the document. When retrieving relevant reference files, the index can be optimized, for example, different indexing strategies can be adopted for high-frequency vocabulary and low-frequency vocabulary, and a cache mechanism can be used to accelerate hot word retrieval. The specific search process may include the following steps:
[0058] The first step is query parsing: After the user enters a query, the search engine first parses the query, such as word segmentation and removal of stop words.
[0059] The second step is index query: use the inverted index to find all document IDs that match the query terms.
[0060] The third step is relevance sorting: Sort the documents according to their relevance to the query. The relevance is calculated using algorithms such as TF-IDF (term frequency-inverse document frequency) and BM25.
[0061] Step 4: Result display: Based on the sorting results, display the most relevant document content or summary to the user.
[0062] In some embodiments of the present application, when searching for relevant information in a database, a text2SQL algorithm may be used to convert a natural language query text into an SQL query instruction, specifically including the following steps:
[0063] The first step is data preprocessing. The input natural language query text is segmented and decomposed into words or phrases. Part-of-speech tagging: Each word after segmentation is tagged with a part-of-speech tag, such as noun, verb, adjective, etc. Named entities in the text are identified, such as table names, field names, etc. Syntactically analyze the sentences, build a dependency tree of the sentences, and understand the relationship between words.
[0064] The second step is to understand the query intent and determine the user's query intent, such as whether to query data, update data, or perform other operations.
[0065] Step 3: SQL query generation and template matching: select the appropriate SQL template according to the logical form. Query construction: fill in the specific table name, field name, conditions, etc. according to the logical form and template to build a complete SQL query statement.
[0066] In the above operating environment, the present application embodiment also provides a financial service processing method, which is applicable to Figure 2 In the financial artificial intelligence management service platform. Figure 3 As shown, the method comprises the following steps:
[0067] Step S302, receiving a business handling text input by a user, wherein the business handling text is used to indicate the financial business that the user wishes to handle;
[0068] In the technical solution provided in step S302, in order to solve the interpretability problem of large model-generated content in the application process of banking business scenarios, vector databases, knowledge graphs, distributed big data, etc. can be used to build a financial value matrix for financial service providers such as banks, thereby gradually forming a financial knowledge graph covering all scenarios of business marketing, operation guidance, asset allocation, compliance and security, and financial analysis. In this way, when semantic recognition is performed on the business processing text entered by the user, the financial knowledge graph can be combined with a knowledge search engine that integrates multiple indexes in forward and reverse order, providing an accurate, reliable, and effective source of knowledge for the large model to generate content, and providing the source of citations during user interaction, so that the content generated by the large model "makes sense". In addition, the large language model can be connected to an external real-time database to improve the large model's real-time question-and-answer capabilities for news media data.
[0069] In addition, as an optional implementation, an expert model in the indicator knowledge field can be constructed, and the large language model can be connected to the expert model to decouple the complex data structure of the underlying data warehouse. This allows the large language model to understand the data more accurately while avoiding the occurrence of AI hallucinations.
[0070] Step S304, using the large language model to process the business handling text, obtaining the candidate businesses output by the large language model based on the business handling text, description information of each candidate business, and a reference data link used by the large language model when processing the business handling text;
[0071] As an optional implementation method, the above-mentioned large language model can also provide basic functions such as copywriting, code generation, knowledge question and answer in a conversational interactive manner, as well as extended functions such as RPA task execution, enterprise information query, and system call.
[0072] In some embodiments of the present application, in order to improve the security and interpretability of the large language model, a semantic vector library can be constructed, and the information organization and language generation capabilities of the large model can be used to achieve accurate and interpretable semantic generation, meet the requirements of banking business scenarios for model accuracy and interpretability, and meet the bank's needs for large model-based knowledge applications.
[0073] In addition, in order to achieve efficient knowledge management and convenient retrieval, the knowledge must undergo strict sorting and structuring before being stored. Each piece of data will be carefully classified and classified into corresponding subject areas according to its content attributes, such as macroeconomic analysis, financial technology trends, personal financial management knowledge, etc. Such partitioning and classification not only helps to improve the systematic nature of the knowledge base, but also greatly facilitates users to quickly locate the required content in massive information.
[0074] Optionally, in order to ensure the timeliness and comprehensiveness of the knowledge base, a scheduled batch task automatic entry mechanism can also be used. This automated process can regularly capture fresh information from the above-mentioned sources, automatically complete information screening, format conversion and standardization, and then seamlessly connect to the knowledge base system, reducing the tediousness and delay of manual operations and ensuring the continuous updating and expansion of knowledge base content.
[0075] In the technical solution provided in step S304, the step of using a large language model to process the business processing text includes: converting the business processing text into a business processing text vector; using a large language model to determine a knowledge vector that matches the business processing text vector in a preset vector library; and using a large language model to process the business processing text vector and the knowledge vector that matches the business processing text vector.
[0076] Specifically, when processing business handling texts, we can first conduct in-depth semantic analysis and vectorization of user questions in the text. This process is equivalent to converting natural language text into mathematical vectors that can be understood by machines. Each vector represents the semantic features of the question, enabling the machine to understand the essence of the question beyond the vocabulary level.
[0077] After that, the vector of the user's question can be efficiently matched with the vectors in the pre-built huge knowledge base. This knowledge base contains structured and unstructured information carefully sorted from various sources. Each knowledge point has been vectorized to form a "knowledge universe" in a high-dimensional space. By calculating the similarity between the user's question vector and the vectors of each knowledge point in the library, the system quickly selects the most relevant answer candidates and achieves accurate matching.
[0078] As an optional implementation method, in order to further expand the knowledge boundary, the large language model can also be connected to various plug-ins and API interfaces on the Internet to actively absorb high-quality content from social media, professional forums, online courses and other wider channels. This not only greatly enriches the coverage of the knowledge base, but also promotes cross-domain knowledge integration.
[0079] In order to ensure data security, if external data needs to be called, the API interface can be screened, and the large language model can only communicate with external data through the screened reliable API interface. Such data can include but is not limited to weather forecasts, financial information, geographic information, etc.
[0080] Step S306, displaying each candidate service and its description information, as well as a reference data link to the user;
[0081] In the technical solution provided in step S306, after displaying each alternative business and its description information, as well as the reference data link to the user, the financial business processing method also includes: updating the reference data set corresponding to the reference data link based on the user's modification operation on the reference data link, wherein the modification operation includes: adding a reference data link, deleting a reference data link; instructing the large language model to process the business handling text again based on the updated reference data set to obtain updated alternative business and its description information.
[0082] Step S308 , determining a target business from among the candidate businesses according to the user's selection operation, and retrieving a target model corresponding to the target business from the model set to process the target business.
[0083] In the technical solution provided in step S308, each model in the model set is trained in the following manner: first-category model parameters and second-category model parameters are determined in the model to be trained, wherein the number of first-category model parameters is greater than the number of second-category model parameters; the second-category model parameters are trained by linear residual adjustment, and the first-category model parameters are frozen during the training process; the first-category model parameters and the bypass matrix parameters obtained after training the second-category model parameters are merged to obtain a trained model.
[0084] In some embodiments of the present application, each model can be fine-tuned by using the linear residual adjustment (LoRA) method, so that only a small number of weights can be adjusted locally while keeping most of the parameters of the pre-trained model unchanged. This helps to enhance its adaptability to financial scenarios without destroying the universality of the original model.
[0085] Specifically, when training the model, the strategy of freezing the parameters of the pre-trained model is first adopted, which means that during the fine-tuning process, the parameters of the main part of the model will not change significantly, so as to protect the original language understanding and generation capabilities of the model from loss. After the fine-tuning training is completed, the original parameters of the pre-trained model are combined with the bypass matrix parameters obtained through training to achieve the superposition of the two. When predicting and generating financial-related text content, the model trained in this way can fully mobilize the extensive knowledge reserve of the pre-trained model and accurately reflect the specialized knowledge and practical rules in the financial business field, thereby greatly improving the generalization ability and application effect of the model for various complex financial scenarios.
[0086] As an optional implementation, retrieving a target model corresponding to a target business from a model set to process the target business includes: obtaining a first screening model set based on the correlation between the models in the model set and the target business, wherein the models in the first screening model set are models whose correlation indicators with the target business meet preset requirements; obtaining the business processing results obtained after each model in the first screening model set processes the target business, and again filtering the target business processing results from each business processing result and displaying them to the user.
[0087] In some embodiments of the present application, the step of retrieving a target model corresponding to a target business from a model set to process the target business includes: displaying a set of alternative models and model description information of each model in the alternative model set to a user; determining a target model from the alternative model set based on a user's selection operation, and instructing the target model to process the data input by the user.
[0088] Specifically, the above model collection can be set as a model application market. Users can flexibly select suitable models for interaction and application development according to their own needs, such as looking for the most matching service model in the fields of text generation, intelligent question answering, automatic summarization, creative writing, code writing, and intelligent customer service.
[0089] It should also be noted that multiple models can also work together to handle the financial business or other needs that users expect to handle. For example, through the collaboration of multiple models, multiple solutions can be provided for customers to choose from when facing complex problems. Each model may give different answers to the same question based on its unique algorithm and training data. In this way, customers can compare and choose from multiple perspectives to find the answer that best meets their needs. This diversified answer selection mechanism greatly improves customer satisfaction and the accuracy of problem solving.
[0090] As an optional implementation, the step of retrieving a target model corresponding to the target business from a model set to process the target business includes: determining a task processing flow corresponding to the target business, wherein the task processing flow is used to indicate the business system required to be called to process the target business; and calling the business system according to the task processing flow through the target model to process the target business.
[0091] As an optional implementation method, when the large language model confirms the business that the user expects to handle, that is, after the distribution and identification of the user's problem, it can further determine the RPA process that matches the user's problem and activate the PRA task corresponding to the process. After that, the corresponding model can be called to complete various operations across systems based on the PRA task, such as data capture, form filling, and information synchronization between systems. This capability enables the system to dynamically respond to user needs and generate or update information in real time, greatly enhancing immediacy and efficiency.
[0092] In some embodiments of the present application, based on the process orchestration capabilities of the large language model, multiple services such as plug-in calls, file parsing, knowledge base entry, and large models can be deeply integrated to realize process-based, intelligent, and efficient AI Agent capabilities. Specifically, the Agent can be connected to knowledge, and various Agent skills can be developed at the same time to understand user intentions, perform task planning, and task decomposition to complete various business tasks. By integrating multiple vectorized models such as BGE and multilingual-e5-large, a business knowledge base can be independently constructed. The business data is precipitated to realize the functions of Agent perception and understanding. In addition, for multi-step task scenarios with complex logic, multiple nodes can be flexibly assembled, with a stable output structure, to support repeated task execution. Existing capabilities within the bank can be inherited or capabilities can be added autonomously as Agent skills to promote the continuous precipitation and improvement of basic skills and business skills, and improve the efficiency of large model service development.
[0093] In some embodiments of the present application, the complete process of handling financial services is as follows: Figure 4 As shown in Figure 1, it includes five steps: user input, intent understanding, knowledge retrieval, model output, and interface return. Figure 4It can be seen that the content input by the user can be text or voice, and when the input content is voice, the voice will be converted into text before further processing. After that, the intention engine based on the large language model can be used to understand the user's intention based on data from data sources such as knowledge graphs, vector databases, relational databases, and Internet knowledge bases. After assembling the prompt template according to the understanding result, another large model determines the corresponding model from the model set according to the prompt template to process the financial business that the user expects to process. The business processing results can then be displayed to the user in a variety of ways, such as text output, voice broadcast, icon output, etc., and the reference data source can also be displayed to the user. When the user thinks that the current processing result is not the expected result, he or she can repeat the execution by choosing to regenerate. Figure 4 The business process shown.
[0094] The invention adopts a method of receiving a business processing text input by a user, wherein the business processing text is used to indicate the financial business that the user expects to handle; adopting a large language model to process the business processing text, obtaining alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link adopted by the large language model when processing the business processing text; displaying each alternative business and its description information, as well as a reference data link to the user; determining a target business from each alternative business based on the user's selection operation, and retrieving a target model corresponding to the target business from a model set to process the target business. By adopting a large language model to determine the alternative business based on the business processing text, and retrieving a target model corresponding to the target business to process the target business after the user selects the target business from the alternative business, the purpose of accurately determining the financial business that the user expects to handle is achieved, thereby realizing the technical effect of efficiently guiding and assisting the user to handle the required financial business, thereby solving the technical problem of low business processing success rate and excessive time consumption caused by the inability to guide and assist the user to handle the required financial business in the related technology.
[0095] It should also be noted that due to Figure 3 The financial business processing methods in Figure 2 The financial artificial intelligence management service platform shown in Figure 3 The relevant explanations of the financial business processing methods in Figure 2 The financial artificial intelligence management service platform shown in .
[0096] The present application embodiment provides a financial service processing device, Figure 5 is a schematic diagram of the structure of the device. Figure 5It can be seen that the device includes: a first processing module 50, which is used to receive a business processing text input by a user, wherein the business processing text is used to indicate the financial business that the user expects to handle; a second processing module 52, which is used to use a large language model to process the business processing text, and obtain the alternative business output by the large language model based on the business processing text, the description information of each alternative business, and the reference data link used by the large language model when processing the business processing text; a third processing module 54, which is used to display each alternative business and its description information, as well as the reference data link to the user; a fourth processing module 56, which is used to determine the target business from each alternative business according to the user's selection operation, and to retrieve the target model corresponding to the target business from the model set to process the target business.
[0097] In some embodiments of the present application, the steps of the second processing module 52 using a large language model to process business processing text include: converting the business processing text into a business processing text vector; using a large language model to determine a knowledge vector that matches the business processing text vector in a preset vector library; and using a large language model to process the business processing text vector and the knowledge vector that matches the business processing text vector.
[0098] In some embodiments of the present application, after displaying various alternative services and their description information, as well as reference data links to the user, the third processing module 54 is also used to: update the reference data set corresponding to the reference data link based on the user's modification operation on the reference data link, wherein the modification operation includes: adding a reference data link, deleting a reference data link; instructing the large language model to process the business handling text again based on the updated reference data set to obtain updated alternative services and their description information.
[0099] In some embodiments of the present application, each model in the model set is trained in the following manner: determining first-category model parameters and second-category model parameters in the model to be trained, wherein the number of first-category model parameters is greater than the number of second-category model parameters; training the second-category model parameters by linear residual adjustment, and freezing the first-category model parameters during the training process; merging the first-category model parameters and the bypass matrix parameters obtained after training the second-category model parameters to obtain a trained model.
[0100] In some embodiments of the present application, the fourth processing module 56 retrieves the target model corresponding to the target business from the model set to process the target business, including the following steps: obtaining a first screening model set based on the correlation between the models in the model set and the target business, wherein the models in the first screening model set are models whose correlation indicators with the target business meet preset requirements; obtaining the business processing results obtained after each model in the first screening model set processes the target business, and again filtering the target business processing results from each business processing result and displaying them to the user.
[0101] In some embodiments of the present application, the step in which the fourth processing module 56 retrieves the target model corresponding to the target business from the model set to process the target business includes: displaying the alternative model set and the model description information of each model in the alternative model set to the user; determining the target model from the alternative model set based on the user's selection operation, and instructing the target model to process the data input by the user.
[0102] In some embodiments of the present application, the step in which the fourth processing module 56 retrieves the target model corresponding to the target business from the model set to process the target business includes: determining the task processing flow corresponding to the target business, wherein the task processing flow is used to indicate the business system that needs to be called to process the target business; and calling the business system according to the task processing flow through the target model to process the target business.
[0103] It should be noted that the various modules in the above-mentioned financial business processing device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.
[0104] According to an embodiment of the present application, a non-volatile storage medium is also provided, in which a program is stored, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the following financial business processing method: receiving a business processing text input by a user, wherein the business processing text is used to indicate the financial business that the user expects to handle; using a large language model to process the business processing text, to obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used when the large language model processes the business processing text; displaying each alternative business and its description information, as well as a reference data link to the user; determining a target business from each alternative business based on the user's selection operation, and retrieving a target model corresponding to the target business from the model set to process the target business.
[0105] According to an embodiment of the present application, an electronic device is also provided, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the following financial business processing method is executed when the program is running: receiving a business processing text input by a user, wherein the business processing text is used to indicate a financial business that the user wishes to handle; using a large language model to process the business processing text, to obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used when the large language model processes the business processing text; displaying each alternative business and its description information, as well as a reference data link to the user; determining a target business from each alternative business based on the user's selection operation, and retrieving a target model corresponding to the target business from the model set to process the target business.
[0106] According to an embodiment of the present application, a computer program product is also provided, including a computer program, which implements the following financial business processing method when executed by a processor: receiving a business processing text input by a user, wherein the business processing text is used to indicate the financial business that the user expects to handle; using a large language model to process the business processing text, to obtain alternative businesses output by the large language model based on the business processing text, description information of each alternative business, and a reference data link used when the large language model processes the business processing text; displaying each alternative business and its description information, as well as a reference data link to the user; determining a target business from each alternative business based on the user's selection operation, and retrieving a target model corresponding to the target business from a model set to process the target business.
[0107] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.
[0108] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.
[0109] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0110] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0111] If the integrated unit is implemented in the form of 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 the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.
[0112] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A financial artificial intelligence management service platform, characterized in that: It includes interaction module, semantic analysis module, retrieval module, model scheduling module and model collection, among which, The interaction module is used to receive a business handling text input by a user and forward the business handling text to the semantic analysis module, wherein the business handling text is used to indicate the financial business that the user expects to handle; The semantic analysis module is used to process the business handling text using a large language model to obtain candidate businesses output by the large language model based on the business handling text, description information of each candidate business, and reference data links used by the large language model when processing the business handling text, and update the reference data set corresponding to the reference data link according to the user's modification operation on the reference data link through the interactive module; The interactive module is further used to display to the user each of the candidate services and their description information obtained by the semantic analysis module, as well as the reference data link, and to determine a target service from each of the candidate services according to a selection operation of the user; The model scheduling module is used to retrieve a target model corresponding to the target business from the model set according to the target business to process the target business, wherein the model set includes a plurality of neural network models with different structures, and the model scheduling module retrieves a target model corresponding to the target business from the model set to process the target business, including: displaying the candidate model set and model description information of each model in the candidate model set to the user; determining the target model from the candidate model set according to the user's selection operation, and instructing the target model to process the data input by the user; Each model in the model set is trained in the following manner: first-category model parameters and second-category model parameters are determined in the model to be trained, wherein the number of the first-category model parameters is greater than the number of the second-category model parameters; the second-category model parameters are trained by linear residual adjustment, and the first-category model parameters are frozen during the training process; the first-category model parameters and the bypass matrix parameters obtained after training the second-category model parameters are merged to obtain a trained model.
2. The financial artificial intelligence management service platform according to claim 1, characterized in that: The semantic analysis module is also used for: According to the modification operation of the reference data link by the user through the interactive module, the reference data set corresponding to the reference data link is updated, wherein the modification operation includes: adding a reference data link and deleting a reference data link; The large language model is instructed to process the service handling text again according to the updated reference data set to obtain the updated candidate services and description information thereof.
3. The financial artificial intelligence management service platform according to claim 1, characterized in that: The semantic analysis module uses a large language model to process the business handling text, including: Converting the business processing text into a business processing text vector; Using the large language model to determine a knowledge vector matching the business handling text vector in a preset vector library; The large language model is used to process the business processing text vector and the knowledge vector matching the business processing text vector.
4. The financial artificial intelligence management service platform according to claim 1, characterized in that: The model scheduling module retrieves a target model corresponding to the target business from a model set to process the target business, including: According to the correlation between the models in the model set and the target business, a first screening model set is obtained, wherein the models in the first screening model set are models whose correlation indicators with the target business meet preset requirements; The business processing results obtained after each model in the first screening model set processes the target business are obtained, and the target business processing results are screened again from the business processing results and displayed to the user.
5. The financial artificial intelligence management service platform according to claim 1, characterized in that: The model scheduling module retrieves a target model corresponding to the target business from a model set to process the target business, including: Determine a task processing flow corresponding to the target business, wherein the task processing flow is used to indicate a business system to be called to process the target business; The target business is processed by calling the business system according to the task processing flow through the target model.
6. The financial artificial intelligence management service platform according to claim 1, characterized in that: The financial artificial intelligence management service platform also includes a retrieval database, wherein the retrieval database includes a financial knowledge graph, a vector database, a relational database, and an Internet knowledge base.
7. The financial artificial intelligence management service platform according to claim 6, characterized in that: The semantic analysis module is also used for: Extracting keywords from the business handling text; Determine document identification information corresponding to the keyword in the search database by using an inverted index according to the keyword; Determine the correlation between the document corresponding to each of the document identification information and the keyword; According to the relevance, a target document is determined from the documents corresponding to the document identifier as a reference document of the large language model, and a document summary of the document with the greatest relevance is generated and presented to the user through the interactive module.
8. A financial business processing method, characterized in that: Applicable to the financial artificial intelligence management service platform described in any one of claims 1 to 7, comprising: Receiving a business handling text input by a user, wherein the business handling text is used to indicate the financial business that the user desires to handle; Processing the business handling text using a large language model to obtain candidate businesses output by the large language model based on the business handling text, description information of each of the candidate businesses, and a reference data link used by the large language model when processing the business handling text; Displaying each of the candidate services and their description information, as well as the reference data link to the user; A target business is determined from each of the candidate businesses according to a selection operation of the user, and a target model corresponding to the target business is retrieved from a model set to process the target business.
Citation Information
Cited By
Message Processing Auxiliary System
TWI923229B
