Financial operation business question answering method based on large language model based on index algorithm

Abstract

The present application discloses a financial operation business question-answering method based on a large language model with an indexing algorithm. The method includes: receiving a query question input by a target object; identifying the semantic information corresponding to the query question, and determining the identity feature information and behavioral feature information of the target object; determining a target large language model that matches the semantic information, the identity feature information and the behavioral feature information of the target object; searching for a document collection in the database in the target large language model; searching for detailed document content in the document collection; determining the frequency of query keywords, document update time and document quality score indicators as sorting factors; sorting the document collection, and using the sorted document collection as the output result of the target large language model. The present application solves the technical problem that the related technology cannot combine the user's semantic information, identity feature information and behavioral feature information to select a specific large language model to process the query question input by the user.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a large language model financial operation business question and answer method based on an indexing algorithm. Background Art

[0002] Natural Language Processing (NLP) is an important branch of artificial intelligence, which focuses on how computers understand, interpret, and generate human language. In recent years, with the development of deep learning technology, the field of NLP has made significant progress.

[0003] Existing technologies often use keyword-based search algorithms to match user queries, or use shallow natural language processing techniques to understand user input. Such methods have limited performance in processing complex semantic structures and understanding user implicit intentions, and cannot fully utilize the deep understanding and generation capabilities of large language models.

[0004] In addition, existing technologies cannot fully consider the identity and behavior characteristics of users to dynamically adjust the model's response strategy. This means that whether it is a novice employee or a senior expert, whether it is handling routine tasks or responding to emergencies, the information provided by the system is one-size-fits-all and lacks personalization and pertinence.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The present application provides a large language model financial operation business question and answer method based on an index algorithm, so as to at least solve the technical problem that the relevant technology cannot combine the user's semantic information, identity feature information and behavior feature information to select a specific large language model to process the query questions input by the user.

[0007] According to one aspect of the present application, a large language model financial operation business question and answer method based on an indexing algorithm is provided, comprising: receiving a query question input by a target object; identifying semantic information corresponding to the query question, and determining identity feature information and behavioral feature information of the target object; determining a target large language model that matches the semantic information, identity feature information and behavioral feature information of the target object; searching for a document collection in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; searching for document details of the document collection through a forward index, wherein the document details include: document update time, document quality score indicators; determining the frequency of the query keywords in the inverted index and the document update time and document quality score indicators in the forward index as sorting factors; sorting the document collection according to the sorting factors, and using the sorted document collection as the output result of the target large language model.

[0008] Optionally, determining a target large language model that matches the semantic information, the identity feature information of the target object, and the behavior feature information includes: inputting the semantic information, the identity feature information of the target object, and the behavior feature information into a routing selection model, the routing selection model being used to perform the following steps: performing feature extraction processing on the semantic information to obtain a first feature vector, performing feature extraction processing on the identity feature information of the target object to obtain a second feature vector, and performing feature extraction processing on the behavior feature information of the target object to obtain a third feature vector; merging the first feature vector, the second feature vector, and the third feature vector to obtain a target feature vector; inputting the target feature vector into a rule base, wherein the rule base stores pre-set routing rules based on data features and business scenarios, the routing rules are presented in the form of a decision tree, each decision node in the decision tree corresponds to a threshold judgment of a feature dimension, and the branch nodes in the decision tree correspond to different model selection paths; matching the target feature vector according to the routing rules to obtain a target model path corresponding to the target feature vector, wherein the target model path points to one or more target large language models.

[0009] Optionally, the method also includes: after completing the route selection, the routing selection model sends the identification information of the target large language model to the task planning module; after receiving the identification information, the task planning module detects the resource utilization efficiency of the target large language model corresponding to the identification information, wherein the resource utilization efficiency includes: CPU utilization rate and memory occupancy rate; when it is detected that the CPU utilization rate of the target large language model corresponding to the identification information is less than a first preset percentage and / or the memory occupancy rate is less than a second preset percentage, the task planning module sends a target instruction to the routing selection model, wherein the target instruction is used to instruct the routing selection model to redetermine the routing rules.

[0010] Optionally, the method further includes: extracting sample data from a historical data repository at a preset time interval, using a feature vector of the sample data as input and a corresponding target model path as a label to train a routing selection model.

[0011] Optionally, determining a target large language model that matches the semantic information, the identity feature information, and the behavior feature information of the target object includes: determining the task urgency of the query question based on the semantic information, wherein the task urgency includes: urgent, not urgent; determining the target object whose identity feature information and behavior feature information meet at least one of the following preset conditions as a first type of target object: the number of credit accounts is less than a first preset threshold, the number of times the credit report is queried by a financial institution is less than a second preset threshold, and the number of bank account transactions within a preset period of time is less than a third preset threshold; determining the target object whose identity feature information and behavior feature information meet at least one of the following preset conditions as a second type of target object: the number of credit accounts is less than a first preset threshold, the number of times the credit report is queried by a financial institution is less than a second preset threshold, and the number of bank account transactions within a preset period of time is less than a third preset threshold; The number is not less than the first preset threshold, the number of times the credit report is queried by the financial institution is not less than the second preset threshold, and the number of transactions of the bank account within the past preset time period is less than the third preset threshold; according to the preset matching rules, determine the first type of target large language model that matches the first type of target object with an urgent task urgency; according to the preset matching rules, determine the second type of target large language model that matches the first type of target object with a non-urgent task urgency; according to the preset matching rules, determine the third type of target large language model that matches the second type of target object with an urgent task urgency; according to the preset matching rules, determine the fourth type of target large language model that matches the second type of target object with a non-urgent task urgency.

[0012] Optionally, the response speed of the first type of target large language model is greater than that of the second type of target large language model; the response speed of the third type of target large language model is greater than that of the fourth type of target large language model; the response speed of the third type of target large language model is greater than that of the first type of target large language model; the response speed of the fourth type of target large language model is greater than that of the second type of target large language model.

[0013] According to another aspect of the present application, a large language model financial operation business question and answer device based on an index algorithm is also provided, including: a receiving module, used to receive a query question input by a target object; a first determination module, used to identify the semantic information corresponding to the query question, and determine the identity feature information and behavior feature information of the target object; a second determination module, used to determine a target large language model that matches the semantic information, the identity feature information and the behavior feature information of the target object; an output module, used to search for a document collection in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; searching for document details of the document collection through a forward index, wherein the document details include: document update time, document quality scoring index; determining the frequency of the query keywords in the inverted index and the document update time and document quality scoring index in the forward index as sorting factors; sorting the document collection according to the sorting factors, and using the sorted document collection as the output result of the target large language model.

[0014] According to another aspect of the present application, a non-volatile storage medium is also provided, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned large language model financial operation business question and answer method based on the index algorithm.

[0015] According to another aspect 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 program executes the above-mentioned large language model financial operation business question and answer method based on the index algorithm when running.

[0016] According to another aspect of the present application, a computer program is also provided, wherein when the computer program is executed by a processor, the above-mentioned large language model financial operation business question and answer method based on the index algorithm is implemented.

[0017] According to another aspect of the present application, a computer program product is also provided, which includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned large language model financial operation business question and answer method based on the index algorithm is implemented.

[0018] In the present application, a query question input by a target object is received; semantic information corresponding to the query question is identified, and identity feature information and behavioral feature information of the target object are determined; a target large language model matching the semantic information, identity feature information and behavioral feature information of the target object is determined; a document collection is searched in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; detailed document content of the document collection is searched through a forward index, wherein the detailed document content includes: document update time, document quality score index; the frequency of the query keywords in the inverted index and the document update time and document quality score index in the forward index are determined as sorting factors; the document collection is sorted according to the sorting factors, and the sorted document collection is used as the output result of the target large language model, so as to achieve the purpose of combining the user's semantic information, identity feature information and behavioral feature information to select a specific large language model to process the query question input by the user, thereby achieving the technical effect of efficiently processing different types of query questions, and further solving the technical problem that the related technology cannot combine the user's semantic information, identity feature information and behavioral feature information to select a specific large language model to process the query question input by the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] 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:

[0020] Figure 1 It is a flowchart of a large language model financial operation business question and answer method based on an index algorithm according to an embodiment of the present application;

[0021] Figure 2 It is a structural diagram of a large language model financial operation business question-answering device based on an index algorithm according to an embodiment of the present application;

[0022] Figure 3 It is a hardware structure block diagram of a computer terminal of a large language model financial operation business question and answer method based on an index algorithm according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] 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.

[0024] 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.

[0025] According to an embodiment of the present application, a method embodiment of a large language model financial operation business question and answer method based on an index algorithm is 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.

[0026] Figure 1 is a flowchart of a large language model financial operation business question answering method based on an index algorithm according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0027] Step S102: receiving a query question input by a target object.

[0028] Receive query requests from users, who can enter their questions or needs in natural language through various channels, such as mobile devices, PCs or counter terminals. The system will immediately capture these inputs and prepare for the next step of processing.

[0029] Step S104, identifying semantic information corresponding to the query question, and determining identity feature information and behavior feature information of the target object.

[0030] When receiving a query, we use natural language processing technology, including but not limited to word vectors, attention mechanisms, and Transformer architecture, to perform deep semantic analysis of the question and understand the core intent and context of the question. At the same time, we analyze the user's identity and behavior characteristics to lay the foundation for subsequent model selection and personalized services.

[0031] Step S106, determining a target large language model that matches the semantic information, the identity feature information and the behavior feature information of the target object.

[0032] Among them, behavioral characteristic information includes: the number of times the credit report has been queried by financial institutions, and the number of transactions in the bank account within a preset period of time in the past.

[0033] Based on the semantic information, identity feature information, and behavioral feature information collected and analyzed in the previous step, one or a group of models are intelligently selected from the preset multiple large language models, which are most suitable for the current query scenario and the characteristics of the target object. The selection process involves steps such as model performance evaluation, domain relevance analysis, and personalized parameter adjustment to ensure that the selected model can not only accurately understand the question, but also provide the most appropriate and accurate answer based on the specific needs of the target object.

[0034] Step S108, searching for a document set in the database of the target large language model through the inverted index, wherein the document set includes the query keywords in the query question; searching for document details of the document set through the forward index, wherein the document details include: document update time, document quality score index; determining the frequency of the query keywords in the inverted index and the document update time and document quality score index in the forward index as sorting factors; sorting the document set according to the sorting factors, and using the sorted document set as the output result of the target large language model.

[0035] Once the target large language model is selected, the query question is passed to the model and processed using its powerful natural language understanding and generation capabilities. The model will generate one or more answers based on the semantic information of the question, the characteristics of the target object, and its own knowledge base. These answers not only directly address the questions raised by the user, but also take into account the user's identity, preferences, and background information of the query, thereby providing a more personalized service that meets actual needs. After processing is completed, the output results of the model are presented to the target object, completing a complete query service.

[0036] In the above step S108, firstly, the inverted index technology is used to quickly locate the document set containing the query keywords in the database of the target large language model. The inverted index is an efficient data structure, which associates the keywords in the document with the document ID list containing these keywords, so that all relevant documents can be found instantly when the keyword query is performed, which greatly improves the search speed. Through this technology, the system can quickly identify all documents that match the keywords in the user's query question and build a candidate document set. After obtaining the candidate document set, the forward index technology is further used to find the detailed information of each document, including the update time of the document, the document quality score index and other metadata. The forward index is a data structure with the document ID as the key and the document content, attributes, etc. as the value. It can provide complete information of the document, which is convenient for the system to conduct a comprehensive evaluation of the document later. The frequency of the query keyword in the inverted index, the update time of the document in the forward index and the document quality score index are used as the key factors for sorting. The frequency of query keywords reflects the closeness between the document and the query; the update time of the document ensures the timeliness of the information, especially for frequently changing information such as regulatory policies and internal regulations; the document quality score index ensures the authority and reliability of the search results and avoids the interference of low-quality documents. Finally, the candidate document collection is comprehensively sorted according to the above sorting factors. This process involves a variety of sorting algorithms, such as weight-based sorting, TF-IDF algorithm, etc., to calculate the comprehensive score of the document. Documents with high scores will be displayed to users first to ensure that users receive the most relevant, latest and highest quality information first. The system outputs the sorted document collection as a result and provides it to users for viewing, helping them to quickly and accurately obtain the required knowledge and support business operations and decision-making processes.

[0037] This search strategy that combines inverted index and forward index can not only quickly locate relevant documents, but also ensure the high quality and timeliness of search results, providing users with a smarter and more efficient information retrieval experience. This mechanism fully reflects the application potential of large model technology in banking business scenarios, and provides users with powerful business support through intelligent search and sorting.

[0038] According to the above steps, a query question input by a target object is received; semantic information corresponding to the query question is identified, and identity feature information and behavioral feature information of the target object are determined; a target large language model matching the semantic information, identity feature information and behavioral feature information of the target object is determined; a document collection is searched in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; detailed document content of the document collection is searched through a forward index, wherein the detailed document content includes: document update time, document quality score index; the frequency of the query keywords in the inverted index and the document update time and document quality score index in the forward index are determined as sorting factors; the document collection is sorted according to the sorting factors, and the sorted document collection is used as the output result of the target large language model, thereby achieving the purpose of combining the user's semantic information, identity feature information and behavioral feature information to select a specific large language model to process the query question input by the user, thereby achieving the technical effect of efficiently processing different types of query questions.

[0039] The following Figure 1 The steps shown are exemplary and explanatory.

[0040] According to some optional embodiments of the present application, determining a target large language model that matches the identity feature information and the behavior feature information of the target object can be achieved by the following method: inputting semantic information, the identity feature information and the behavior feature information of the target object into a routing selection model, the routing selection model being used to perform the following steps: performing feature extraction processing on the semantic information to obtain a first feature vector, performing feature extraction processing on the identity feature information of the target object to obtain a second feature vector, and performing feature extraction processing on the behavior feature information of the target object to obtain a third feature vector; merging the first feature vector, the second feature vector and the third feature vector to obtain a target feature vector; inputting the target feature vector into a rule base, wherein the rule base stores pre-set routing rules based on data features and business scenarios, the routing rules are presented in the form of a decision tree, each decision node in the decision tree corresponds to a threshold judgment of a feature dimension, and the branch nodes in the decision tree correspond to different model selection paths; matching the target feature vector according to the routing rules to obtain a target model path corresponding to the target feature vector, wherein the target model path points to one or more target large language models.

[0041] In the above embodiment, the semantic information input by the user is deeply analyzed, key semantic features are extracted, and converted into a vector representation that can be understood by the computer. Technologies such as word embedding and sentence vectors can be used to ensure that the vector can reflect the complex semantics and intentions of the problem. Feature extraction is performed on the identity feature information of the target object, and feature extraction is performed on the behavior feature information of the target object, including historical query records, preference settings, common query types, etc., which are also converted into vector representations. The behavior feature vector reflects the personalized needs and habits of the user. The above three feature vectors are merged to obtain a comprehensive target feature vector. This vector integrates all the information of semantic information, identity features and behavior features, providing a comprehensive basis for subsequent model selection.

[0042] The rule base stores routing rules based on data features and business scenarios, and the rules are presented in the form of a decision tree. Each decision node in the decision tree is set with a threshold judgment for a specific feature dimension, such as the complexity of the problem, the role level of the user, etc. The target feature vector is input into the decision tree of the rule base and matched according to the preset routing rules. The branch nodes in the decision tree correspond to different model selection paths, which indicate one or more large language models that should be selected based on the feature vector matching results. Ultimately, the decision tree will output the path information of one or more target large language models. This step ensures that the most appropriate model is selected to handle the problem based on the characteristics, identity, and historical behavior of the user's query. If it is a path of multiple models, the system may further use ensemble learning or model fusion technology to achieve a better response based on the specific needs of the problem.

[0043] Furthermore, after completing the route selection, the routing selection model sends the identification information of the target large language model to the task planning module; after receiving the identification information, the task planning module detects the resource utilization efficiency of the target large language model corresponding to the identification information, wherein the resource utilization efficiency includes: CPU utilization rate and memory occupancy rate; when it is detected that the CPU utilization rate of the target large language model corresponding to the identification information is less than a first preset percentage and / or the memory occupancy rate is less than a second preset percentage, the task planning module sends a target instruction to the routing selection model, wherein the target instruction is used to instruct the routing selection model to redetermine the routing rules.

[0044] It is understandable that after receiving the target large language model identification information sent by the routing model, the task planning module's primary task is to monitor the resource utilization efficiency of the selected model. This includes real-time detection of CPU usage and memory occupancy, and evaluation of the hardware resource consumption of the model in the current task execution. The first preset percentage and the second preset percentage are thresholds pre-set according to system performance, hardware specifications, and business requirements, and are used to determine whether the resource utilization of the model is in an ideal state.

[0045] If the test results show that the CPU usage of the target large language model is lower than the first preset percentage and / or the memory usage is lower than the second preset percentage, this means that the selected model has low resource consumption when processing the current task, and there may be insufficient resource utilization. In this case, the task planning module will send a "target instruction" to the routing selection model, instructing it to re-evaluate the routing rules to select a more appropriate model to ensure efficient use of resources. The sending of the target instruction prompts the routing selection model to restart the selection process, dynamically adjust the model selection based on the latest resource status and business needs, and optimize resource allocation.

[0046] By monitoring the resource usage of the model in real time and dynamically adjusting the model selection, we can avoid the waste of resources caused by fixed model selection and ensure the efficiency and flexibility of the system in resource utilization. By selecting the model that best suits the current needs and resource status, we can provide faster and more accurate responses, thereby significantly improving the user's satisfaction and experience during the query process.

[0047] Furthermore, sample data is extracted from the historical data repository at a preset time interval, and the routing selection model is trained using the feature vector of the sample data as input and the corresponding target model path as a label.

[0048] It is worth noting that the system automatically extracts sample data from the historical data repository at preset time intervals. These sample data include historical query questions, corresponding query results, user feedback, and information about the selected large language model. The extraction of sample data not only takes into account the distribution in time series, but also ensures the diversity and representativeness of the data, covering various business scenarios and user feature combinations. For each sample data extracted, the system will reconstruct its feature vector, including the feature vector of semantic information, the feature vector of identity feature information, and the feature vector of behavioral feature information. These feature vectors comprehensively reflect the entire background information of the historical query. At the same time, based on the "target large language model" information actually used by the sample data in the historical query process, the system will set a corresponding label for each sample data, namely the target model path. This label indicates which large language model or model combination has proven to be the best choice under a specific feature vector. Using these labeled feature vector data, the system begins to train the routing model. During the training process, the model continuously learns and adjusts to improve its ability to predict the correct target model path when given a feature vector. Through training with a large amount of historical data, the routing model can gradually master the association rules between different feature vectors and the target model path, optimize its decision logic, and enable it to more accurately and quickly select the most suitable large language model to handle the query when facing a new user query. After training, the routing model will be evaluated for performance to check its prediction accuracy and resource efficiency on new data. If the model's performance does not meet the expected standards, or there is a deviation in the selection in certain specific scenarios, the system will automatically adjust the training strategy, introduce new sample data or adjust the way the feature vector is constructed, and retrain to achieve iterative optimization of the model to ensure its continuous improvement to adapt to changing business needs and user behavior.

[0049] The above-mentioned routing selection model training mechanism based on historical data can not only ensure that the decision-making process of the system to select a large language model is based on a large amount of actual business data and user feedback, but also continuously improve the efficiency and accuracy of routing selection through continuous learning and optimization, providing a more intelligent, personalized and efficient service experience for the bank and beyond.

[0050] According to some other optional embodiments of the present application, determining the target large language model that matches the semantic information, the identity feature information and the behavior feature information of the target object can be achieved by the following method: determining the task urgency of the query question according to the semantic information, wherein the task urgency includes: urgent, not urgent; determining the target object whose identity feature information and behavior feature information meet at least one of the following preset conditions as the first category of target objects: the number of credit accounts is less than a first preset threshold, the number of times the credit report is queried by the financial institution is less than a second preset threshold, and the number of transactions of the bank account within the past preset time period is less than a third preset threshold; determining the target object whose identity feature information and behavior feature information meet at least one of the following preset conditions as the second category of target objects Class target objects: the number of credit accounts is not less than a first preset threshold, the number of times the credit report is queried by a financial institution is not less than a second preset threshold, and the number of transactions of a bank account within a preset period of time in the past is less than a third preset threshold; according to the preset matching rules, a first class target large language model matching the first class target object with an urgent task urgency is determined; according to the preset matching rules, a second class target large language model matching the first class target object with a non-urgent task urgency is determined; according to the preset matching rules, a third class target large language model matching the second class target object with an urgent task urgency is determined; according to the preset matching rules, a fourth class target large language model matching the second class target object with a non-urgent task urgency is determined.

[0051] Preferably, the response speed of the first type of target large language model is greater than that of the second type of target large language model; the response speed of the third type of target large language model is greater than that of the fourth type of target large language model; the response speed of the third type of target large language model is greater than that of the first type of target large language model; the response speed of the fourth type of target large language model is greater than that of the second type of target large language model.

[0052] In the above embodiment, first, the task urgency of the query question is evaluated based on semantic information and classified as "urgent" or "not urgent". This classification can be based on the keywords, tone, or contextual information related to time sensitivity contained in the question. For example, a query involving instant transaction confirmation or urgent compliance consultation will be marked as "urgent", while a query about business process guidance or general policy interpretation may be marked as "not urgent".

[0053] Next, the system divides the target objects (i.e., the query initiators) into "first-category target objects" and "second-category target objects." This classification is based on preset conditions such as the number of credit accounts of the target objects, the number of times their credit reports have been queried, and the transaction frequency of their bank accounts. Specifically: First-category target objects: users with a small number of credit accounts, a low number of credit report inquiries, and a low frequency of recent transactions. Such users may not be highly dependent on banking services, or their account activities are relatively small. Second-category target objects: users with a large number of credit accounts, a high number of credit report inquiries, or a high frequency of recent transactions. Such users may have a higher demand for banking services, have frequent account activities, and have high expectations for timely responses.

[0054] According to the preset matching rules, the corresponding types of large language models are matched for different types of users and task urgency. The selection of these models takes into account the response speed and resource utilization efficiency to ensure the efficiency of the service: The first type of target large language model: matched with urgent tasks and the first type of target objects. This type of model is characterized by a fast response speed and can quickly process and reply to urgent queries, but there may be a trade-off in resource usage and the ability to handle complex problems. The second type of target large language model: matched with non-urgent tasks and the first type of target objects. This type of model has a slower response speed, but may achieve a better balance between processing power and resource efficiency. The third type of target large language model: matched with urgent tasks and the second type of target objects. This type of model not only has a fast response speed, but also has a high ability to handle complex problems. It is suitable for high-demand users and emergency scenarios. The fourth type of target large language model: matched with non-urgent tasks and the second type of target objects. This type of model finds a balance between response speed, resource efficiency and processing power, and is suitable for handling daily queries from the second type of target objects.

[0055] Figure 2 is a structural diagram of a large language model financial operation business question-answering device based on an index algorithm according to an embodiment of the present application, such as Figure 2 As shown, the device comprises:

[0056] The receiving module 22 is used to receive a query question input by the target object.

[0057] The first determination module 24 is used to identify the semantic information corresponding to the query question and determine the identity feature information and behavior feature information of the target object.

[0058] The second determination module 26 is used to determine a target large language model that matches the semantic information, the identity feature information and the behavior feature information of the target object.

[0059] The output module 28 is used to search for a document set in the database of the target large language model through an inverted index, wherein the document set includes the query keywords in the query question; search for detailed document content of the document set through a forward index, wherein the detailed document content includes: document update time, document quality score index; determine the frequency of the query keywords in the inverted index and the document update time and document quality score index in the forward index as sorting factors; sort the document set according to the sorting factors, and use the sorted document set as the output result of the target large language model.

[0060] Optionally, the second determination module 26 is also used to perform the following steps: inputting semantic information, identity feature information and behavior feature information of the target object into a routing selection model, the routing selection model is used to perform the following steps: performing feature extraction processing on the semantic information to obtain a first feature vector, performing feature extraction processing on the identity feature information of the target object to obtain a second feature vector, and performing feature extraction processing on the behavior feature information of the target object to obtain a third feature vector; merging the first feature vector, the second feature vector and the third feature vector to obtain a target feature vector; inputting the target feature vector into a rule base, wherein the rule base stores pre-set routing rules based on data features and business scenarios, the routing rules are presented in the form of a decision tree, each decision node in the decision tree corresponds to a threshold judgment of a feature dimension, and the branch nodes in the decision tree correspond to different model selection paths; matching the target feature vector according to the routing rules to obtain a target model path corresponding to the target feature vector, wherein the target model path points to one or more target large language models.

[0061] Optionally, the second determination module 26 is further used to perform the following steps: after completing the route selection, the routing selection model sends the identification information of the target large language model to the task planning module; after receiving the identification information, the task planning module detects the resource utilization efficiency of the target large language model corresponding to the identification information, wherein the resource utilization efficiency includes: CPU utilization rate and memory occupancy rate; when it is detected that the CPU utilization rate of the target large language model corresponding to the identification information is less than a first preset percentage and / or the memory occupancy rate is less than a second preset percentage, the task planning module sends a target instruction to the routing selection model, wherein the target instruction is used to instruct the routing selection model to redetermine the routing rules.

[0062] Optionally, the second determination module 26 is further configured to perform the following steps: extracting sample data from the historical data repository at preset time intervals, taking the feature vector of the sample data as input and the corresponding target model path as a label, and training the routing selection model.

[0063] Optionally, the second determination module 26 is further used to perform the following steps: determine the task urgency of the query question based on the semantic information, wherein the task urgency includes: urgent, not urgent; determine the target object whose identity characteristic information and behavior characteristic information meet at least one of the following preset conditions as the first type of target object: the number of credit accounts is less than a first preset threshold, the number of times the credit report is queried by the financial institution is less than a second preset threshold, and the number of bank account transactions within the past preset time period is less than a third preset threshold; determine the target object whose identity characteristic information and behavior characteristic information meet at least one of the following preset conditions as the second type of target object: the number of credit accounts is not less than the first preset threshold , the number of times the credit report is queried by the financial institution is not less than the second preset threshold, and the number of transactions of the bank account within the past preset time period is less than the third preset threshold; according to the preset matching rules, determine the first type of target large language model that matches the first type of target object with an urgent task urgency; according to the preset matching rules, determine the second type of target large language model that matches the first type of target object with a non-urgent task urgency; according to the preset matching rules, determine the third type of target large language model that matches the second type of target object with an urgent task urgency; according to the preset matching rules, determine the fourth type of target large language model that matches the second type of target object with a non-urgent task urgency.

[0064] Optionally, the response speed of the first type of target large language model is greater than that of the second type of target large language model; the response speed of the third type of target large language model is greater than that of the fourth type of target large language model; the response speed of the third type of target large language model is greater than that of the first type of target large language model; the response speed of the fourth type of target large language model is greater than that of the second type of target large language model.

[0065] It should be noted that the above Figure 2 Each module in the can be a program module (for example, a set of program instructions that implement a specific function) or a hardware module. 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 modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0066] It should be noted that Figure 2 The preferred implementation of the illustrated embodiment can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0067] Figure 3 The hardware structure block diagram of a computer terminal for implementing a large language model financial operation business question-answering method based on an index algorithm is shown. Figure 3As shown, the computer terminal 30 may include one or more (302a, 302b, ..., 302n are used to illustrate) processors 302 (the processor 302 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission module 306 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 3 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 3 More or fewer components as shown, or with Figure 3 Different configurations are shown.

[0068] It should be noted that the one or more processors 302 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 30. 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).

[0069] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the large language model financial operation business question-answering method based on the index algorithm in the embodiment of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, the large language model financial operation business question-answering method based on the index algorithm is realized. The memory 304 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 instances, the memory 304 may further include a memory remotely arranged relative to the processor 302, and these remote memories may be connected to the computer terminal 30 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.

[0070] The transmission module 306 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 30. In one example, the transmission module 306 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 module 306 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0071] 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 30 .

[0072] It should be noted that, in some optional embodiments, the above Figure 3 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 3 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0073] It should be noted that Figure 3 The computer terminal shown is used to execute Figure 1 The large language model financial operation business question and answer method based on the index algorithm is shown, so the relevant explanations in the execution method of the above command are also applicable to the electronic device and will not be repeated here.

[0074] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned large language model financial operation business question and answer method based on the index algorithm.

[0075] A program for a non-volatile storage medium to perform the following functions: receiving a query question input by a target object; identifying semantic information corresponding to the query question, and determining identity feature information and behavior feature information of the target object; determining a target large language model that matches the semantic information, identity feature information and behavior feature information of the target object; searching for a document collection in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; searching for document details of the document collection through a forward index, wherein the document details include: document update time, document quality score index; determining the frequency of the query keywords in the inverted index and the document update time and document quality score index in the forward index as sorting factors; sorting the document collection according to the sorting factors, and using the sorted document collection as the output result of the target large language model.

[0076] An embodiment of the present application also provides an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the program executes the above-mentioned large language model financial operation business question and answer method based on the index algorithm when running.

[0077] The processor is used to run a program that performs the following functions: receiving a query question input by a target object; identifying semantic information corresponding to the query question, and determining identity feature information and behavior feature information of the target object; determining a target large language model that matches the semantic information, identity feature information and behavior feature information of the target object; searching for a document collection in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; searching for document details of the document collection through a forward index, wherein the document details include: document update time, document quality score index; determining the frequency of the query keywords in the inverted index and the document update time and document quality score index in the forward index as sorting factors; sorting the document collection according to the sorting factors, and using the sorted document collection as the output result of the target large language model.

[0078] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0079] 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.

[0080] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0081] 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.

[0082] 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.

[0083] 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.

[0084] 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.

[0085] 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 large language model financial operation business question answering method based on index algorithm, characterized in that: include: Receive a query question input by a target object; Identify the semantic information corresponding to the query question, and determine the identity feature information and behavior feature information of the target object; Determining a target large language model that matches the semantic information, the identity feature information, and the behavior feature information of the target object, including: inputting the semantic information, the identity feature information, and the behavior feature information of the target object into a routing selection model, wherein the routing selection model is used to perform the following steps: performing feature extraction processing on the semantic information to obtain a first feature vector, performing feature extraction processing on the identity feature information of the target object to obtain a second feature vector, and performing feature extraction processing on the behavior feature information of the target object to obtain a third feature vector; merging the first feature vector, the second feature vector, and the third feature vector to obtain a target feature vector; inputting the target feature vector into a rule base, wherein the rule base stores pre-set routing rules based on data features and business scenarios, wherein the routing rules are presented in the form of a decision tree, wherein each decision node in the decision tree corresponds to a threshold judgment of a feature dimension, and the branch nodes in the decision tree correspond to different model selection paths; matching the target feature vector according to the routing rule to obtain a target model path corresponding to the target feature vector, wherein the target model path points to one or more target large language models; A document collection is searched in a database in the target large language model through an inverted index, wherein the document collection includes query keywords in the query question; detailed document content of the document collection is searched through a forward index, wherein the detailed document content includes: document update time, document quality score index; the frequency of the query keyword in the inverted index and the document update time and the document quality score index in the forward index are determined as sorting factors; the document collection is sorted according to the sorting factors, and the sorted document collection is used as the output result of the target large language model.

2. The method according to claim 1, characterized in that The method further comprises: After completing the route selection, the routing selection model sends the identification information of the target large language model to the task planning module; After receiving the identification information, the task planning module detects the resource utilization efficiency of the target large language model corresponding to the identification information, wherein the resource utilization efficiency includes: CPU utilization rate and memory occupancy rate; When it is detected that the CPU usage of the target large language model corresponding to the identification information is less than a first preset percentage and / or the memory occupancy is less than a second preset percentage, the task planning module sends a target instruction to the routing selection model, wherein the target instruction is used to instruct the routing selection model to redetermine the routing rules.

3. The method according to claim 1, characterized in that The method further comprises: Sample data is extracted from a historical data repository at a preset time interval, and the routing model is trained using a feature vector of the sample data as input and a corresponding target model path as a label.

4. The method according to claim 1, characterized in that: Determining a target large language model that matches the semantic information, the identity feature information, and the behavior feature information of the target object includes: Determine the task urgency of the query question according to the semantic information, wherein the task urgency includes: urgent and not urgent; The target object whose identity characteristic information and behavior characteristic information meet at least one of the following preset conditions is determined as a first type of target object: the number of credit accounts is less than a first preset threshold, the number of times the credit report is queried by a financial institution is less than a second preset threshold, and the number of bank account transactions in the past preset period of time is less than a third preset threshold; The target object whose identity characteristic information and behavior characteristic information meet at least one of the following preset conditions is determined as a second type of target object: the number of credit accounts is not less than the first preset threshold, the number of times the credit report is queried by the financial institution is not less than the second preset threshold, and the number of transactions of the bank account in the past preset period of time is less than the third preset threshold; According to a preset matching rule, determining a first-category target large language model that matches a first-category target object whose task urgency is urgent; According to the preset matching rule, determining a second type of target large language model that matches the first type of target object whose task urgency is not urgent; According to a preset matching rule, determining a third-category target large language model that matches the second-category target object whose task urgency is urgent; According to the preset matching rule, a fourth target large language model matching the second target object whose task urgency is not urgent is determined.

5. The method according to claim 4, characterized in that The response speed of the first target large language model is greater than that of the second target large language model; the response speed of the third target large language model is greater than that of the fourth target large language model; the response speed of the third target large language model is greater than that of the first target large language model; the response speed of the fourth target large language model is greater than that of the second target large language model.

6. A large language model financial operation business question-answering device based on an index algorithm, characterized in that: include: A receiving module, used for receiving a query question input by a target object; A first determination module is used to identify semantic information corresponding to the query question and determine identity feature information and behavior feature information of the target object; The second determination module is used to determine a target large language model that matches the semantic information, the identity feature information of the target object, and the behavior feature information, including: inputting the semantic information, the identity feature information of the target object, and the behavior feature information into a routing selection model, wherein the routing selection model is used to perform the following steps: performing feature extraction processing on the semantic information to obtain a first feature vector, performing feature extraction processing on the identity feature information of the target object to obtain a second feature vector, and performing feature extraction processing on the behavior feature information of the target object to obtain a third feature vector; merging the first feature vector, the second feature vector, and the third feature vector to obtain a target feature vector; inputting the target feature vector into a rule base, wherein the rule base stores pre-set routing rules based on data features and business scenarios, wherein the routing rules are presented in the form of a decision tree, wherein each decision node in the decision tree corresponds to a threshold judgment of a feature dimension, and the branch nodes in the decision tree correspond to different model selection paths; matching the target feature vector according to the routing rule to obtain a target model path corresponding to the target feature vector, wherein the target model path points to one or more target large language models; An output module is used to search for a document set in a database in the target large language model through an inverted index, wherein the document set includes query keywords in the query question; search for document details of the document set through a forward index, wherein the document details include: document update time, document quality score index; determine the frequency of the query keywords in the inverted index and the document update time and the document quality score index in the forward index as sorting factors; sort the document set according to the sorting factors, and use the sorted document set as the output result of the target large language model.

7. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the large language model financial operation business question and answer method based on the index algorithm as described in any one of claims 1 to 5.

8. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program, when running, executes the large language model financial operation business question and answer method based on an index algorithm as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the large language model financial operation business question and answer method based on an index algorithm as described in any one of claims 1 to 5 is implemented.

Images