Language text system co-processing method and device and storage medium

Abstract

The invention discloses a language text system co-processing method and device and a storage medium, and relates to the field of financial science and technology and the field of artificial intelligence. The method comprises the following steps: receiving a natural language text input by a target object through a user interface; determining a system access intention of the target object according to the natural language text; querying from a knowledge graph according to the system access intention to obtain target information; the system access intention and the target information are converted into a probability distribution matrix through the target agent, and the probability distribution matrix represents the probability that each service system is used for processing the natural language text; and determining Y target systems from the X service systems to cooperatively process the natural language text according to the probability distribution matrix. The technical problem that in the prior art, when a plurality of independent systems are used for processing natural language text requests submitted by a user, due to the fact that the cooperation efficiency between the systems is low, the natural language text processing time is long is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and the field of financial technology, in particular, to a language text system collaborative processing method and device and storage medium. BACKGROUND

[0002] In the existing technical framework of the financial field, when processing natural language text requests submitted by employees or customers, significant efficiency challenges are faced. Due to the complexity and diversity of business needs, banks and other financial institutions often rely on multiple independent systems to manage different business processes, such as credit management, payment settlement, etc. However, the information silo phenomenon between these systems seriously limits the collaborative efficiency, resulting in a significant extension of natural language processing time when facing cross-system requests.

[0003] For example, when an employee or customer submits a request involving multiple business areas in the form of natural language, the existing system has difficulty in quickly and accurately understanding the full connotation of the request, and thus cannot seamlessly interface to the relevant business modules to achieve efficient processing. This problem not only increases the complexity and processing time of the operation, but also may cause process errors and data security risks, seriously affecting the operational efficiency and customer satisfaction of the bank. Therefore, solving the problem of low system collaboration efficiency is of great significance for optimizing the natural language text request processing process and improving the intelligent level of bank business.

[0004] At present, there is no effective solution to the above problems. SUMMARY

[0005] The embodiments of the present application provide a language text system collaborative processing method, device and storage medium, to at least solve the technical problem of long natural language text processing time due to low system collaboration efficiency when using multiple independent systems to process natural language text requests submitted by users in the prior art.

[0006] According to an aspect of an embodiment of the present application, a language text system collaborative processing method is provided, comprising: receiving a natural language text input by a target object through a user interface; determining a system access intention of the target object according to the natural language text; querying target information from a knowledge graph according to the system access intention, wherein the target information is information with an association degree greater than a preset association degree with the system access intention; converting the system access intention and the target information into a probability distribution matrix through a target agent, wherein the probability distribution matrix represents the probability of each business system being used to process the natural language text; determining Y target systems from X business systems for collaborative processing of the natural language text according to the probability distribution matrix, wherein X and Y are both integers greater than 1, and X is greater than or equal to Y.

[0007] Optionally, the training process of the target agent comprises the following steps:

[0008] obtaining a training set, wherein the training set comprises historical work orders submitted by users and business system information called when processing the historical work orders; performing a plurality of training operations on the neural network according to the training set until the neural network enters a convergent state, to obtain the target agent; wherein each training operation comprises: predicting, by the neural network, a business system to be called according to a historical work order, and when detecting that a correctness rate of a prediction result is less than a preset threshold, updating a function of the neural network and entering a next training operation; and when detecting that the correctness rate of the prediction result is greater than or equal to the preset threshold, determining that the neural network enters the convergent state.

[0009] Optionally, each training operation comprises the following steps:

[0010] extracting, by an input layer of the neural network, N-dimensional natural language features from each historical work order in the training set, wherein N is an integer greater than 1; performing feature compression on the natural language features output by the input layer through a fully connected layer of the neural network, and establishing a function relationship between the feature compression result and the business system information in the training set; converting, by an output layer of the neural network, the function relationship output by the fully connected layer into probability distribution information, wherein the probability distribution information is used to represent probabilities corresponding to calling each business system according to the feature compression result; and determining a correctness rate of the neural network when predicting a business system to be called according to a historical work order according to the probability distribution information.

[0011] Optionally, after determining Y target systems from X business systems to cooperatively process the natural language text according to the probability distribution matrix, detecting a processing time length of the target systems for the natural language text, a processing success rate of the target systems for the natural language text, and a satisfaction degree of a target object for a processing result of the natural language text; determining a reward coefficient according to the processing time length of the target systems for the natural language text, the processing success rate of the target systems for the natural language text, and the satisfaction degree of the target object for the processing result of the natural language text; and adjusting the target agent according to the reward coefficient.

[0012] Optionally, before querying the target information from the knowledge graph according to the system access intention, classifying data in a database of each business system, and inducing data of the same type into a data entity; detecting a similarity between the data entity and each node in the knowledge graph; and associating the data entity with a node with the highest similarity in the knowledge graph.

[0013] Optionally, querying the target information from the knowledge graph according to the system access intention comprises: determining a target node from the knowledge graph according to a semantic similarity between the system access intention and each node in the knowledge graph; and determining the target information according to information of the target node itself and a data entity associated with the target node.

[0014] Optionally, after determining Y target systems to cooperatively process the natural language text from the X business systems according to the probability distribution matrix, at least one of the following M types of information is collected and displayed on the user interface, where the M types of information include: jump link information between the Y target systems; processing results of the natural language text; processing operation information of each of the Y target systems on the natural language text.

[0015] According to another aspect of the embodiments of the present application, a system cooperative processing device for language text is further provided. The device comprises: a receiving unit configured to receive a natural language text input by a target object through a user interface; a determining unit configured to determine a system access intention of the target object according to the natural language text; a querying unit configured to query target information from a knowledge graph according to the system access intention, where the target information is information having an association degree greater than a preset association degree with the system access intention; a first processing unit configured to convert the system access intention and the target information into a probability distribution matrix through a target intelligent agent, where the probability distribution matrix represents a probability of each business system being used to process the natural language text; and a second processing unit configured to determine Y target systems to cooperatively process the natural language text from X business systems according to the probability distribution matrix, where X and Y are both integers greater than 1, and X is greater than or equal to Y.

[0016] According to another aspect of the embodiments of the present application, a computer readable storage medium is further provided. The computer readable storage medium stores a computer program. When the computer program is executed, the computer readable storage medium causes a device where the computer readable storage medium is located to perform the system cooperative processing method for language text.

[0017] According to another aspect of the embodiments of the present application, an electronic device is further provided. The electronic device comprises one or more processors and a memory. The memory is configured to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the system cooperative processing method for language text.

[0018] According to another aspect of the embodiments of the present application, a computer program product is further provided. The computer program product comprises a computer program or instructions. When the computer program or instructions are executed by a processor, the system cooperative processing method for language text is implemented.

[0019] In the present application, a system collaborative processing method for language text is proposed. After receiving a natural language text input by a target object through a user interface, the system access intention of the target object is first determined according to the natural language text, and then the target information is obtained from the knowledge graph according to the system access intention, wherein the target information is information with an association degree greater than a preset association degree with the system access intention. Subsequently, the system access intention and the target information are converted into a probability distribution matrix by a target agent, wherein the probability distribution matrix represents the probability of each business system being used to process the natural language text. Finally, Y target systems are determined from X business systems to collaboratively process the natural language text according to the probability distribution matrix, wherein X and Y are both integers greater than 1, and X is greater than or equal to Y.

[0020] From the above, according to the technical solution of the present application, after determining the user intention implied in the natural language text, the target information highly relevant to the intention is efficiently queried from the knowledge graph. This process avoids the tedious steps of manual searching and inputting information, significantly shortening the time from text input to information acquisition. After determining the system access intention of the user, the target agent can convert the intention and the queried target information into a probability distribution matrix, which intuitively presents the possibility of different business systems processing a specific request. Based on this, the most suitable combination of business systems can be automatically identified and selected for collaborative processing, without the need for manual judgment and manual switching, greatly optimizing the efficiency of multi-system collaborative processing.

[0021] Since the present method can intelligently identify user intention, quickly retrieve information and automatically assign to the most suitable system for processing, the processing time of natural language text requests can be significantly shortened. Compared with the traditional way, the user does not need to wait for data transmission between systems and manual confirmation, and the entire processing flow is more smooth and efficient.

[0022] Therefore, the system collaborative processing method of the present application effectively solves the problem of system collaboration in processing natural language text requests in the banking industry, and achieves the technical effect of improving the processing efficiency of natural language text requests. BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings, which are included to provide a further understanding of the present application, constitute a part of the present application and illustrate the illustrative embodiments of the present application and their description serve to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0024] Figure 1 is a flowchart of an optional system collaborative processing method for language text according to an embodiment of the present application;

[0025] Figure 2is a flowchart of an optional training method of a target agent according to an embodiment of the present application;

[0026] Figure 3 is a schematic diagram of an optional system cooperative processing device of language text according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative work should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to the process, method, product or device.

[0029] It should also be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) collected by the present application are information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards in relevant regions, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for user to choose authorization or refusal. For example, an interface is provided between the system and the related users or agencies, and before obtaining the relevant information, the interface needs to send a request to the aforementioned user or agency, and after receiving the consent information feedback from the aforementioned user or agency, the relevant information is obtained.

[0030] According to the embodiment of the present application, an embodiment of a system collaborative processing method of language text is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from here.

[0031] According to the embodiment of the present application, an intelligent office assistant service can be used as the execution subject of the system collaborative processing method of language text, which can be a software system or a combination of software and hardware embedded system. Of course, the method execution subject in the present application can also be other forms of execution subject, such as device, equipment, etc. Those skilled in the art should know that the present application does not particularly limit the specific form of the method execution subject.

[0032] Figure 1 is a flowchart of an optional system collaborative processing method of language text according to the embodiment of the present application, as shown in Figure 1 The method comprises the following steps:

[0033] Step S101, receiving natural language text input by a target object through a user interface.

[0034] Optionally, in the present application, the target object can be an employee within the bank or a customer directly interacting with the bank. The target object is the direct recipient of the intelligent office assistant service, and the target object can submit requirements, query information or operation requests in the form of natural language.

[0035] It should be noted that all user data including natural language text have been authorized by the user for use. The authorization can be performed by the user checking the agreement terms, signing the authorization letter, etc. when registering or using the intelligent office assistant service. Authorization is the basis for the legality and compliance of data collection. In addition, the target object can retrieve its own authorization at any time, and the system needs to delete the information of the target object after receiving the authorization revocation application of the target object.

[0036] Optionally, the user interface is a visual or operation interface for the intelligent office assistant service to interact with the target object, which can be a graphical interface (such as window, button, drop-down menu, etc.) or a text interface (such as command line interface, chat window, etc.). The purpose of user interface design is to provide a friendly and intuitive operation environment, so that users can easily submit instructions or requests through the interface.

[0037] Optionally, natural language text refers to the form of language used in daily communication, such as Chinese, English, etc. Unlike programming languages or specific command formats, natural language has rich context and expression ability, and can express complex intentions and information. In the intelligent office assistant service, users can use natural language text to ask questions, describe requirements or issue instructions without the need to learn specific command formats, greatly reducing the use threshold.

[0038] Step S102, determining the system access intention of the target object according to the natural language text.

[0039] Optionally, after receiving the natural language text input by the target object (such as bank staff or customers) through the user interface, the intelligent office assistant service can use natural language processing technology to analyze and understand the text content. This process aims to parse the real needs or intentions of the user from the text, i.e. the "system access intention", which can be querying specific business information, submitting a certain business application, seeking guidance for a certain operation, etc.

[0040] Natural language processing technology can accurately capture key information in the text through semantic analysis, entity recognition, and intention classification, and understand the specific bank business system or function that the user wants to access or operate. For example, when the user asks "how to apply for a personal loan", the intelligent assistant can identify the intention of "personal loan application" and quickly locate the relevant business system to provide accurate guidance or operation links for the user, thereby improving the speed and accuracy of service response and simplifying the process of finding answers in multiple independent systems.

[0041] Step S103, querying the target information from the knowledge graph according to the system access intention, wherein the target information is information with an association degree greater than a preset association degree with the system access intention.

[0042] Optionally, in the intelligent office assistant service, after determining the "system access intention" of the target object, the service will start the knowledge graph query mechanism to obtain target information highly related to the intention. Knowledge graph is a structured data storage form that not only stores information but also depicts the association and relationship between information. For the banking industry, the knowledge graph covers various business rules, operation processes, product information, etc., forming a rich information network.

[0043] Optionally, the intelligent office assistant can search for information in the knowledge graph that has a higher correlation with the intent than a pre-set threshold by converting the system access intent into query conditions. The "pre-set correlation" here is a standard for measuring information relevance, and the system sets this threshold according to semantic similarity, business logic correlation, and other factors to ensure that the returned information is the most targeted and useful. For example, if the intent is "cross-border remittance process consultation", the intelligent office assistant will look for information that is directly related to cross-border remittance or indirectly related (such as relevant regulations, required materials, etc.), rather than irrelevant personal savings operation guidelines.

[0044] The above query process makes full use of the deep correlation of the knowledge graph, can quickly and accurately locate the target information, avoids the problems of information overload and irrelevant results in traditional search, provides a direct and fast path to the required knowledge for the user, significantly improves the efficiency and accuracy of information acquisition, and further optimizes the bank's internal business processes and customer experience.

[0045] Step S104, converting the system access intent and the target information into a probability distribution matrix through the target agent, wherein the probability distribution matrix represents the probability of each business system being used to process natural language text.

[0046] Optionally, when the intelligent office assistant service understands the system access intent of the target object and obtains a series of highly relevant target information from the knowledge graph, the target agent can map these intents and information into a probability distribution matrix, where each element of the matrix represents the probability of a specific business system in the bank being used to process the current natural language text request. The construction of the matrix is based on the correlation strength between the target information and each business system, as well as the prediction results of historical data and learning algorithms.

[0047] The probability distribution matrix takes into account the complexity and diversity of bank business, and each business system has the potential to be the best choice for processing a specific request. For example, for a request to "query financial product yield", the agent will not only consider the information management system of the financial product, but also consider the customer relationship management system, risk assessment system, etc., because these systems may also contain key data that affect the yield of the financial product. Each probability value in the matrix is calculated by the agent, reflecting the agent's evaluation of the matching degree between the system and the request. A system with a higher probability value means it is more suitable as the current request handler, while a lower probability value indicates that the system has a lower correlation with the request. In this way, the agent can dynamically recommend the best system combination for each request, thereby achieving fast and accurate business processing.

[0048] The above process not only saves the time of employees in manually searching and selecting between multiple systems, but also ensures the optimal matching between business requests and processing systems through probabilistic evaluation, thereby improving processing efficiency and accuracy.

[0049] Step S105, determining Y target systems for collaborative processing of the natural language text from the X business systems according to the probability distribution matrix, wherein X and Y are both integers greater than 1, and X is greater than or equal to Y.

[0050] Optionally, the probability distribution matrix assigns a value to the likelihood of each business system (a total of X) being selected to process a specific request through quantitative analysis. These probability values are obtained by the agent through algorithm processing based on system access intent and target information, reflecting the relevance and suitability of a business system to the current demand. In the matrix, business systems with higher probabilities are considered more suitable as the main participants in this processing.

[0051] The intelligent office assistant service uses the matrix to select the most suitable business systems (a total of Y) for collaborative processing. The selection of Y depends on the complexity of the specific demand and the processing requirements. For example, for a request involving cross-border payment and foreign exchange, the payment system, exchange rate system, and compliance review system (assuming X = 10, Y = 3) may need to work together; while for a relatively simple account balance inquiry, only a single account management system may be needed (Y = 1 at this time).

[0052] Through dynamic evaluation and intelligent decision-making, the intelligent office assistant service can quickly locate the most relevant Y systems among multiple business systems, achieving efficient allocation of resources and high-efficiency collaboration. Such a mechanism not only greatly reduces the time for manual judgment and system switching, but also ensures that each processing is the optimal solution, thereby improving the overall business processing speed and quality.

[0053] In an optional embodiment, Figure 2 is a flowchart of an optional training method of a target agent according to an embodiment of the present application, as Figure 2 shown, comprising the following steps:

[0054] Step S201, obtaining a training set, wherein the training set includes historical work orders submitted by users and business system information called when processing the historical work orders.

[0055] Step S202, performing multiple training operations on the neural network according to the training set until the neural network enters a convergent state, obtaining the target agent.

[0056] In each training operation, the neural network is used to predict the business system to be called based on historical work orders. If the accuracy of the prediction result is less than a preset threshold, the function of the neural network is updated and the next training operation is entered. If the accuracy of the prediction result is greater than or equal to the preset threshold, it is determined that the neural network enters a convergent state.

[0057] Optionally, first, the intelligent office assistant service builds or obtains a training set containing work order records submitted by bank employees or users in the past and the business system information called when processing the work orders. The work order records include the original demand description of the user, the system access intention, and the final processing result and the called business system.

[0058] Then, the intelligent office assistant service trains the neural network model multiple times based on the training set. In each training cycle, the agent attempts to predict the business system required to process a historical work order through the neural network. The model analyzes the natural language text in the work order, converts it into a numerical feature vector, and then inputs it into the neural network for forward propagation to predict the business system to be called.

[0059] The prediction result of the model is then compared with the actual business system call to check whether the accuracy of the prediction meets a preset threshold. If the accuracy is below the threshold, it means that the prediction of the model is not accurate enough, and the weights and biases of the neural network need to be adjusted through the backpropagation algorithm to improve its prediction performance. This process is repeated until the accuracy of the prediction result reaches or exceeds the preset threshold, and the model is considered to have entered a convergent state.

[0060] The convergent state is an important node in the training of the neural network, indicating that the model has been fully trained and has learned how to accurately infer the business system access intention from natural language text and make reasonable system call predictions. The neural network that has reached the convergent state, i.e., the "target agent" mentioned above, will be used in the intelligent office assistant service to provide instant and accurate business guidance and support to users.

[0061] In an optional embodiment, the target agent can dynamically optimize the reply strategy using reinforcement learning technology. The target agent includes three core components: a policy network module, an experience replay pool, and a reward calculation engine. The target agent receives feedback information such as user interaction logs and business processing results through a real-time data bus, forming a closed-loop optimization mechanism of "decision-making - execution - feedback". The specific technical solutions are as follows:

[0062] (1) State space design: Collect employee portrait labels, employee to-do list quantities, task types handled, and in-house business system classifications, etc. data, and use feature embedding technology to encode multi-dimensional data into multiple high-dimensional vectors.

[0063] (2) Action space design: According to the questions consulted by employees, define executable operations of direct decision class and process optimization class, direct decision class includes approval pass, approval rejection, supplementary materials and other strategies, process optimization class includes business system recommendation, approval process adjustment, task allocation optimization and other strategies.

[0064] (3) Reward coefficient design: Set two types of positive and negative incentives, positive incentives include shortening employee business processing time, high employee satisfaction evaluation, improving pending task processing efficiency, etc., negative incentives include increasing employee business processing time, low employee satisfaction evaluation, error recommendation of related business systems, error processing of related transactions, business process flow error, etc.

[0065] (4) Algorithm implementation: Use historical work orders and operation logs to build offline data sets, extract desensitized feature data as state space parameters, build action, state, and reward triple data sets, and implement policy network initialization. Build a virtual bank business environment to simulate employee question and answer, task allocation, operation feedback and other processes. Introduce random disturbance to simulate abnormal scenarios such as sudden high-priority tasks and system failures.

[0066] The policy network is designed to receive 256-dimensional state features (such as question and answer assistant text input, employee portrait feature vector, etc.) at the input layer, pass through a fully connected network of two layers of 256 neurons, and finally use an activation function to enhance non-linear expression ability. The output layer uses the Softmax function to convert the output into a probability distribution, representing the priority of each execution action.

[0067] The value network is designed to receive 256-dimensional state features at the input layer, pass through a fully connected network of 256 neurons and a fully connected network of 128 neurons, and use an activation function to gradually compress the feature dimension. The output estimates the value of the current state, guiding the policy network to update the policy.

[0068] The core hyperparameters of the algorithm are set as follows: the clipping threshold is 0.2, the discount factor is 0.99, and the GAE (Generalized Advantage Estimation) parameter is 0.95. The batch training settings are as follows: the number of sampling steps per round is 2048 steps, the number of update rounds is 3 to 4 rounds, and the number of samples for each gradient update is 64.

[0069] In an optional embodiment, at each training operation, the intelligent office assistant service can extract N-dimensional natural language features from each historical work order in the training set through the input layer of the neural network, where N is an integer greater than 1, then perform feature compression on the natural language features output by the input layer through the fully connected layer of the neural network, and establish a functional relationship between the feature compression result and the business system information in the training set, and convert the functional relationship output by the fully connected layer into probability distribution information through the output layer of the neural network, where the probability distribution information is used to represent the probability corresponding to the calling of each business system according to the feature compression result. Finally, the intelligent office assistant service determines the accuracy of the neural network in predicting the business system to be called according to the historical work order according to the probability distribution information.

[0070] Optionally, in each training operation, the input layer of the neural network is responsible for extracting natural language features from historical work orders. These features include but are not limited to: vocabulary, syntax structure, context information, etc. Through technologies such as word embedding, unstructured text is converted into N-dimensional feature vectors, which enables the neural network to understand and process text information in a numerical way. The size of N depends on the complexity of the features and the design of the model, and is usually a large integer, ensuring that the model can capture the rich semantic information of the text.

[0071] The extracted N-dimensional features are then input to the fully connected layer of the neural network. The fully connected layer performs non-linear transformation on the feature vector through multiple layers of neurons to achieve feature compression. The compressed feature result, i.e. the dimension of the vector, is reduced, but the core information is retained. This process helps the model to remove redundant information and focus on the features most relevant to business system calls. At the same time, the fully connected layer is also responsible for establishing a functional relationship between the compressed features and the business system call information, i.e. the model learns how to infer the business system that may need to be called from the feature vector.

[0072] After feature compression and function relationship establishment, the output of the fully connected layer is passed to the output layer of the neural network. The output layer uses a specific function (such as Softmax) to convert the output of the fully connected layer into probability distribution information, which describes the probability of calling each business system based on the feature compression result. In this way, the model can assign a probability value to each business system, intuitively reflecting the degree of association between the work order and each business system.

[0073] Finally, the intelligent office assistant service determines the accuracy of the model in predicting the business system to be called based on the probability distribution information generated by the output layer. Specifically, the model will select the business system with the highest probability as the prediction result based on the probability distribution, and compare it with the actual business system called in the historical work order. If the prediction matches the actual, the model prediction is considered correct; otherwise, the prediction is incorrect. By calculating the prediction accuracy of multiple work orders, the current performance of the model can be evaluated. If the accuracy is lower than the preset threshold, the model needs to continue training and optimization; once it reaches or exceeds the threshold, the model is considered to have converged and can accurately predict the call of the business system. At this time, the model is the "target agent" and can be deployed to the intelligent office assistant service for real-time processing of new work orders submitted by users.

[0074] This training process not only demonstrates how neural networks process and understand natural language, but also embodies the mechanism of automatic learning and optimization of prediction ability through a large amount of data training, ensuring that the intelligent office assistant service can provide accurate and efficient business system calls and processing suggestions when facing diverse user needs.

[0075] In an alternative embodiment, after determining Y target systems to cooperatively process the natural language text from X business systems according to the probability distribution matrix, the intelligent office assistant service can detect the processing time of the target system for the natural language text, the processing success rate, and the satisfaction of the target object with the processing result of the natural language text, and then determine the reward coefficient according to the processing time of the target system for the natural language text, the processing success rate, and the satisfaction of the target object with the processing result of the natural language text. Finally, the intelligent office assistant service adjusts the target agent according to the reward coefficient.

[0076] Optionally, the intelligent office assistant service can automatically track and record various indicators of the target system processing specific work orders or requests, including the following indicators:

[0077] Processing time: the time required from receiving the request to completing the processing, which directly affects the response speed and overall operational efficiency within the bank.

[0078] Processing success rate: measures the ability of the target system to successfully complete user requests, reflecting the reliability of the system and the integrity of the business process.

[0079] User satisfaction: collects feedback from the target object (usually bank employees or customers) on the processing result, obtained through questionnaires, star ratings, etc., which is an important indicator for evaluating service quality and user experience.

[0080] Based on the detected processing indicators, the intelligent office assistant service calculates a reward coefficient to evaluate the effectiveness of the target system in processing requests. The reward coefficient is a mechanism in reinforcement learning that guides and optimizes the behavior of the agent. In this application, if the target system processes requests quickly, with high success rate and user satisfaction, it will be given a higher positive reward coefficient; conversely, if the processing time is long, the success rate is low, and the user is not satisfied, it will be given a lower negative reward coefficient.

[0081] Finally, the intelligent office assistant service adjusts the decision and behavior of the target agent according to the calculated reward coefficient. In the reinforcement learning framework, the agent learns how to maximize cumulative rewards through interaction with the environment (i.e. processing requests and system responses). Positive rewards will encourage the agent to repeat similar successful behaviors, while negative rewards will encourage the agent to correct errors and explore better processing strategies. This process usually involves updating neural network weights, and through continuous trial and error and optimization, the target agent can gradually learn to make better system choices in similar situations, improving future processing efficiency and user satisfaction.

[0082] In an alternative embodiment, before querying the target information from the knowledge graph according to the system access intention, the intelligent office assistant service can classify the data in the database of each business system, and summarize the data of the same type into a data entity, then detect the similarity between the data entity and each node in the knowledge graph, and finally associate the data entity with the node with the highest similarity in the knowledge graph.

[0083] Optionally, the intelligent office assistant service first classifies the data in the database of each business system in detail. This process involves data preprocessing and labeling, dividing the data by business type, functional attribute or other meaningful dimensions, such as classifying according to account management, loan application, foreign exchange transaction, etc. Subsequently, the service summarizes the data with the same characteristics into a data entity, which helps to improve the organization and processability of the data, and also facilitates subsequent similarity detection and node association.

[0084] The intelligent office assistant service can then detect the similarity between these data entities and the nodes in the knowledge graph. The knowledge graph is a structured data representation that expresses entities and their relationships through nodes and edges. By calculating the semantic similarity between data entities and knowledge graph nodes, the service can identify the most matching graph nodes for the data entities. The key here is that similarity detection not only considers the surface matching of text, but also involves deep semantic understanding and logical reasoning to ensure that the associated results accurately reflect the actual meaning of the data.

[0085] Finally, the intelligent office assistant service associates the data entity with the node in the knowledge graph that has the highest similarity. This association process means that when the intelligent agent encounters a user's system access intent, it can quickly locate the most relevant node in the knowledge graph to obtain or infer the target information required. Node association is a key link in the entire knowledge graph query process, and it directly determines whether the intelligent agent can provide accurate and timely information support, which is an important guarantee to ensure service quality and user satisfaction.

[0086] Through the above process, the intelligent office assistant service effectively establishes a bridge between the bank's internal data and the knowledge graph, improving data accessibility and information availability. In practical applications, this helps the intelligent agent respond more quickly and accurately to users' business inquiries and operational needs.

[0087] In an optional embodiment, the target information is obtained from the knowledge graph according to the system access intent, including: the intelligent office assistant service can determine the target node from the knowledge graph according to the semantic similarity between the system access intent and each node in the knowledge graph, and then determine the target information according to the information of the target node itself and the data entity associated with the target node.

[0088] Optionally, first, the intelligent office assistant service analyzes the user's system access intent, which is usually expressed in the form of natural language text. Then, the service will calculate the semantic similarity between this intent and each node in the knowledge graph. The nodes in the knowledge graph represent specific entities or concepts, such as bank business types, policies and regulations, and operational processes. Semantic similarity calculation is based on text analysis, comparing the semantic similarity between user intent and node description, and quantifying the degree of association between the two.

[0089] Based on the calculated semantic similarity, the intelligent office assistant service can filter out the target node that best matches the user's intent from the knowledge graph. This target node is a collection of entities or concepts in the knowledge graph that are most relevant to the user's needs, providing a starting point for subsequent information retrieval and business processing. Selecting the highest similarity node as the target ensures that the intelligent agent can accurately understand and respond to the user's real needs, avoiding errors and ambiguities in information queries.

[0090] Once the target node is determined, the intelligent office assistant service further determines the target information sought by the user based on the information contained in the node itself and the data entities associated with the node in the knowledge graph. The information contained in the node itself may include detailed business process descriptions, relevant regulatory provisions, or operation guidelines, etc., while the associated data entities cover specific data instances related to the node, such as account details, transaction records, etc. By integrating node information and data entities, the service can provide rich, comprehensive and directly targeted feedback to user needs. This process not only improves the efficiency of information query, but also ensures the accuracy and practicality of the results, providing users with a high-quality service experience.

[0091] Finally, the intelligent office assistant service integrates the above collected information to form a complete and structured response, which is presented to the user in the form of text, links or operation suggestions, etc. This response not only contains the answer to the user's demand, but also may include additional guidance, suggestions or links to more detailed materials to help users better understand the business or issues of their queries, thereby improving the efficiency of the bank's internal work and customer satisfaction.

[0092] In an optional embodiment, after determining Y target systems to collaboratively process the natural language text from X business systems according to the probability distribution matrix, the intelligent office assistant service can collect at least one of the following M types of information and display the collected information on the user interface, wherein the M types of information include:

[0093] Jump link information between Y target systems;

[0094] Processing result of the natural language text;

[0095] Processing operation information of each of the Y target systems for the natural language text.

[0096] Optionally, the intelligent office assistant service can provide jump link information between Y target systems. This means that when a system has completed its part of the processing task, the user can directly jump to the next system that needs to be collaboratively processed through the link provided by the intelligent assistant, without the need to manually search or re-login. This seamless jump mechanism greatly simplifies the user's navigation work between different systems and improves office efficiency.

[0097] The intelligent office assistant service can also display the results of the natural language text after being processed by the selected Y target systems. The processing results may include answers to user queries, status of business operations (such as success or failure), notifications requiring additional user information, etc. By clearly displaying the processing results, users can quickly understand the progress of their requests and take necessary measures such as supplementing materials, confirming information or adjusting operation strategies in a timely manner.

[0098] In addition to the processing result, the intelligent office assistant service can also provide specific processing operation information of each of the Y target systems for the natural language text. This includes how the system parses the text, which operation steps are performed, which data or resources are called, etc. For complex business processes, this detailed operation information is crucial for users to understand the entire processing flow, troubleshoot problems, and also provides a basis for internal audit and compliance checks.

[0099] Optionally, the window of the intelligent office assistant service can be divided into a bubble area and a chat interaction area. The bubble area is responsible for displaying personal to-do messages of employees, pushing real-time information in the industry or industry. During real-time office work, employees can pay attention to the information in this area, reminding employees to handle related matters in a timely manner, providing cutting-edge knowledge in the financial field, and enriching the knowledge field of employees. The chat interaction area is the main area of the intelligent assistant. Employees can ask related business questions to the assistant by text or select common question list options. The assistant front end receives the request and establishes a conversation process. The request is transmitted to the background center through the real-time interface. The center reads data from each business system through an agent and returns to the front end to provide to the employee.

[0100] The intelligent assistant also has intelligent prompting and recommendation functions, which can provide personalized operation suggestions based on the operation habits and needs of employees.

[0101] When designing the intelligent assistant, the present application fully considers data security and privacy protection issues. For example, by using data encryption, access control and other security mechanisms, the security of employee and customer data during transmission and storage is ensured. The intelligent assistant also has data desensitization function, which can process sensitive data to protect customer privacy.

[0102] According to another aspect of the embodiments of the present application, a system cooperative processing device for language text is also provided, wherein, Figure 3 is a schematic diagram of an optional system cooperative processing device for language text according to an embodiment of the present application, as Figure 3 shown, the device comprises a receiving unit 301, a determining unit 302, a querying unit 303, a first processing unit 304 and a second processing unit 305.

[0103] Optionally, the receiving unit 301 is configured to receive natural language text input by a target object through a user interface; the determining unit 302 is configured to determine a system access intention of the target object according to the natural language text; the querying unit 303 is configured to query target information from a knowledge graph according to the system access intention, wherein the target information is information having an association degree greater than a preset association degree with the system access intention; the first processing unit 303 is configured to convert the system access intention and the target information into a probability distribution matrix through a target agent, wherein the probability distribution matrix represents a probability that each business system is used to process the natural language text; and the second processing unit 304 is configured to determine Y target systems from X business systems to cooperatively process the natural language text according to the probability distribution matrix, wherein X and Y are both integers greater than 1, and X is greater than or equal to Y.

[0104] Optionally, the system cooperative processing apparatus for natural language text further includes: an obtaining unit configured to obtain a training set, wherein the training set includes historical work orders submitted by users and business system information called when the historical work orders are processed; and a training unit configured to perform multiple training operations on the neural network according to the training set until the neural network enters a convergent state, to obtain the target agent. Each training operation includes: predicting a business system that needs to be called according to the historical work orders by the neural network, updating a function of the neural network and entering a next training operation when a correctness of a prediction result is detected to be less than a preset threshold, and determining that the neural network enters the convergent state when the correctness of the prediction result is detected to be greater than or equal to the preset threshold.

[0105] Optionally, the training unit includes: a first training subunit configured to extract N-dimensional natural language features from each historical work order in the training set through an input layer of the neural network, wherein N is an integer greater than 1; a second training subunit configured to perform feature compression on the natural language features output by the input layer through a fully connected layer of the neural network, and establish a function relationship between the feature compression result and the business system information in the training set; a third training subunit configured to convert the function relationship output by the fully connected layer into probability distribution information through an output layer of the neural network, wherein the probability distribution information is used to represent probabilities corresponding to calling each type of business system according to the feature compression result; and a fourth training subunit configured to determine a correctness of the neural network in predicting a business system that needs to be called according to the historical work orders.

[0106] Optionally, the system cooperative processing apparatus for language text further comprises: a detection unit configured to detect a processing time length of the target system for the natural language text, a processing success rate of the target system for the natural language text, and a satisfaction degree of the target object for a processing result of the natural language text; a coefficient determination unit configured to determine a reward coefficient according to the processing time length of the target system for the natural language text, the processing success rate of the target system for the natural language text, and the satisfaction degree of the target object for the processing result of the natural language text; and an adjustment unit configured to adjust the target intelligent agent according to the reward coefficient.

[0107] Optionally, the system cooperative processing apparatus for language text further comprises: a classification unit configured to classify data in a database of each business system, and to summarize data of a same type as a data entity; a similarity detection unit configured to detect a similarity between the data entity and each node in the knowledge graph; and a node association unit configured to associate the data entity with a node with a highest similarity in the knowledge graph.

[0108] Optionally, the query unit 303 comprises: a first determination sub-unit configured to determine a target node from the knowledge graph according to a semantic similarity between the system access intention and each node in the knowledge graph; and a second determination sub-unit configured to determine the target information according to information of the target node and a data entity associated with the target node.

[0109] Optionally, the system cooperative processing apparatus for language text further comprises: an acquisition unit configured to acquire at least one of M types of information, and to display the acquired information on a user interface, wherein the M types of information comprise:

[0110] jump link information between Y target systems;

[0111] a processing result of the natural language text;

[0112] processing operation information of each of the Y target systems for the natural language text.

[0113] According to another aspect of the embodiments of the present application, a computer readable storage medium is provided, wherein the computer readable storage medium stores a computer program, and when the computer program is executed, the computer readable storage medium causes a device where the computer readable storage medium is located to perform the system cooperative processing method for language text.

[0114] According to another aspect of the embodiments of the present application, an electronic device is provided, wherein the electronic device comprises one or more processors and a memory, and the memory is configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors perform the system cooperative processing method for language text.

[0115] According to another aspect of the embodiments of the present application, a computer program product is also provided, which includes a computer program or instructions, and the computer program or instructions, when executed by a processor, implement the above-mentioned system cooperative processing method of language text.

[0116] The above-mentioned serial numbers of the embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0117] In the above-mentioned embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0118] In the several embodiments provided by the present application, it should be understood that the disclosed technology can be implemented in other ways. Of course, the unit embodiment described above is only schematic. For example, the division of the units can be a logical function division. There can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, units or modules. The electric or other forms can be used.

[0119] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiments of the present application.

[0120] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be realized in the form of hardware, or in the form of software functional unit.

[0121] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0122] The above is only the preferred embodiment of the present application, and it should be pointed out that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should be considered as the protection scope of the present application.

Claims

1. A method for systematic collaborative processing of language texts, characterized in that: include: receiving natural language text input by a target object through a user interface; determining the system access intention of the target object according to the natural language text; Obtain target information from the knowledge graph according to the system access intention, wherein the target information is information having a correlation with the system access intention greater than a preset correlation; Converting the system access intention and the target information into a probability distribution matrix through a target agent, wherein the probability distribution matrix represents the probability of each business system being used to process the natural language text; Y target systems are determined from X business systems according to the probability distribution matrix to collaboratively process the natural language text, where X and Y are both integers greater than 1, and X is greater than or equal to Y.

2. The method for systematic collaborative processing of language text according to claim 1, characterized in that: The training process of the target agent includes the following steps: Obtaining a training set, wherein the training set includes historical work orders submitted by users and business system information called when processing the historical work orders; Performing multiple training operations on the neural network according to the training set until the neural network enters a convergence state, thereby obtaining the target intelligent agent; Among them, each training operation includes: predicting the business system that needs to be called based on the historical work order through the neural network, and when it is detected that the accuracy of the prediction result is less than a preset threshold, updating the function of the neural network and entering the next training operation; when it is detected that the accuracy of the prediction result is greater than or equal to the preset threshold, determining that the neural network has entered a convergence state.

3. The method for systematic collaborative processing of language texts according to claim 2, characterized in that: Each training operation consists of the following steps: Extracting N-dimensional natural language features from each historical work order in the training set through the input layer of the neural network, where N is an integer greater than 1; Performing feature compression on the natural language features output by the input layer through the fully connected layer of the neural network, and establishing a functional relationship between the feature compression result and the business system information in the training set; Converting the functional relationship output by the fully connected layer into probability distribution information through the output layer of the neural network, wherein the probability distribution information is used to represent the probability corresponding to calling each business system according to the feature compression result; The accuracy of the neural network in predicting the business system that needs to be called based on the historical work orders is determined based on the probability distribution information.

4. The method for systematic collaborative processing of language text according to claim 1, characterized in that: After determining Y target systems from X business systems to collaboratively process the natural language text according to the probability distribution matrix, the method for system collaborative processing of the language text further includes: Detecting the processing time and processing success rate of the target system for the natural language text and the target subject's satisfaction with the processing result of the natural language text; Determining a reward coefficient based on the target system's processing time and processing success rate for the natural language text and the target subject's satisfaction with the processing result of the natural language text; The target agent is adjusted according to the reward coefficient.

5. The method for systematic collaborative processing of language texts according to claim 1, characterized in that: Before obtaining target information from the knowledge graph according to the system access intention, the system collaborative processing method of the language text further includes: Classify the data in the database of each business system and group the data of the same type into one data entity; Detecting the similarity between the data entity and each node in the knowledge graph; Associating the data entity with the node with the highest similarity in the knowledge graph.

6. The method for systematic collaborative processing of language texts according to claim 5, characterized in that: According to the system access intention, the target information is queried from the knowledge graph, including: Determining a target node from the knowledge graph based on the semantic similarity between the system access intention and each node in the knowledge graph; The target information is determined according to the target node's own information and a data entity associated with the target node.

7. The method for systematic collaborative processing of language text according to claim 1, characterized in that: After determining Y target systems from X business systems to collaboratively process the natural language text according to the probability distribution matrix, the method for system collaborative processing of the language text further includes: Collect at least one of the following M types of information and display the collected information on the user interface, wherein the M types of information include: Jump link information between the Y target systems; The processing result of the natural language text; The processing operation information of each target system in the Y target systems for the natural language text.

8. A system collaborative processing device for language text, characterized in that: include: A receiving unit, configured to receive a natural language text input by a target object through a user interface; a determining unit, configured to determine the system access intention of the target object according to the natural language text; A query unit, configured to query and obtain target information from a knowledge graph according to the system access intention, wherein the target information is information having a correlation with the system access intention greater than a preset correlation; A first processing unit is configured to convert the system access intention and the target information into a probability distribution matrix through a target agent, wherein the probability distribution matrix represents the probability that each business system is used to process the natural language text; The second processing unit is configured to determine, according to the probability distribution matrix, Y target systems from the X business systems to collaboratively process the natural language text, where X and Y are both integers greater than 1, and X is greater than or equal to Y.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the method for systematic collaborative processing of language texts according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The system comprises one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the system collaborative processing method of language text according to any one of claims 1 to 7.

11. A computer program product, characterized in that The method comprises a computer program or instructions, which, when executed by a processor, implements the method for systematic collaborative processing of language texts according to any one of claims 1 to 7.

Images