Customer bill intelligent analysis and paraphrasing method based on large model agent Agent

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to a customer bill intelligent analysis and paraphrasing method based on a large model agent Agent. According to the method, a large model agent technology is utilized, a user bill related text question is received through a front end, the problem quality is optimized by utilizing a large model, and deep analysis of complex bill data is realized by virtue of the large model contained in a control end, a memory module for storing specific configuration information and operations of retrieving a knowledge base, calling a tool library and the like by virtue of an action end. According to the method, consumption data and package information can be converted into clear and understandable natural language explanation, a personalized consumption report can be dynamically generated, a user is supported to inquire a cost generation reason through conversation, practical suggestions and optimization strategies for consumption behaviors are provided, the user interaction experience and information transparency are effectively improved, and the user experience is improved. The user is assisted in cost management and consumption optimization, and an intelligent innovative scheme is provided for solving the existing bill analysis problem.

Description

Technical Field

[0001] The invention belongs to the field of artificial intelligence technology, and in particular relates to a method for intelligent parsing and interpretation of customer bills based on a large model intelligent agent. Background Art

[0002] In today's highly competitive market environment, telecom operators and public cloud vendors face huge challenges and opportunities. In order to gain a favorable position in the market, enterprises are constantly committed to the innovation and upgrading of products and services.

[0003] With the development of diversified services, telecom operators continue to expand their business areas, and the composition of new packages is becoming increasingly complex. Traditional voice and data services can no longer meet the needs of users, and operators have included value-added services such as home broadband, IPTV, and IoT device connection into the package system. For example, the addition of home broadband services allows users to enjoy high-speed and stable network connections in the home environment, facilitating the use of home entertainment and smart home devices; IPTV services provide users with a rich resource of TV programs, changing the traditional way of watching TV; IoT device connection services provide basic support for the development of smart homes, smart cities and other fields.

[0004] With the addition of these value-added services, users have more choices when choosing telecom packages. Different users can choose packages with different value-added service combinations according to their actual needs. For example, young users may prefer packages that include high-speed data traffic and IoT device connection services to meet their needs for mobile Internet and smart device use; while home users may pay more attention to the quality and cost-effectiveness of home broadband and IPTV services.

[0005] Service segmentation and innovation: Public cloud vendors are constantly segmenting and innovating their services in the field of cloud computing. In addition to basic computing, storage, and network services, they have also delved into areas such as security and AI. In terms of computing services, public cloud vendors provide virtual machine instances with different performance and configurations to meet the computing power needs of different users; in storage services, there are a variety of options such as object storage, block storage, and file storage, and different pricing strategies are provided for different storage needs; network services include virtual private clouds, load balancing, etc., which ensure users' network connections and data transmission in cloud environments; security services cover data encryption, firewalls, intrusion detection, and other aspects to ensure the security of user data and applications; AI services provide users with artificial intelligence algorithms and platforms such as machine learning and deep learning to help users analyze and predict data.

[0006] Diversified pricing models. Public cloud vendors have developed diversified pricing models for various services. In addition to the common pay-as-you-go and annual and monthly subscription models, they have also introduced various pricing methods such as tiered pricing based on usage and reserved instance pricing. For example, the pay-as-you-go model is suitable for small and medium-sized enterprises and startups that are more sensitive to cost control in the early stages of their business. They can pay flexibly based on actual usage; while the annual and monthly subscription model is more suitable for enterprises with relatively stable resource needs, and they can obtain certain price discounts through long-term purchases.

[0007] Disadvantages of the existing bill presentation method:

[0008] 1. Although telecom operators and public cloud vendors have made significant progress in products and services, there is a clear lag in bill presentation. For example, the bill structure is simple; the content of traditional bills is limited. Traditional bills are usually provided to users in the form of paper or electronic documents, and their structure is relatively simple. The main content includes basic information of the user, such as name, account number, billing cycle, total amount of consumption, and a simple classification of various expenses. In telecom bills, common expense categories include voice calls, text messages, and data traffic; in public cloud bills, expenses are generally classified according to service categories such as computing, storage, and network. Each expense item usually only lists its corresponding usage and unit price.

[0009] 2. Lack of in-depth analysis. This simple billing structure lacks detailed analysis of user consumption behavior. For example, in a telecommunications bill, users can only see how many minutes of voice calls they have used, how many text messages they have sent, and how much data traffic they have used in a certain period of time, but they cannot know the specific scenarios in which these consumption behaviors occurred. For example, users cannot know which voice calls are made locally and which are long-distance, or which data traffic is used for video playback and which is used for software updates. In public cloud bills, although users can see the charges for each service category, they cannot understand how these charges are generated based on their business operations, such as which computing tasks consume more resources and which storage operations lead to increased costs.

[0010] 3. Poor interactivity and inconvenience for users to query. The traditional bill display method lacks interactivity, and users can only passively view the information on the bill. If users have questions about certain charges on the bill, it is difficult to find the answer through the bill itself. For example, a user finds that the telecommunications fee suddenly increased in a certain month, but cannot find the specific reason for the increase in fees on the bill. The only way to solve the problem is to call the customer service phone or go to the business hall for consultation. In terms of public cloud bills, if users have questions about the cost calculation of certain cloud services, they cannot directly obtain relevant explanations and analysis on the bill.

[0011] 4. It is not conducive to cost management. Due to the lack of detailed analysis and interactivity of consumption behavior, it is difficult for users to conduct in-depth cost management and consumption optimization through traditional bills. In telecommunications consumption, users cannot make reasonable consumption plans based on bill information, such as how to reduce unnecessary voice calls and data traffic costs without affecting normal use; in public cloud consumption, enterprise users cannot optimize the use of cloud resources based on bill information, such as how to adjust the allocation of computing tasks to reduce computing costs, or how to reasonably choose storage types to reduce storage costs.

[0012] This situation results in users being unable to intuitively and comprehensively understand their consumption details and cost structure when faced with complex packages and diverse services. There is an urgent need for a technology that can intelligently parse and interpret bills to solve these problems. Summary of the invention

[0013] In view of the significant deficiencies in the current related prior art in the processing of customer bills, the present invention provides a method for intelligent parsing and interpretation of customer bills based on a large model intelligent agent, which is used to realize the intelligent parsing and interpretation of customer bills. With the help of advanced intelligent data analysis technology and natural language generation capabilities, the present invention can automatically analyze complex consumption data and diversified package information, and convert them into clear and easy-to-understand natural language expressions to provide users with intuitive explanations. At the same time, this patent has the ability to dynamically generate highly personalized consumption reports to meet the specific needs of different users. Users can not only conveniently query the source of various expenses through dialogue, but also obtain practical suggestions and optimization strategies carefully customized for their specific consumption behaviors, thereby effectively improving users' understanding of bills and assisting users in accurate cost management and consumption decision optimization.

[0014] To achieve the above object, the present invention provides a method for intelligent parsing and interpretation of customer bills based on a large model agent, which at least comprises the following steps:

[0015] Input perception step: After the front end receives the text input about the bill from the user, it uses the big model to process the input text. Through grammatical analysis, semantic understanding and keyword extraction operations, it improves the clarity and pertinence of the question and thus improves the quality of the question.

[0016] Control end setting steps: The control end module includes a basic large model that can be provided by a cloud platform or deployed locally, and a memory module with a storage function. The memory module stores long-term prompt word configurations customized for bill parsing scenarios and short-term user question context information, wherein the prompt word configuration is also set with bill parsing thought chain steps to guide the analysis logic of the large model;

[0017] Action-side execution steps: When the action-side module receives questions related to user bills, it first searches the knowledge base, which stores background knowledge about bills, metadata about related products and packages ordered by users. The big model combines the information retrieved from the knowledge base and automatically calls the bill item time interval query function and different dimension sub-item query functions in the tool library. If other data references are needed during the processing, other APIs in the tool library will continue to be called. Finally, the big model generates and outputs answers in real time based on the user's question requirements. When the user has further questions about the answer content, the above steps are repeated to continue responding and processing.

[0018] Furthermore, in the action-side execution step, based on the analysis of the user's billing data and the understanding of consumption behavior, a highly personalized consumption report can be dynamically generated. The report is customized based on the user's historical consumption data, current billing information, and the user's personalized demand characteristics to meet the specific needs of different users in terms of cost detail query, consumption trend analysis, and package content interpretation;

[0019] At the same time, it provides users with practical suggestions and optimization strategies customized for their specific consumption behaviors, including suggestions for package upgrades, downgrades or changes based on the usage of various services in telecom user packages, resource allocation optimization suggestions based on the resource utilization efficiency of public cloud users, and cost optimization strategies for users based on their consumption patterns, business needs and bill data characteristics, including resource allocation plans and pricing model selection strategies based on business peaks and valleys for public cloud users, thereby effectively improving users' understanding of bills and assisting users in accurate cost management and consumption decision optimization.

[0020] Furthermore, in the input perception step, the large model performs semantic clarification and intent refinement on ambiguous user questions, converting the user's original questions into a form that better meets the analysis requirements, and the conversion process is based on the understanding of the bill data structure and common consumption scenarios.

[0021] Furthermore, the long-term prompt word configuration in the control end setting step covers a variety of bill analysis scenarios, including detailed analysis of different types of expenses, comparative evaluation between different packages, and one or more relevant prompt words in exploring the causes of abnormal fluctuations in expenses, so as to assist the large model in performing effective analysis in different bill problem scenarios.

[0022] Furthermore, the knowledge base data in the action-end execution step is updated regularly, and its update frequency can be dynamically adjusted according to the billing data change frequency and product package update cycle factors, and efficient indexing algorithms and search strategies are used when searching the knowledge base to ensure fast and accurate positioning of knowledge content related to user problems.

[0023] Furthermore, the bill item time interval query function and the different dimension sub-item query function in the tool library in the action end execution step have high stability and accuracy. When calling other external APIs, data encryption transmission and identity authentication security mechanisms are adopted to ensure data compatibility and transmission security, and prevent data leakage and illegal access.

[0024] Furthermore, in the execution step on the action side, when the big model processes data according to the customized thinking chain steps, data mining and machine learning algorithms are used to assist in the analysis of complex data, to mine potential patterns and correlations in the data, and in the result generation stage, the analysis results are explained in detail in natural language, and the data comparison and analysis details are displayed in intuitive visual forms such as charts and tables for easy user understanding.

[0025] Furthermore, it supports users to ask questions in a spoken, natural language style and has the ability to respond quickly, with the response time being within an acceptable range of user interaction experience, to meet users' flexible and diverse query needs for bill information, including one or more of expense detail inquiries, consumption trend analysis, and package content interpretation.

[0026] Furthermore, during the execution of the entire method, the user's interaction history is continuously recorded, including one or more information of question content, question time and answer results, so that the large model can fully combine historical interaction information when processing subsequent questions, better understand the context and intention of the user's current question, and improve the accuracy and relevance of the answer.

[0027] The present invention adopts the above technical solution, which has at least the following beneficial effects:

[0028] 1. Optimization of user interaction experience. The dialogue interaction mechanism built based on the large language model gives users the ability to ask questions conveniently in colloquial natural language, and the large model can respond instantly and give accurate answers. This feature completely breaks the static and limitations of traditional bill presentation methods, and transforms the originally obscure bill information into a smooth and intuitive interactive experience. Users are no longer limited to passively viewing the bill content in a fixed format, but can flexibly initiate diversified inquiries according to their own needs. Whether it is in-depth questioning of expense details, macro-grasp of consumption trends, or detailed interpretation of package details, they can get real-time and clear feedback. This highly flexible interaction mode greatly improves the communication efficiency between users and the billing system, allowing users to easily and independently explore and understand their own consumption situation, significantly improving the user experience, and enhancing user satisfaction and trust in billing services.

[0029] 2. Improved information transparency. With the help of intelligent agent technology and large model thinking chain, the system has pre-set a variety of capability functions and professional plug-ins, carefully customized scenario prompts, and built a comprehensive and accurate private domain knowledge base, successfully building an efficient bill interpretation intelligent agent application architecture. When facing various questions from users, the system can quickly and accurately filter and associate relevant information from massive data with its powerful intelligent integration capabilities, deeply explore the logical relationship behind the data, and present complex consumption data and package rules to users in a clear and easy-to-understand way. This not only effectively eliminates the understanding barriers caused by the simple listing of information in traditional bills, but also provides users with a full range of multi-level consumption information interpretation, allowing users to deeply understand the source of each expense, consumption scenarios and the relationship with the overall package, greatly improving the transparency of information, and allowing users to have a more thorough and comprehensive understanding of their own consumption status.

[0030] 3. Cost management and expenditure optimization. Based on the in-depth analysis and intelligent interpretation of user bill data, the system can provide users with highly targeted personalized consumption suggestions and scientific and reasonable cost optimization strategies. For telecom users, the system can accurately determine whether there is a package mismatch or resource waste based on the frequency, duration, traffic consumption and other multi-dimensional data of each service in their package, and timely propose package upgrade, downgrade or change suggestions to ensure that users can avoid unnecessary expenses while enjoying high-quality services. For public cloud users, the system can tailor resource allocation optimization solutions and optimal pricing model selection strategies for them through detailed analysis of key factors such as resource utilization efficiency and business peak and valley periods, helping users to maximize resource utilization and reduce operating costs while meeting business needs. Through these precise suggestions and strategies, users can manage their own consumption costs more scientifically and effectively, plan budgets reasonably, and achieve refined optimization of expenditures, thereby transforming bills from simple expense records into powerful financial management tools, helping users make wise decisions in a complex and changing consumption environment and improve financial benefits and resource utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 It is a structural diagram of the bill intelligent parsing and interpretation system of the present invention;

[0033] Figure 2It is a flow chart of intelligent parsing and interpretation of customer bills of the present invention. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0035] like Figure 1 and Figure 2 As shown, in this embodiment, a method for intelligent parsing and interpretation of customer bills based on a large model agent includes at least the following steps:

[0036] Input perception step: After the front end receives the text input about the bill from the user, it uses the big model to process the input text. Through grammatical analysis, semantic understanding and keyword extraction operations, it improves the clarity and pertinence of the question and thus improves the quality of the question.

[0037] Control end setting steps: The control end module includes a basic large model that can be provided by a cloud platform or deployed locally, and a memory module with a storage function. The memory module stores long-term prompt word configurations customized for bill parsing scenarios and short-term user question context information, wherein the prompt word configuration is also set with bill parsing thought chain steps to guide the analysis logic of the large model;

[0038] Action-side execution steps: When the action-side module receives questions related to user bills, it first searches the knowledge base, which stores background knowledge about bills, metadata about related products and packages ordered by users. The big model combines the information retrieved from the knowledge base and automatically calls the bill item time interval query function and different dimension sub-item query functions in the tool library. If other data references are needed during the processing, other APIs in the tool library will continue to be called. Finally, the big model generates and outputs answers in real time based on the user's question requirements. When the user has further questions about the answer content, the above steps are repeated to continue responding and processing.

[0039] As an implementation mode, in the execution step of the action end, this embodiment can dynamically generate a highly personalized consumption report based on the analysis of the user's billing data and the understanding of the consumption behavior. The report is customized according to the user's historical consumption data, current billing information and the user's personalized demand characteristics to meet the specific needs of different users in terms of cost detail query, consumption trend analysis and package content interpretation;

[0040] At the same time, it provides users with practical suggestions and optimization strategies customized for their specific consumption behaviors, including suggestions for package upgrades, downgrades or changes based on the usage of various services in telecom user packages, resource allocation optimization suggestions based on the resource utilization efficiency of public cloud users, and cost optimization strategies for users based on their consumption patterns, business needs and bill data characteristics, including resource allocation plans and pricing model selection strategies based on business peaks and valleys for public cloud users, thereby effectively improving users' understanding of bills and assisting users in accurate cost management and consumption decision optimization.

[0041] As an implementation method, in the input perception step in this embodiment, the big model performs semantic clarification and intent refinement on ambiguous user questions, converting the original user questions into a form that better meets the analysis requirements, and the conversion process is based on the understanding of the bill data structure and common consumption scenarios.

[0042] As an implementation method, the long-term prompt word configuration in the control end setting step described in this embodiment covers a variety of bill analysis scenarios, including detailed analysis of different types of charges, comparative evaluation between different packages, and one or more relevant prompt words in exploring the causes of abnormal fluctuations in charges, so as to assist the large model in performing effective analysis in different bill problem scenarios.

[0043] As an implementation method, the knowledge base data in the action end execution step described in this embodiment is updated regularly, and the update frequency can be dynamically adjusted according to the billing data change frequency and product package update cycle factors, and efficient indexing algorithms and search strategies are used when searching the knowledge base to ensure fast and accurate positioning of knowledge content related to user problems.

[0044] As an implementation method, the bill item time interval query function and the different dimension sub-item query function in the tool library in the action end execution step described in this embodiment have high stability and accuracy. When calling other external APIs, data encryption transmission and identity authentication security mechanisms are adopted to ensure data compatibility and transmission security, and prevent data leakage and illegal access.

[0045] As an implementation method, in the action end execution step of this embodiment, when the big model processes the data according to the customized thinking chain steps, data mining and machine learning algorithms are used to assist in the analysis of complex data to mine the potential patterns and associations in the data. In the result generation stage, the analysis results are explained in detail in natural language, and the data comparison and analysis details are displayed in intuitive visual forms such as charts and tables for easy user understanding.

[0046] As an implementation method, this embodiment supports users to ask questions in a spoken, natural language style and has the ability to respond quickly. The response time is within an acceptable range of user interaction experience, thereby realizing users' flexible and diverse query needs for bill information, including one or more of expense detail query, consumption trend analysis, and package content interpretation.

[0047] As an implementation method, in this embodiment, the user's interaction history is continuously recorded during the execution of the entire method, including one or more information of question content, question time and answer results, so that the large model can fully combine historical interaction information when processing subsequent questions, better understand the context and intention of the user's current question, and improve the accuracy and relevance of the answer.

[0048] As an application scenario, for example:

[0049] 1. User-initiated questions

[0050] After checking the monthly mobile phone bill, the user becomes suspicious because he finds that the charges are higher than usual. He enters a natural language question through the interactive interface provided by the system (such as a mobile application, a web customer service portal, etc.): "Why is my phone bill so much higher this month?"

[0051] 2. Agent Problem Optimization

[0052] After receiving the original question from the user, the agent immediately starts the big model to process it. The big model uses the semantic understanding and intention analysis modules in natural language processing technology to identify that the user is mainly concerned about the reasons for the increase in phone charges and comparative analysis.

[0053] Based on its understanding of common bill analysis scenarios and user intent, the agent optimized the original question to: "Compare the phone bills of this month with those of last month, and list the reasons for the differences." This optimized question is more targeted and actionable, and can guide the subsequent system's precise data query and analysis.

[0054] 3. Data Acquisition and Information Analysis

[0055] Fee details:

[0056] According to the optimized problem, the agent calls the user fee query interface in the tool library through a function. This interface interacts with the telecom operator's billing system in a safe and efficient manner to accurately obtain the detailed call fee details of the user this month and last month, including the specific amount of each fee, billing time and other information. For example, the voice call fee this month is [X] yuan, and it was [Y] yuan last month; the data traffic fee this month is [M] yuan, and it was [N] yuan last month, etc.

[0057] Package and product association analysis: The agent further analyzes the package or product list associated with the expense item. By interacting with the knowledge base, it determines which specific services are included in the package used by the user (such as the free call time, data traffic quota, number of SMS, etc. included in the package) and whether there are additional value-added products (such as ringtones, video membership rights, etc.). For example, the user is using the [package name] package, which includes [X] minutes of free voice calls and [Y] GB of data traffic per month, and this month the user additionally ordered a [value-added product name].

[0058] Knowledge base search: Continue to search the knowledge base for meta information of packages or products, including detailed rules of packages, billing standards of various services, functions and prices of value-added products, etc. For example, the knowledge base stores information such as the billing standard for this package exceeding the free call time is [Z] yuan per minute, and the charge for exceeding the data traffic quota is [P] yuan per [GB].

[0059] 4. Data Processing and Result Generation

[0060] Big model data analysis: The big model conducts a comprehensive analysis of the acquired call fee data and package information for this month and last month based on the customized thinking chain steps. First, the data is structured and the various fees are classified and sorted according to categories such as voice calls, data traffic, text messages, and value-added services. Then, by comparing the data of this month and last month, the difference in the cost changes is found. For example, it is found that the data traffic fee this month has increased by [Q] yuan because users watched more videos this month, which exceeded the data traffic quota in the package, and additional fees were incurred according to the billing standard for the excess part.

[0061] Results display and explanation: This month's and last month's call charges are compared and presented in a clear and intuitive table format, with the table listing in detail the amount of each fee this month, the amount of last month, and the changes. At the same time, explain in detail in natural language the origin of the bill items that are higher than last month's call charges below or next to the table. For example, "This month, your data traffic fee increased by [Q] yuan compared to last month because the total amount of data traffic you used this month was [specific value] GB, which exceeded the [Y] GB limit of [specific excess value] GB in the package, and additional traffic charges were incurred according to the excess billing standard of [P] yuan per [GB]."

[0062] Generation of consumption suggestions: Finally, based on the user's call usage and package features, the large model makes personalized recommendations on call usage. If the user frequently exceeds the data traffic quota, it is recommended that the user upgrade to a package with a larger data traffic, or recommend some short-term traffic refueling packages. At the same time, it will also pay attention to other preferential packages on the market that meet the needs of users, and provide relevant information to users, such as "You can consider upgrading to the [recommended package name] package, which provides [more traffic value] GB of data traffic per month, and the price is only [specific price difference] yuan higher than your current package, which is more cost-effective. Or you can also order a [traffic refueling package name] for the rest of this month, the price is [refueling package price] yuan, and you can get an additional [refueling package traffic value] GB of data traffic."

[0063] Through the above detailed steps, users can gain an in-depth understanding of the reasons for the increase in their phone bills and obtain targeted consumption suggestions, so as to better manage their mobile phone bill expenses and improve user experience and consumption benefits.

[0064] Another example: a company uses public cloud services to support its e-commerce business operations, including computing resources for website server operation, storage resources for product images and order data storage, network resources to ensure smooth user access, etc. After receiving the monthly bill, the company's financial staff found that the expenses were higher than before, so they asked the bill intelligent parsing and interpretation system of the present invention: "Why did the public cloud fee increase so much this month?"

[0065] After receiving the question, the agent optimizes it to: "Compare the public cloud costs this month with last month, and list the reasons for the charges for the different items by computing, storage, and network resources." The agent then calls the cost query interface in the tool library to obtain detailed cost data for two months. It then parses the list of cloud service packages and products associated with the cost items, and retrieves detailed metadata of these packages and products from the knowledge base, such as the unit price of different computing resource specifications, the billing method of storage capacity, and the charging standard for network traffic. The large model performs structured processing and in-depth analysis of the data based on customized thinking chain steps, and compares and displays the changes in the costs of various resources this month and last month in an intuitive table format. At the same time, it explains in detail the reasons for the increased bill items, such as the surge in website traffic due to this month's e-commerce promotion activities, which has increased network traffic costs and the cost of elastic expansion of computing resources. Finally, cost optimization strategies are proposed based on the company's business model and resource usage. For example, it is recommended to appropriately reduce the configuration of computing resources during the non-promotion period, purchase a certain amount of network traffic packages in advance to enjoy discounts, and clean up or archive some storage data that has not been used for a long time to reduce storage costs, etc., to help companies better manage public cloud service costs and improve financial benefits.

[0066] Another example: A user has subscribed to a telecom package that includes home broadband, mobile phone calls, text messages, and data traffic shared by multiple family members. When the user receives the bill one month, he finds that the total cost exceeds his expectations, so he asks the system: "Why did my telecom package cost exceed the limit this month?"

[0067] The agent optimizes the problem to: "Analyze the difference between the service fees and the package standards in this month's package, and list the reasons for overspending". Then the agent obtains the usage details of each service in the user's package this month through function calls, such as the length of mobile phone calls, the number of text messages, and the data traffic usage of each family member, and parses the association rules and preferential policies of each service in the package. Find the detailed information of the package from the knowledge base, including the basic quota of each service in the package, the billing standard after exceeding the quota, etc. The large model conducts a comprehensive analysis of the data according to the set thinking chain steps, and uses clear charts to show the comparison between the service fees and the package standards, and explains the reasons for overspending, such as a family member's data traffic exceeding the package quota due to watching high-definition videos, resulting in additional fees. Finally, according to the user's usage habits and family needs, personalized consumption suggestions are put forward, such as recommending upgrading to a package with larger data traffic, or opening a specific traffic package to meet the large traffic needs of individual family members, and recommending some preferential package combinations suitable for the family, so that users can plan their telecommunications consumption more reasonably, avoid unnecessary expenses, and achieve consumption optimization.

[0068] Although the embodiments of the present invention have been shown and described above, it will be appreciated that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, substitute and vary the above embodiments within the scope of the present invention.

Claims

1. A method for intelligent parsing and interpretation of customer bills based on a large model agent, characterized by: At least the following steps are included: Input perception step: After the front end receives the text input about the bill from the user, it uses the big model to process the input text. Through grammatical analysis, semantic understanding and keyword extraction operations, it improves the clarity and pertinence of the question and thus improves the quality of the question. Control end setting steps: The control end module includes a basic large model that can be provided by a cloud platform or deployed locally, and a memory module with a storage function. The memory module stores long-term prompt word configurations customized for bill parsing scenarios and short-term user question context information, wherein the prompt word configuration is also set with bill parsing thought chain steps to guide the analysis logic of the large model; Action-side execution steps: When the action-side module receives questions related to user bills, it first searches the knowledge base, which stores background knowledge about bills, metadata about related products and packages ordered by users. The big model combines the information retrieved from the knowledge base and automatically calls the bill item time interval query function and different dimension sub-item query functions in the tool library. If other data references are needed during the processing, other APIs in the tool library will continue to be called. Finally, the big model generates and outputs answers in real time based on the user's question requirements. When the user has further questions about the answer content, the above steps are repeated to continue responding and processing.

2. The method according to claim 1, characterized in that: In the action execution step, based on the analysis of user billing data and understanding of consumption behavior, a highly personalized consumption report can be dynamically generated. The report is customized based on the user's historical consumption data, current billing information, and the user's personalized demand characteristics to meet the specific needs of different users in terms of cost detail query, consumption trend analysis, and package content interpretation; At the same time, it provides users with practical suggestions and optimization strategies customized for their specific consumption behaviors, including suggestions for package upgrades, downgrades or changes based on the usage of various services in telecom user packages, resource allocation optimization suggestions based on the resource utilization efficiency of public cloud users, and cost optimization strategies for users based on their consumption patterns, business needs and bill data characteristics, including resource allocation plans and pricing model selection strategies based on business peaks and valleys for public cloud users, thereby effectively improving users' understanding of bills and assisting users in accurate cost management and consumption decision optimization.

3. The method according to claim 1, characterized in that: In the input perception step, the big model performs semantic clarification and intent refinement on ambiguous user questions, converting the original user questions into a form that better meets the analysis requirements. This conversion process is based on an understanding of the bill data structure and common consumption scenarios.

4. The method according to claim 1, characterized in that The long-term prompt word configuration in the control end setting step covers a variety of bill analysis scenarios, including detailed analysis of different expense types, comparative evaluation between different packages, and one or more relevant prompt words in exploring the causes of abnormal fluctuations in expenses, to assist the large model in performing effective analysis in different bill problem scenarios.

5. The method according to claim 1, characterized in that: The knowledge base data in the action-side execution step is updated regularly, and its update frequency can be dynamically adjusted according to factors such as the billing data change frequency and the product package update cycle. In addition, efficient indexing algorithms and search strategies are used when searching the knowledge base to ensure fast and accurate positioning of knowledge content related to user problems.

6. The method according to claim 1, characterized in that The bill item time interval query function and the different dimension sub-item query function in the tool library in the action end execution step have high stability and accuracy. When calling other external APIs, data encryption transmission and identity authentication security mechanisms are adopted to ensure data compatibility and transmission security, and prevent data leakage and illegal access.

7. The method according to claim 1, characterized in that In the action-end execution steps, when the big model processes data according to the customized thinking chain steps, it uses data mining and machine learning algorithms to assist in the analysis of complex data, explores the potential patterns and correlations in the data, and in the result generation stage, explains the analysis results in detail in natural language, and combines charts and tables to display data comparison and analysis details in an intuitive and visual form for easy user understanding.

8. The method according to claim 1, characterized in that It supports users to ask questions in a spoken, natural language style and has the ability to respond quickly. The response time is within the acceptable range of user interaction experience, and realizes users' flexible and diverse query needs for bill information, including one or more of expense detail query, consumption trend analysis and package content interpretation.

9. The method according to claim 1, characterized in that: During the execution of the entire method, the user's interaction history is continuously recorded, including one or more information such as question content, question time, and answer results, so that the large model can fully combine historical interaction information when processing subsequent questions, better understand the context and intention of the user's current question, and improve the accuracy and relevance of the answer.

Images