A telecommunications customer service intelligent system based on fusion of CoT and RAG technology

Abstract

The application discloses a kind of based on the intelligent system of telecommunication customer service of fusion CoT and RAG technology, comprising the following steps: step Y1: environment configuration;Step Y2: telecommunication user first asks question;Step Y3: generates query;Step Y4: loads the inventory data of telecommunication user;Step Y5: updates query string;Step Y6: queries telecommunication knowledge base;Step Y7: modifies perfect result;Step Y8: loads rule base, compliance checks;Step Y9: check whether candidate answer is empty;Step Y10: reply final result.The application needs to call CoT thinking chain technology to simulate the logical way of artificial customer service to solve the problem first, then call RAG retrieval enhancement to generate more professional knowledge in the field of telecommunications and telecommunication user personal service call conditions, such as telecommunication user package consumption, call, short message, traffic, and finally get the required answer result through the screening of rule base.

Description

Technical Field

[0001] This invention relates to the field of customer service Q&A technology in the telecommunications industry, and in particular to a telecommunications customer service intelligent system based on the integration of CoT and RAG technologies. Background Technology

[0002] In the telecommunications customer service field, intelligent assistant technology typically uses prompt word engineering to optimize responses when providing services. This method guides the conversation through preset prompt words or templates to improve the standardization and accuracy of responses. However, this system relying on prompt word engineering has the following drawbacks:

[0003] Lack of in-depth understanding: Systems that only use prompt word engineering may not be able to fully understand the complexity and context of user queries, resulting in answers that are too superficial and fail to delve into the core of the problem.

[0004] Unstable response quality: The prompt word engineering cannot pre-plan all dialogue scenarios or customer needs. If the customer's question or way of asking the question is unusual, this method often fails to obtain a satisfactory answer.

[0005] Insufficient personalized services: Due to a lack of in-depth analysis of users' individual needs, these systems may be unable to provide truly customized service recommendations, reducing user satisfaction.

[0006] Lagging Updates: Systems that rely on preset templates may lag behind in updating when faced with rapidly changing service products and regulations in the telecommunications industry, failing to reflect the latest service information in a timely manner.

[0007] At the same time, the unique characteristics of telecom operators' user data further highlight the shortcomings of existing systems. Compared to general customer service systems, telecom user data not only includes general customer service interaction information but also encompasses additional user attributes, service characteristics, and behavioral characteristics, such as number affiliation, communication habits, and package consumption. This complexity of data, coupled with the diversity and complexity of telecom operators' service product systems, presents significant challenges to existing customer service systems.

[0008] Existing systems often fail to provide effective and accurate response suggestions within a single conversation, making it difficult to meet the characteristics of telecom operators' users and service products, and consequently hindering accurate service recommendations and personalized user support. This limitation is particularly pronounced in the telecom industry, as users expect a fast, accurate, and personalized service experience, expectations that existing systems often fail to meet.

[0009] To address the aforementioned issues, this invention proposes a telecommunications customer service intelligent system based on the integration of CoT (Cooperation of Thought) and RAG (Retrieval Enhancement Generation) technologies. This system first uses CoT technology to simulate the problem-solving logic of a human customer service representative. Then, it uses RAG to load more professional telecommunications knowledge and individual service usage data of the user, such as their data plan consumption, call volume, SMS usage, and data usage. Finally, it uses a rule-based database to filter and obtain responses that meet the requirements. Summary of the Invention

[0010] This invention aims to break through the limitations of existing telecommunications customer service systems and proposes for the first time in the telecommunications industry an innovative method that integrates CoT (Coding-to-Track) and RAG (Retrieval-Enhanced Generation) technologies, specifically tailored for a smart assistant for telecommunications customer service.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies includes the following steps:

[0013] Step Y1: Environment Configuration;

[0014] Step Y2: The telecom user submits their first inquiry;

[0015] Step Y3: Generate query;

[0016] Step Y4: Load existing data of telecom users;

[0017] Step Y5: Update the query string;

[0018] Step Y6: Query the telecommunications knowledge base;

[0019] Step Y7: Modify and improve the results;

[0020] Step Y8: Load the rule base and perform compliance checks;

[0021] Step Y9: Check if the candidate answer is empty;

[0022] Step Y10: Reply with the final result.

[0023] As a further improvement to this technical solution, it also includes the Nth question from the telecommunications user, specifically including the following steps:

[0024] Step N1: Environment Configuration;

[0025] Step N2: The telecom user asks questions N times;

[0026] Step N20: The system detects the problem and triggers the start of the process;

[0027] Step N21: Generate a draft answer;

[0028] Step N210: Use the large-scale telecommunications model to interpret the problems of telecommunications users;

[0029] Step N211: Generate a draft answer using the large telecommunications model;

[0030] Step N3: Generate query;

[0031] Step N30: ​​Summarize and merge the questions and draft answers;

[0032] Step N300: Use the large-scale telecommunications model to summarize the synthesis problem and draft answers;

[0033] Step N31: Generate the query string;

[0034] Step N310: Invoke the large telecom model to generate a query string for searching;

[0035] Step N4: Load existing data of telecom users;

[0036] Step N40: Load telecom user data via third-party API;

[0037] Step N5: Update the query string;

[0038] Step N50: Analyze telecommunications user data using the large telecommunications model and output the update query string;

[0039] Step N6: Query the telecommunications knowledge base;

[0040] Step N60: Invoke the updated query to search for relevant content in the telecommunications knowledge base;

[0041] Step N61: Set to query once;

[0042] Step N62: Calculate the similarity between the query string and the document;

[0043] Step N63: Obtain the first search result;

[0044] Step N7: Load the conversation history;

[0045] Step N8: Modify and improve the results;

[0046] Step N80: Call the telecom big data model to modify and improve the draft answer based on the above search results and dialogue records, and output candidate results;

[0047] Step N9: Load the rule base and perform compliance checks;

[0048] Step N10: Check if the candidate answer is empty;

[0049] Step N11: Reply with the final result.

[0050] As a further improvement to this technical solution: step Y2: the telecommunications user's first inquiry includes:

[0051] Step Y20: The system detects the problem and triggers the start of the process;

[0052] Step Y21: Generate a draft answer;

[0053] Step Y210: Use the large-scale telecommunications model to interpret the problems of telecommunications users;

[0054] Step Y211: Generate a draft answer using the large telecom model.

[0055] As a further improvement to this technical solution: step Y3: generating a query, includes:

[0056] Step Y30: Summarize and merge the questions and draft answers;

[0057] Step Y300: Use the large-scale telecommunications model to summarize the synthesis problem and draft answers;

[0058] Step Y31: Generate the query string;

[0059] Step Y310: Call the large telecom model to generate a query string for searching.

[0060] As a further improvement to this technical solution: Step Y4: Load the existing data of telecommunications users, including:

[0061] Step Y40: Load user-specific data via a third-party API.

[0062] As a further improvement to this technical solution: Step Y5: Update the query string, including:

[0063] Step Y50: Analyze telecommunications user data using the large telecommunications model and output the update query string.

[0064] As a further improvement to this technical solution: Step Y6: Query the telecommunications knowledge base, including:

[0065] Step Y60: Invoke the updated query to search for relevant content in the telecommunications knowledge base;

[0066] Step Y61: Set to query once;

[0067] Step Y62: The system calculates the similarity between the query string and the document based on the cosine similarity formula: cos(θ)=(A·B) / (|A||B|), where A and B are the TF-IDF vectors of the query and knowledge base documents, respectively.

[0068] Step Y63: Obtain the first search result.

[0069] As a further improvement to this technical solution: Step Y7: Modify and improve the results, including:

[0070] Step Y70: Based on the understanding of the search results for telecommunications-related knowledge, the large-scale telecommunications model is invoked to modify and improve the draft answer, ensuring the professionalism and accuracy of the answer, and outputting candidate results.

[0071] As a further improvement to this technical solution: Step Y8: Load the rule base and perform compliance checks, including:

[0072] Step Y80: Are there any banned words in the answer?

[0073] Step Y81: Has the recommended service expired?

[0074] Step Y82: Answer whether there have been any new changes to the service process involved;

[0075] Step Y83: Determine whether the answer complies with telecommunications regulations.

[0076] As a further improvement to this technical solution: Step Y9: Check whether the candidate answer is empty, including:

[0077] Step Y90: Call the telecommunications big data model to understand the semantics and check whether the semantics contain information;

[0078] Step Y91: If there is information, save the results;

[0079] Step Y92: If no useful information is found, return to step Y63 and obtain the second one;

[0080] Step Y93: Continue in this manner until meaningful content is obtained.

[0081] Compared with the prior art, the beneficial effects of the present invention are:

[0082] Compared with existing technologies, this invention proposes a method for creating a smart assistant for telecom operator customer service based on the integration of CoT (Coding-of-Thinking) and RAG (Retrieval-Enhanced Generation) technologies. Specifically designed for telecom user service scenarios, it offers the following advantages and benefits:

[0083] (1) Precise service recommendations:

[0084] Traditional customer service systems struggle to effectively address the complex needs of telecom operator users in a single conversation. This invention uses thought chain technology to simulate the problem-solving logic of human customer service representatives, combined with RAG technology to incorporate telecom expertise and personalized service data from telecom users. This allows for continuous refinement of answers through multiple interactions, providing more accurate service recommendations.

[0085] (2) In-depth user profile of telecom users:

[0086] In the process of data collection and knowledge construction, this invention integrates telecommunications user attributes, service characteristics, and behavioral characteristics, such as number affiliation, communication habits, and package consumption. By analyzing telecommunications user profiles through data traffic sources, the personalization and accuracy of recommendations are further improved.

[0087] (4) Intelligent filtering and optimization:

[0088] Before generating the final answer, the system undergoes a thought process training process to ensure that the answer complies with the latest updated rules of the telecommunications company, does not contain prohibited words, and that the recommended service prices are updated in a timely manner and the service processes involved are accurate. Combined with the semantic understanding capabilities of the telecommunications company's large-scale model, the answer is intelligently filtered and optimized to guarantee its quality and reliability.

[0089] (5) Dynamic adjustment and iterative improvement:

[0090] In response to multiple inquiries from telecom users, the system will incorporate the conversation history and the user's background information as context when revising the answer, iterating and improving the response until the user is satisfied. This dynamic adjustment and iterative improvement mechanism better meets the needs of telecom users and enhances their user experience.

[0091] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0092] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0093] Figure 1 This is a flowchart of the main steps in scenario one proposed by the present invention;

[0094] Figure 2 This is a flowchart of the main steps in scenario two proposed in this invention;

[0095] Figure 3This is a detailed flowchart of the steps in scenario one proposed in this invention;

[0096] Figure 4 This is a detailed flowchart of the steps in Scenario 2 proposed in this invention;

[0097] Figure 5 This is a flowchart of an example scenario proposed in this invention;

[0098] Figure 6 This is a flowchart of an example of scenario two proposed in this invention. Detailed Implementation

[0099] This invention proposes a method for applying CoT (Coding Instructions) and RAG (Retrieval Enhanced Generation) technologies to a telecommunications operator customer service intelligent system, comprising the following steps:

[0100] Y1: Environment Configuration

[0101] Y10: Configure the Telecom Large Model (Starry Sky) Interface

[0102] Configure the Xingchen Telecom Big Data Language Model ID and API key. This is used for system calls. The Xingchen interface, specifically designed for the telecom industry, is configured here because this model uses a large amount of internal telecom industry data for pre-training to ensure the system can better understand telecom industry-related semantics.

[0103] Scenario 1: A telecom user asks a question for the first time:

[0104] Step Y2: Telecom users ask questions for the first time

[0105] Step Y20: The system detects the problem and triggers the start of the process.

[0106] Note: This step is the starting point of the entire interaction process and needs to ensure that the system can accurately capture the telecom user's intent and question. If the telecom user's question is complex or ambiguous, the model can request the customer to provide additional information. This can be achieved through prompt word engineering for the model. Therefore, this step needs to ensure that the system can receive and correctly parse the telecom user's input. If there are delays or errors in the system's reception or parsing of the telecom user's question, it may be necessary to optimize the front-end input processing logic. This aspect will not be elaborated upon in this article.

[0107] Step Y21: Generate a draft answer

[0108] Step Y210: Using a large-scale telecommunications model to interpret the problems of telecommunications users

[0109] Step Y211: Generate a draft answer using the large telecom model.

[0110] Note: Generating draft answers is a crucial step in the system's understanding of telecom user questions and providing initial feedback. It relies on the language understanding capabilities of the large-scale telecom model to ensure an accurate grasp of user questions.

[0111] Step Y3: Generate query

[0112] Step Y30: Summarize and merge the questions and draft answers.

[0113] Step Y300: Invoke the telecommunications big data model to summarize and synthesize the question and draft answer. Here, the Xingchen telecommunications big data model can be used to understand and generate natural language text. The technology is Natural Language Processing (NLP), particularly text summarization and fusion techniques. The logic is to take the telecommunications user's question and the system's draft answer as input, analyze the semantic content of both through the model, extract key information, and fuse them to generate a coherent text. The algorithm uses a sequence-to-sequence (Seq2Seq) model, which can transform an input sequence into an output sequence, suitable for text summarization and fusion tasks.

[0114] Step Y31: Generate query string

[0115] Step Y310: The telecom big data model is invoked to generate a query string for searching. Information retrieval (IR) technology is used, focusing on generating a query string that effectively retrieves information. The Xingchen big data model analyzes the fused text, extracts keywords and phrases, and constructs the query string. As a language big data model, Xingchen is based on the Transformer architecture, uses an attention mechanism to understand the semantics of language, and generates queries suitable for retrieval.

[0116] Note: The process of receiving input text and generating query strings in a large telecommunications model is actually quite complex.

[0117] Preprocessing: The input text first needs to be preprocessed, which may include word segmentation, stop word removal, stemming, part-of-speech tagging, etc., to prepare it for input into the model.

[0118] Understanding the input: The large telecom model understands the semantic content of the input text by learning language patterns during its training process.

[0119] Information extraction: The model identifies key information in the input text, such as entities, attributes, and intents, which are crucial for generating query strings.

[0120] Query construction: Based on the extracted information, the model constructs one or more query candidates. This process may involve techniques such as template filling, keyword extraction, and phrase recombination.

[0121] Optimize Query (Optional): Uses built-in optimization algorithms to refine the query string and improve its retrieval performance. This may include adjusting the order of keywords, adding synonyms or related words, etc. It may also require formatting, removing redundant words, and ensuring grammatical correctness.

[0122] Output generation: The model generates one or more query strings designed to capture the core intent of the original input text and be used for subsequent information retrieval or database queries.

[0123] Implementation example: A China Telecom user inputs a question: "My mobile phone has been frequently disconnecting from the internet lately, what should I do?" The China Telecom big data model will perform the following operations:

[0124] Preprocessing: The question is segmented into words such as "mobile phone", "recently", "frequently", and "internet disconnection".

[0125] Information extraction: Identify key information, such as "mobile phone" and "internet disconnection".

[0126] Query building: Generate query strings such as "mobile phone network disconnection solution".

[0127] Optimize and generate output: Add "recent" to emphasize time relevance, generating the final query string "recent mobile network outage solutions".

[0128] Step Y4: Load existing data of telecom users

[0129] Step Y40: Load user-specific data, such as package usage and communication habits, via a third-party API to support the generation of personalized answers.

[0130] Note: The loaded data includes, but is not limited to, telecommunications user attributes (public / government / enterprise attributes, network tenure, whether a user is a 5G telecommunications user, age, gender), current telecommunications package (e.g., price range, category, whether it's a bundled package, whether it's an internet card, tier, whether it can be purchased repeatedly), historical records of telecommunications service usage (call volume, SMS volume, data usage, overseas roaming, etc.), and telecommunications user insurance history. This step is necessary because it provides the system with detailed information about the telecommunications user, which is crucial for generating personalized and accurate responses.

[0131] Step Y5: Update the query string

[0132] Step Y50: Analyze telecommunications user data using the large telecommunications model and output the update query string.

[0133] Explanation: Step Y5 is the starting point of the CoT (CoT) thinking chain in this invention. This step is crucial and is one of the innovations of this invention. The query string is not finalized; it needs to be updated by loading telecommunications user data. This allows the system to customize the query string according to the specific circumstances and needs of telecommunications users. Personalized query strings can improve the accuracy and relevance of relevant information retrieved from the knowledge base. Innovation: In step Y5, we first use a large telecommunications model to conduct in-depth analysis of the user's original query. Based on the user's query and existing draft answers, the model generates an initial query string. This process involves not only understanding the user's intent but also identifying telecommunications-specific terms and concepts. This step is the starting point of the CoT thinking chain technology, laying the foundation for subsequent in-depth analysis and answer generation.

[0134] Additional notes: This section lists some possible scenarios, and when these scenarios occur, we can take corresponding solutions.

[0135] (1) The slow response time of third-party APIs affects the overall response speed of the customer service system. We can optimize the API call process, which may include caching telecom user information, asynchronously loading data, or finding a faster API service provider.

[0136] (2) Some aspects of the telecommunications user data are not being fully utilized, or some parts of the data are outdated or inaccurate. A mechanism needs to be established to regularly verify and refresh the telecommunications user data to ensure its accuracy and timely updates.

[0137] (3) The current large-scale telecommunications model performs poorly in analyzing telecommunications user data and generating query strings. First, adjust it using prompt word engineering, then fine-tune the large-scale telecommunications model to better adapt to specific types of telecommunications user queries and data analysis. If neither approach is effective, other models can be selected for testing.

[0138] Step Y6: Query the telecommunications knowledge base

[0139] Step Y60: Invoke the updated query to search for relevant content in the telecommunications knowledge base.

[0140] Step Y61: Set to query once

[0141] Step Y62: The system calculates the similarity based on cosine similarity using the formula: cos(θ)=(A·B) / (|A||B|), where A and B are the TF-IDF vectors of the query string and the knowledge base document, respectively. The similarity between the query string and the document is then calculated.

[0142] Step Y63: Obtain the first search result (the most relevant or most similar answer).

[0143] Note: During the search process in the telecommunications knowledge base, the system will first retrieve the updated query string. Utilizing the telecommunications industry's unique weighting mechanism and deep semantic understanding capabilities, it will retrieve the most relevant documents from the knowledge base. This step ensures that the selected documents are not only highly similar to the query semantically but also closely aligned with the professionalism and practicality of telecommunications services. If the initial search results fail to meet the criteria, the system will adjust its search strategy based on feedback and continue searching until it finds the answer that best meets the user's needs.

[0144] Step Y7: Modify and improve the results

[0145] Step Y70: Based on the understanding of the search results for telecommunications-related knowledge, the large-scale telecommunications model is invoked to modify and improve the draft answer, ensuring the professionalism and accuracy of the answer, and outputting candidate results.

[0146] Note: This is the second innovation of this invention: revising and improving the previous draft answer. The application of CoT (Coding in Thinking) technology is deepened in step Y7, where the large-scale telecommunications model further analyzes and reasons about the information retrieved from the telecommunications knowledge base. The model combines this information with the user's original query and draft answer, gradually constructing a professional and accurate answer through logical reasoning and information integration. Innovation: In step Y7, the large-scale telecommunications model further applies CoT technology to analyze and reason about the information retrieved from the telecommunications knowledge base. The model combines this information with the user's original query and draft answer, gradually constructing a professional and accurate answer through logical reasoning and information integration. This means that the model will simulate the human thought process when solving problems, gradually reasoning and constructing the answer.

[0147] To ensure the compliance of the output content, the following verification steps have been added:

[0148] Step Y8: Load the rule base and perform a compliance check. The rule base contains compliance requirements specific to the telecommunications industry, such as terms of service and user privacy protection standards, ensuring that the response complies with the latest industry regulations, the latest package plans, the latest promotional prices, and the latest service adjustment explanations, etc.

[0149] Step Y80: Example: Are there any prohibited words in the answer? This ensures professionalism and respect in communication.

[0150] Step Y81: Example: Has the recommended service expired? This ensures the timeliness of the recommended service.

[0151] Step Y82: Example: Answer whether there have been any new changes to the service process involved. This ensures the accuracy of the information provided.

[0152] Step Y83: Example: Does the answer comply with telecommunications regulations? This ensures the legality and standardization of the answer.

[0153] Step Y9: Check if the candidate answers are empty. We use a large telecommunications model to perform in-depth semantic understanding of the answers, evaluating whether they provide meaningful content and ensuring that the solution comprehensively covers the user's question. This step is crucial for ensuring user experience, avoiding invalid or empty answers.

[0154] Step Y90: Call the telecommunications big data model to understand the semantics and check whether the semantics contain information.

[0155] Step Y91: If there is information, save the result.

[0156] Step Y92: If no useful information is found, return to step Y63 and obtain the second one.

[0157] Note: If the current candidate answer does not meet the expected information quality and relevance standards, the system will re-evaluate the query string based on the specific needs of the telecommunications industry and return to step Y63 to obtain the next most relevant knowledge base document. During this process, the system will utilize a deep understanding of telecommunications industry terminology and a weighting adjustment mechanism to ensure that the selected document is highly relevant to the user's queried telecommunications service needs.

[0158] Step Y93: Continue in this manner until meaningful content is obtained. Through continuous iteration, a meaningful answer that conforms to both the logical reasoning of the CoT thinking chain and meets compliance requirements is obtained.

[0159] Step Y10: Reply with the final result. This marks the completion of the entire answer generation and verification process. At this point, the intelligent assistant will provide an answer that has undergone logical reasoning, compliance checks, and semantic verification to ensure that it is accurate, professional, legal, compliant, and fully in line with the characteristics of the telecommunications industry and the needs of users.

[0160] Note: This is the third innovative aspect of this invention: performing another round of rule checks on the modified results. This step aligns with the characteristics of the telecommunications industry. It not only ensures the accuracy and professionalism of the answer but also guarantees its legality and compliance with the latest industry regulations. Filtering through the rule base has the following six key benefits:

[0161] Rule check: Loads the telecommunications industry's rule base, which covers key compliance requirements such as terms of service, user privacy protection, the latest package plans, promotional prices, and telecommunications regulations.

[0162] Compliance matching: Matching answers against a rule base to ensure all recommended services and information comply with the latest industry regulations and company policies.

[0163] Risk Assessment: Conduct a risk assessment on the answers to ensure that no prohibited words are used, recommended services are not expired, service processes are accurately updated, and the overall content complies with telecommunications regulations.

[0164] Enhancing Trust: By demonstrating a strong commitment to compliance requirements, smart assistants can increase user trust in the service and improve brand image.

[0165] Improved efficiency: The rule-checking process helps to quickly identify and correct unqualified answers, reducing the need for manual review and improving the efficiency of the overall service process.

[0166] Continuous optimization: Through continuous rule checks and feedback, the intelligent assistant can learn and adapt, constantly optimizing its answer generation mechanism to better meet the specific needs of the telecommunications industry.

[0167] Through these three innovations, this invention not only improves the service quality of the intelligent telecom customer service assistant, but also ensures the security, compliance, and effectiveness of the service, meeting the strict requirements of the telecom industry for high-standard services.

[0168] Scenario 2: The telecom user asks the Nth question:

[0169] (Detailed explanations have already been provided in Scenario 1. Scenario 2 will be briefly introduced below to avoid repetition.)

[0170] Step N2: Telecom users ask questions N times.

[0171] Step N20: The system detects the problem and triggers the start of the process.

[0172] Step N21: Generate a draft answer

[0173] Step N210: Using a large-scale telecommunications model to interpret the problems of telecommunications users

[0174] Step N211: Generate a draft answer using the large telecommunications model.

[0175] Step N3: Generate query

[0176] Step N30: ​​Summarize and merge the questions and draft answers.

[0177] Step N300: Use the large-scale telecommunications model to summarize the synthesis problem and draft answers.

[0178] Step N31: Generate query string

[0179] Step N310: Call the large telecom model to generate a query string for searching.

[0180] Step N4: Load existing data of telecom users

[0181] Step N40: Load telecom user data via third-party API

[0182] Step N5: Update the query string

[0183] Step N50: Analyze telecommunications user data using the large-scale telecommunications model and output the update query string. Step N6: Query the telecommunications knowledge base.

[0184] Step N60: Invoke the updated query to search for relevant content in the telecommunications knowledge base.

[0185] Step N61: Set to query once

[0186] Step N62: Calculate the similarity between the query string and the document.

[0187] Step N63: Get the first search result (the most relevant or most similar answer).

[0188] Step N7: Load conversation history

[0189] Note: The system needs to load all dialogue records related to the current user session. These records include, but are not limited to, the user's previous questions, the system's answers, the user's feedback, and any related interaction information. The purpose of loading dialogue records is to enable the system to:

[0190] (1) Understand the context and development of the user's problem.

[0191] (2) Identify user needs and preference patterns.

[0192] (3) Detect and adapt to the user’s language style and expression habits.

[0193] Step N8: Modify and improve the results

[0194] Step N80: Call the telecom big data model to modify and improve the draft answer based on the above search results and dialogue records, and output candidate results.

[0195] To ensure the compliance of the output content, the following verification steps have been added:

[0196] Step N9: Load the rule base and perform compliance checks.

[0197] Step N90: Example: Are there any banned words in the answer?

[0198] Step N91: Example: Has the recommended service expired?

[0199] Step N92: Example: Answer whether there have been any new changes to the service process involved.

[0200] Step N93: Example: Does the answer comply with telecommunications regulations?

[0201] Step N10: Check if the candidate answer is empty.

[0202] Step N100: Call the telecommunications big data model to understand the semantics and check whether the semantics contain information.

[0203] Step N101: If there is information, save the result.

[0204] Step N102: If no useful information is found, return to step N63 and obtain the second one.

[0205] Step N103: Continue in this manner until informational content is obtained.

[0206] Step N11: Reply with the final result.

[0207] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies, characterized in that, Includes the following steps: Step Y1: Environment Configuration; Step Y2: The telecom user submits their first inquiry; Step Y3: Generate query; Step Y4: Load existing data of telecom users; Step Y5: Update the query string; Step Y6: Query the telecommunications knowledge base; Step Y7: Modify and improve the results; Step Y8: Load the rule base and perform compliance checks; Step Y9: Check if the candidate answer is empty; Step Y10: Reply with the final result.

2. The intelligent telecommunications customer service system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, This also includes the Nth question from the telecom user, specifically including the following steps: Step N1: Environment Configuration; Step N2: The telecom user asks questions N times; Step N20: The system detects the problem and triggers the start of the process; Step N21: Generate a draft answer; Step N210: Use the large-scale telecommunications model to interpret the problems of telecommunications users; Step N211: Generate a draft answer using the large telecommunications model; Step N3: Generate query; Step N30: ​​Summarize and merge the questions and draft answers; Step N300: Use the large-scale telecommunications model to summarize the synthesis problem and draft answers; Step N31: Generate the query string; Step N310: Invoke the large telecom model to generate a query string for searching; Step N4: Load existing data of telecom users; Step N40: Load telecom user data via third-party API; Step N5: Update the query string; Step N50: Analyze telecommunications user data using the large telecommunications model and output the update query string; Step N6: Query the telecommunications knowledge base; Step N60: Invoke the updated query to search for relevant content in the telecommunications knowledge base; Step N61: Set to query once; Step N62: Calculate the similarity between the query string and the document; Step N63: Obtain the first search result; Step N7: Load the conversation history; Step N8: Modify and improve the results; Step N80: Call the telecom big data model to modify and improve the draft answer based on the above search results and dialogue records, and output candidate results; Step N9: Load the rule base and perform compliance checks; Step N10: Check if the candidate answer is empty; Step N11: Reply with the final result.

3. The intelligent telecommunications customer service system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y2: The telecom user asks a question for the first time, including: Step Y20: The system detects the problem and triggers the start of the process; Step Y21: Generate a draft answer; Step Y210: Use the large-scale telecommunications model to interpret the problems of telecommunications users; Step Y211: Generate a draft answer using the large telecom model.

4. The intelligent telecommunications customer service system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y3: Generate a query, including: Step Y30: Summarize and merge the questions and draft answers; Step Y300: Use the large-scale telecommunications model to summarize the synthesis problem and draft answers; Step Y31: Generate the query string; Step Y310: Call the large telecom model to generate a query string for searching.

5. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y4: Load existing data of telecom users, including: Step Y40: Load user-specific data via a third-party API.

6. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y5: Update the query string, including: Step Y50: Analyze telecommunications user data using the large telecommunications model and output the update query string.

7. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y6: Search the telecommunications knowledge base, including: Step Y60: Invoke the updated query to search for relevant content in the telecommunications knowledge base; Step Y61: Set to query once; Step Y62: The system calculates the similarity between the query string and the document based on the cosine similarity formula: cos(θ)=(A·B) / (|A||B|), where A and B are the TF-IDF vectors of the query and knowledge base documents, respectively. Step Y63: Obtain the first search result.

8. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y7: Modify and improve the results, including: Step Y70: Based on the understanding of the search results for telecommunications-related knowledge, the large-scale telecommunications model is invoked to modify and improve the draft answer, ensuring the professionalism and accuracy of the answer, and outputting candidate results.

9. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y8: Load the rule base and perform compliance checks, including: Step Y80: Are there any banned words in the answer? Step Y81: Has the recommended service expired? Step Y82: Answer whether there have been any new changes to the service process involved; Step Y83: Determine whether the answer complies with telecommunications regulations.

10. A telecommunications customer service intelligent system based on the integration of CoT and RAG technologies according to claim 1, characterized in that, Step Y9: Check if the candidate answer is empty, including: Step Y90: Call the telecommunications big data model to understand the semantics and check whether the semantics contain information; Step Y91: If there is information, save the results; Step Y92: If no useful information is found, return to step Y63 and obtain the second one; Step Y93: Continue in this manner until meaningful content is obtained.

Images