Communication service request processing method, system and device and storage medium

Abstract

The invention provides a communication service request processing method, system and device and a storage medium, after a communication service request is preliminarily processed to obtain related information, similarity retrieval is performed on the related information, and a formatted retrieval result group is generated based on a preset condition and a similarity retrieval result; the complexity of related information can be simplified, and the semantic understanding deviation condition of processing the related information by the system is reduced; besides, the formatted retrieval result group is verified based on a preset algorithm and a preset condition, it is ensured that the formatted retrieval result group is generated based on the user communication service request, and the situation that the matching result of related information is disordered due to the fact that the system faces a large number of service requests at the same time can be reduced.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a communication service request processing method, system, device and storage medium. Background Art

[0002] In the current digital age, communication business processing methods, relying on advanced technology background, are gradually showing their strong potential. With the rapid development of big data, cloud computing and artificial intelligence technology, communication business processing systems can deeply understand complex and changing business scenarios. Big data analysis can dig out key information such as user behavior and market trends, and provide accurate data support for decision-making; while artificial intelligence algorithms can simulate human thinking, predict and evaluate potential business risks, and ensure the accuracy of decision-making.

[0003] The existing communication service processing system has also demonstrated outstanding capabilities in processing a large number of communication service requests. By building an efficient network architecture, such as distributed systems and load balancing technology, the system can quickly respond to and process service requests, ensuring the stability and reliability of communication services. At the same time, the application of automated communication service processing systems and intelligent decision-making engines has further improved processing efficiency, reduced manual intervention, and reduced operating costs.

[0004] In the existing communication service processing process, the communication service request is first preliminarily processed to extract relevant information, and then the system identifies and matches the relevant information to obtain the corresponding processing results and outputs them. However, there are certain problems in the process of the system identifying and matching relevant information. On the one hand, when faced with relevant information after processing complex business requests, the system cannot match the correct processing results, and it is easy to have deviations in the semantic understanding of relevant information; on the other hand, with the continuous expansion of communication services, the amount of data has exploded, posing a higher challenge to the system's processing capabilities. When a large number of users' communication service requests need to be processed at the same time, the burden on the system will increase, resulting in disorder in the processing results obtained by the system matching relevant information. Summary of the invention

[0005] In view of this, in order to solve one of the above problems, the purpose of an embodiment of the present invention is to provide a communication service request processing method, system, device and storage medium, which can reduce the deviation in semantic understanding and matching disorder of the relevant information obtained after preliminary processing of the communication service request.

[0006] In a first aspect, the present invention provides a method for processing a communication service request, the method comprising the following steps:

[0007] Obtaining a communication service request, and determining a query vector according to the communication service request;

[0008] Performing a similarity search on the query vector based on a preset vector database to obtain a similarity search result;

[0009] Generate a formatted search result group based on the similarity search result, the first preset condition and the second preset condition;

[0010] The formatted search result group is verified according to the communication service request and the first preset condition, and a processing result is output.

[0011] The present invention provides a communication service request processing method. After preliminarily processing the communication service request to obtain relevant information, the method performs similarity search on the relevant information and generates a formatted search result group based on preset conditions and similarity search results, thereby simplifying the complexity of the relevant information and reducing the semantic understanding deviation of the system in processing the relevant information. In addition, the formatted search result group is verified based on a preset algorithm and preset conditions to ensure that the formatted search result group is generated based on the user's communication service request, thereby avoiding the situation where the system encounters a large number of service requests at the same time and the related information matching results are disordered.

[0012] Optionally, determining a query vector according to the communication service request includes:

[0013] Based on the communication service request, obtaining key information;

[0014] Parsing the communication service request based on the key information to generate a structured query statement;

[0015] Semantic similarity matching is performed based on the structured query statement to generate the query vector.

[0016] Optionally, performing similarity search on the query vector based on a preset vector database to obtain a similarity search result includes:

[0017] Parsing the query vector based on the preset vector database;

[0018] Perform similarity search based on the analysis result and return matching results;

[0019] The matching results are screened based on a preset standard, and the similarity search results are obtained based on the screened results.

[0020] Optionally, the preset vector database is constructed by the following method:

[0021] Collect data based on preset data sources to obtain original data;

[0022] Cleaning the original data, and structuring the cleaned original data to obtain structured data;

[0023] Based on the trained conditional random field, the text data in the structured data is split to obtain word block data;

[0024] The word block data is converted into vector data based on the trained text embedding model to form a preset vector database.

[0025] Optionally, generating a formatted search result group based on the similarity search result, the first preset condition and the second preset condition includes:

[0026] Calculating the parameters to be examined and the temporary search parameters based on the similarity search result and the first preset condition;

[0027] If the parameter to be examined and the temporary search parameter meet a second preset condition, a formatted search result group is generated based on the similarity search result, the parameter to be examined and the temporary search parameter.

[0028] Optionally, the calculating of the to-be-examined parameter and the temporary search parameter based on the similarity search result and the first preset condition includes:

[0029] Based on the first preset condition, obtaining a temporary token value;

[0030] Calculate a first parameter based on the temporary token value and a first preset formula;

[0031] Calculate the parameters to be reviewed based on the first parameter and the second preset formula;

[0032] A temporary search parameter is calculated based on the similarity search result, the first parameter, the parameter to be examined and a third preset formula.

[0033] Optionally, verifying the formatted search result group according to the communication service request and the first preset condition and outputting a processing result includes:

[0034] Based on the first preset condition, checking the temporary search parameter and the parameter to be examined;

[0035] The inspection result meets the first preset condition, and the audit factor is calculated based on a fourth preset formula and the first preset condition;

[0036] Calculate a search result summary value based on the hash function and the similarity search result;

[0037] Calculate a first auxiliary audit parameter based on a fifth preset formula, the search result summary value and the temporary search parameter;

[0038] Calculate a second auxiliary review parameter based on a sixth preset formula, the parameter to be reviewed and the temporary search parameter;

[0039] Calculate a third auxiliary audit parameter based on a seventh preset formula, the first auxiliary audit parameter, the second auxiliary audit parameter, and the audit factor;

[0040] Calculating an audit parameter based on an eighth preset formula and the third auxiliary audit parameter;

[0041] If the parameters to be examined are the same as the audit parameters, a processing result is output based on the formatted search result group.

[0042] In a second aspect, the present invention provides a communication service request processing system, characterized in that it includes:

[0043] The first module is used to obtain a communication service request and determine a query vector according to the communication service request;

[0044] The second module is used to perform similarity search on the query vector based on a preset vector database to obtain a similarity search result;

[0045] A third module is used to generate a formatted search result group based on the similarity search result, the first preset condition and the second preset condition;

[0046] The fourth module is used to verify the formatted search result group according to the communication service request and the first preset condition, and output a processing result.

[0047] In a third aspect, the present invention provides a communication service request processing device, comprising:

[0048] at least one processor;

[0049] at least one memory for storing at least one program;

[0050] When the at least one program is executed by the at least one processor, the at least one processor implements the processing method as described above.

[0051] In a fourth aspect, a computer-readable storage medium stores a program executable by a processor, wherein the program executable by the processor is used to execute the processing method described above when executed by the processor.

[0052] In summary, the beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0053] The present invention provides a communication service request processing method, system, device and storage medium. After preliminarily processing the communication service request to obtain relevant information, the method performs similarity search on the relevant information and generates a formatted search result group based on preset conditions and similarity search results, thereby simplifying the complexity of the relevant information and reducing the semantic understanding deviation of the system in processing the relevant information. In addition, the formatted search result group is verified based on a preset algorithm and preset conditions to ensure that the formatted search result group is generated based on the user's communication service request, which can reduce the disorder of the relevant information matching results caused by the system facing a large number of service requests at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the steps of a method for processing a communication service request provided by an embodiment of the present invention;

[0055] Figure 2 It is a schematic diagram of a step flow of determining a query vector according to a communication service request provided by an embodiment of the present invention;

[0056] Figure 3 It is a schematic flow chart of a step of performing similarity search on a query vector based on a preset vector database and obtaining a similarity search result provided by an embodiment of the present invention;

[0057] Figure 4 It is a schematic diagram of a step flow of a method for forming a preset vector database provided by an embodiment of the present invention;

[0058] Figure 5 It is a schematic diagram of a step flow of a method for generating a formatted search result group provided by an embodiment of the present invention;

[0059] Figure 6 It is a schematic diagram of a step flow of a method for calculating parameters to be examined and temporary search parameters provided by an embodiment of the present invention;

[0060] Figure 7 It is a schematic diagram of the steps of a method for verifying a formatted search result group provided by an embodiment of the present invention;

[0061] Figure 8 It is a system architecture block diagram of a communication service request processing system provided by an embodiment of the present invention;

[0062] Fig. 9 It is a structural block diagram of a communication service request processing system provided by an embodiment of the present invention;

[0063] Fig.10 It is a schematic diagram of the flow structure of another communication service processing method provided by an embodiment of the present invention;

[0064] Fig.11It is a schematic diagram of the steps of another method for processing a communication service request provided by an embodiment of the present invention;

[0065] Fig.12 It is a structural diagram of a communication service request processing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0067] First, the technical field and technical background of this application design are introduced:

[0068] Communication service request processing refers to the communication enterprises such as China Mobile, China Unicom, China Telecom, etc., formulating and selecting the best solutions or processing results for customer problems or needs during the course of business operations; the communication service request processing system provides an interactive platform for customers to input communication service request data. Based on the preset database and algorithm, the system processes the information of the questions or needs raised by the customers and matches the processed information to obtain and output the processing results. Application examples include "intelligent customer service" and "risk assessment".

[0069] Digital AI technology is a cutting-edge technology that integrates digitization and intelligence. It aims to efficiently process and analyze massive amounts of data through advanced algorithms and models, and realize intelligent decision-making and automated operations. This application provides a communication service request processing method, system, device and storage medium, which are based on digital AI technology. Through the collaborative work of multiple core modules, and the use of multiple intelligent data models (such as RAG model, LLM model, etc.) and AI intelligent technology, intelligent decision support from user input to data analysis and feedback is realized.

[0070] Several terms involved in this application are explained as follows:

[0071] Natural language processing (NLP) technology: As a core branch of artificial intelligence, it is dedicated to enabling computers to understand and process human natural language. Through the comprehensive application of deep learning, statistical methods and linguistic rules, NLP can achieve a full range of functions from text preprocessing, semantic understanding to text generation. It is widely used in machine translation, sentiment analysis, question-answering systems, text classification and other fields, greatly promoting the intelligent process of human-computer interaction and providing strong support for information retrieval, data mining and intelligent decision-making.

[0072] RAG (Retrieval-Augmented Generation) model: A technical architecture that combines information retrieval and generation capabilities, designed to enhance the output of the generation model by retrieving relevant information from an external knowledge base. It uses the generation capabilities of large language models and introduces a retrieval mechanism to obtain more accurate and relevant information, improving the quality and accuracy of generated content. It is widely used in multiple fields such as question-answering systems, chatbots, and content creation.

[0073] LLM (Large Language Model) model: A large natural language processing model based on deep learning, with a large number of parameters and powerful language understanding capabilities. It can learn the grammar, semantics and context information of the language by training massive text data, and can handle various natural language processing tasks such as text generation, machine translation, question-answering systems, etc. LLM models usually have billions to hundreds of billions of parameters and can generate human-readable and coherent text content.

[0074] ChatGLM3 model: A pre-trained dialogue model based on the Transformer architecture, containing about 600 million parameters, with powerful language understanding and generation capabilities. The model is pre-trained on a large-scale text corpus and can handle a variety of complex language tasks. ChatGLM3 provides a rich API interface and supports multiple development methods. It can be applied to a variety of practical scenarios such as intelligent customer service, knowledge question and answer, and text creation.

[0075] M3E text embedding model: an advanced Chinese embedding model that supports both Chinese and English text processing. The model uses a contrastive learning method and performs deep training on a large-scale sentence pair dataset, which can efficiently convert natural language text into dense vector representations. The M3E model not only has powerful text semantic understanding capabilities, but also supports multiple functions such as homogeneous text similarity calculation and heterogeneous text retrieval, and can efficiently solve natural language related problems.

[0076] Chroma Vector Database: A database system designed for storing, managing, and efficiently retrieving embedded vectors. It uses advanced vector indexing technology to support fast similarity search of large-scale data sets. It is widely used in image recognition, natural language processing, recommendation systems, and intelligent search engines, providing users with efficient, scalable, and flexible vector data storage and retrieval solutions.

[0077] Hash Function: Also known as hash function or hash algorithm, it is a function that maps input data of arbitrary length (usually called "message" or "keyword") to output data of fixed length (called "hash value" or "hash value" or "message digest").

[0078] Data disorder: Data disorder in the communication service request processing system refers to the situation that when using the system, due to various reasons (such as system design defects, network problems, operational errors, etc.), the processing results are inconsistent with the received data, lost, wrong or abnormal. These disordered data will directly affect the normal operation of the system and the smooth progress of the business process.

[0079] Vector space model Word2Vec: A technology that represents vocabulary as vectors in a high-dimensional space. By training a large amount of text data, each word is mapped to a continuous vector of a fixed length, so that semantically similar words are also close in distance in the vector space. It is widely used in tasks such as text classification, sentiment analysis, and machine translation in natural language processing.

[0080] API (Application Programming Interface) format: refers to a data format that defines a set of rules and structures for communication and data exchange between different software applications. These formats usually include the structure of requests and responses, data types, field names and values, etc., to ensure that data is correctly parsed and processed between different systems. Common API formats include JSON, XML, etc.

[0081] Deduplication algorithm: A data processing technology that aims to identify and remove duplicate elements or records from a data set to ensure the uniqueness and accuracy of the data. This technology is widely used in data processing, set operations, data cleaning and other scenarios.

[0082] Conditional Random Field (CRF): A discriminative probability model that is often used to annotate or analyze sequence data, such as natural language text or biological sequences. It combines the characteristics of the maximum entropy model and the hidden Markov model, can express long-distance dependencies and overlapping features, and has achieved remarkable results in sequence annotation tasks such as part-of-speech tagging and named entity recognition in natural language processing.

[0083] like Figure 1 As shown, Figure 1 1 is a flow chart of steps of a method for processing a communication service request provided by an embodiment of the present invention, which includes the following steps:

[0084] S100: Obtain a communication service request, and determine a query vector according to the communication service request;

[0085] Acquire data on the communication service request submitted by the user, where the communication service request data source includes: natural language text (such as "Please provide me with a decision-making report on the current communication network operation status") or select a specific service option (such as "intelligent customer service", "risk assessment", etc.); based on the acquired communication service request data, extract key information from the data, perform data parsing and matching, and obtain a query vector corresponding to the communication service request.

[0086] S200: performing a similarity search on the query vector based on a preset vector database to obtain a similarity search result;

[0087] The query vector is encapsulated into a form acceptable to a vector database (such as a specific API form); the query vector is parsed through a preset vector database and a similarity search is performed on the parsed result, specifically: corresponding features in the query vector are extracted and matched with corresponding features in a preset vector database; preliminary screening is performed based on the matching relevance; results whose relevance does not meet the requirements are screened out and arranged according to the matching relevance; and similarity retrieval results are obtained based on the arranged result.

[0088] S300: generating a formatted search result group based on the similarity search result, the first preset condition and the second preset condition;

[0089] By presetting conditions and algorithms, corresponding calculation parameters are constructed, and the similarity retrieval results are formatted based on the obtained similarity retrieval results and the constructed calculation parameters to obtain simplified retrieval results. In the subsequent process of matching the retrieval results to obtain processing results, the complexity of related information can be simplified and the semantic deviation of the system in processing related information can be reduced.

[0090] S400: verifying the formatted search result group according to the communication service request and the first preset condition, and outputting a processing result.

[0091] The formatted search result group is verified based on the preset algorithm and preset conditions. Specifically, the corresponding verification parameters are calculated based on the preset algorithm, the formatted search results and the constructed calculation parameters, and it is determined whether the relationship between the verification parameters and the constructed calculation parameters meets the preset conditions. If so, it can be determined that the obtained formatted search result group corresponds to the communication service request, and the processing result is output.

[0092] In some embodiments, Figure 2 As shown, the process of determining the query vector according to the communication service request in step S100 can be implemented by the following steps:

[0093] S110: Acquire key information based on the communication service request;

[0094] Extract key information from the communication service requests submitted by users, specifically identify the intent and subject of the communication service requests, segment the submitted text, remove stop words, and perform semantic understanding.

[0095] S120: parsing the communication service request based on the key information to generate a structured query statement;

[0096] Natural language processing (NLP) technology is used to parse the key information of user queries, extract keywords, phrases and topics, and build structured query statements based on the analysis results.

[0097] S130: Perform semantic similarity matching based on the structured query statement to generate the query vector.

[0098] Use the vector space model (Word2Vec) to perform semantic similarity matching, find the corresponding Word2Vec vector for each word in the structured query statement, average and sum the word vectors, and get the vector representation of the entire query:

[0099]

[0100] Wherein, N is the number of words in the query, and wi is each word; optionally, the vector space model includes a Word2Vec vector space model.

[0101] In some embodiments, Figure 3 As shown, the retrieval process of performing similarity retrieval on the query vector based on the preset vector database and obtaining the similarity retrieval result in step S200 can be implemented by the following steps:

[0102] S210: parsing the query vector based on the preset vector database;

[0103] The generated query vector is encapsulated into a format acceptable to the vector database (specific API format), including the query vector and related meta-information, and the communication service request is sent to the vector database. The vector database receives the request and parses the query vector to obtain the retrieval vector.

[0104] S220: Perform similarity search based on the analysis result and return matching results;

[0105] Based on the parsed retrieval vector, a similarity search is performed according to the following formula, and the vector database returns several document IDs with the highest similarity to the query vector and their similarity scores:

[0106]

[0107] Where wi is the search vector, wj is the vector in the vector database, ||wi||, ||wj|| are the moduli of vectors wi and wj;

[0108] S230: Filter the matching results based on a preset standard, and obtain the similarity search results based on the filtered results.

[0109] Receive the similarity score results returned by the vector database, including the ID and similarity score of the relevant documents, perform preliminary filtering, remove the results that are irrelevant to the query or of low quality, sort the results, and arrange them in descending order by similarity score to ensure that the most relevant information is displayed first, and obtain the similarity retrieval results.

[0110] In some embodiments, Figure 4 As shown, the preset vector database in step S210 can be constructed by the following method:

[0111] S211: Collect data based on preset data sources (including but not limited to industry databases, public APIs, and real-time data streams) to obtain original data.

[0112] S212: Clean the original data, and structure the cleaned original data to obtain structured data; data cleaning includes but is not limited to deduplication, filtering irrelevant information, and processing missing values.

[0113] S213: Based on the trained conditional random field (CRF), the text data in the structured data is split to obtain word block data;

[0114] S214: Based on the trained M3E text embedding model, the word block data is converted into vector data through a data vectorization network to form a preset vector database.

[0115] In some embodiments, Figure 5 As shown, the method of generating the formatted search result group in step S300 can be implemented by the following steps:

[0116] S310: Calculating the parameters to be examined and the temporary search parameters based on the similarity search result and the first preset condition;

[0117] S320: If the parameter to be examined and the temporary search parameter meet a second preset condition, a formatted search result group is generated based on the similarity search result, the parameter to be examined and the temporary search parameter.

[0118] In some embodiments, Figure 6 As shown, the method of calculating the parameters to be examined and the temporary search parameters in step S310 can be set according to a preset algorithm based on the RAG model, and the setting method includes the following steps:

[0119] S311: Obtain the string form of the similarity search result, defined as resulti;

[0120] S312: Randomly select a temporary token value Tokeni that meets the first preset condition [1, n-1];

[0121] S313: Calculate a first parameter Pconi according to a first preset formula Pconi=Tokeni×P;

[0122] Wherein, P is a base point of order n on the elliptic curve E(Fp), E(Fp) is an elliptic curve defined on a finite field Fp, p is a large prime number, and the elliptic curve is expressed as: y2=x3+ax+b(mod p), a, b∈Fp and are constants, for any (x, y)∈E(Fp);

[0123] S314: Calculate the parameter to be examined PendPi according to the second preset formula PendPi=x(Pconi)(mod n);

[0124] Wherein, x(Pconi) is the x coordinate of a point Pconi on the elliptic curve E(Fp). If PendPi=0, a temporary token value is reselected for calculation;

[0125] S315: Calculate the temporary search parameter sti according to the third preset formula sti=(Tokeni)-1×[H(resulti)+d×(PendPi)](mod n), where H is a hash function; if sti=0, reselect the temporary token value for calculation;

[0126] S316: Construct a formatted search result group {PendPi, sti, resulti} based on the similarity search results, the parameters to be reviewed and the temporary search parameters.

[0127] It should be noted that the formatted search results are as follows, taking the decision report on the current communication network operation status queried by the user as an example:

[0128] {

[0129] "user_query": "Please provide a decision report on the current operation status of the communication network.",

[0130] "retrieved_info": [{

[0131] "title": "Equipment Operation Status Analysis",

[0132] “content”: “The average uptime of base station equipment in a certain area is 98%.”

[0133] },

[0134] {

[0135] "title": "Network Performance Report",

[0136] "content": "Network latency has averaged 45ms over the past 7 days, exceeding the standard threshold by 10%."]

[0137] };

[0138] In some embodiments, Figure 7 As shown, in step S400, the formatted search result group is verified according to the communication service request and the first preset condition, and the processing result is output, which can be implemented by the following steps:

[0139] S410: According to the first preset condition [1, n-1], check whether sti, PendPi∈[1, n-1] are established. If not, the verification fails;

[0140] S420: Calculate the audit factor FAuditi according to a fourth preset formula FAuditi=dP;

[0141] Where, d∈[1,n-1]; P is a base point of order n on the elliptic curve E(Fp), E(Fp) is an elliptic curve defined on a finite field Fp, p is a large prime number, and the elliptic curve is represented by: y2=x3+ax+b(mod p), a, b∈Fp and are constants, for any (x, y)∈E(Fp);

[0142] S430: Calculate the search result summary value Absi=H(resulti) according to the hash function and the string form of the similarity search result resulti;

[0143] S440: Calculate the first auxiliary audit parameter IRAui according to the fifth preset formula IRAui=Absi×(sti)-1(mod n);

[0144] S450: Calculate the second auxiliary audit parameter IIRAui according to the sixth preset formula IIRAui=PendPi×(sti)-1(mod n);

[0145] S460: Calculate the third auxiliary audit parameter IIIRAui according to the seventh preset formula IIIRAui=ⅠRAui×P+ⅡRAui×FAuditi:

[0146] S470: Calculate the audit parameter PAui according to the eighth preset formula PAui=x(IIIRAui)(mod n);

[0147] S480: If the equation PAui=PendPi, the verification is passed;

[0148] If the equation is satisfied, it can be confirmed that the received search result group {PendPi, sti, resulti} is generated based on the user's communication service request and the system data is not disordered.

[0149] S490: After verification, output the processing result.

[0150] After confirmation and verification, the large language model is called to perform text generation operations. Based on the prompt template and the acquired communication service request data, the generated text is basically cleaned up to remove extra spaces, punctuation or incoherent sentences. According to the user's communication service request, the goal of generating text is clarified, the user's original query is combined with the retrieved information to form a complete context, and the processing results are output.

[0151] like Figure 8 As shown, Figure 8 It is a system architecture block diagram of a communication service request processing system provided by an embodiment of the present invention. The system plans the capability architecture based on the AI ​​big model as the underlying driver. The core modules of the overall architecture include but are not limited to RAG (Retrieval-Augmented Generation, RAG), LLM (Large Language Model), model training and fine-tuning, and industry applications, so as to realize the upper-level application functions of vertical fields, data analysis, and industry reports.

[0152] like Fig. 9 As shown, Fig. 9 It is a structural block diagram of a communication service request processing system provided by an embodiment of the present invention, including a user interaction module, a vector database, a RAG module, an LLM module, a model training and fine-tuning module, a prompt template module and a log file database;

[0153] The user interaction module is used to provide a user interaction interface, support natural language input and query, and display analysis results, reports and visualized data;

[0154] The vector database is used to store vector forms of data from a preset data source;

[0155] The RAG module is used to quickly retrieve relevant information from the vector database and generate a text reply based on the retrieval results and the communication service request data entered by the user;

[0156] The LLM module is used to process natural language understanding and generation, support dialogue systems and question-answering systems, and generate text data;

[0157] The prompt template module is used to guide the LLM module to generate templated output;

[0158] The model training and fine-tuning module supports large-scale model training and fine-tuning, optimizes model performance, and performs model evaluation and tuning to ensure that it meets the specific needs of communication network operation and maintenance decision-making;

[0159] The log file database is used to record the log files generated during the operation of each module of the system.

[0160] like Fig.10 is a flow chart of another communication service processing method provided by an embodiment of the present invention. Fig.11 is a schematic diagram of the steps of another method for processing a communication service request provided by an embodiment of the present invention, combined with Fig.10 and Fig.11 The specific implementation steps of the communication service processing method are as follows:

[0161] Step S501: The user submits a communication service request through the user interaction module, inputs a natural language text or selects a specific service option, and the user interaction module transmits the user communication service request to the RAG module.

[0162] Step S502: The RAG module quickly retrieves information related to the user query from the vector database.

[0163] The specific implementation process is as follows:

[0164] The RAG module extracts key information from user input, identifies the intent and subject of the query, tokenizes the query text, removes stop words, and performs semantic understanding.

[0165] Furthermore, the RAG module uses natural language processing (NLP) technology to parse user queries, extract keywords, phrases and topics, and construct structured query statements based on the analysis results.

[0166] Furthermore, the RAG module uses the vector space model (Word2Vec) to perform semantic similarity matching on structured query statements, specifically:

[0167] Find the corresponding Word2Vec vector for each word, average and sum the word vectors to get the vector representation of the entire query:

[0168]

[0169] Where N is the number of words in the query, and wi is each word;

[0170] Furthermore, the RAG module encapsulates the generated query vector into a format acceptable to the vector database (specific API format), including the query vector and related meta-information, and sends the communication service request to the vector database. The vector database receives the request and parses the query vector, performs a similarity search, and returns several document IDs with the highest similarity to the query vector and their similarity scores:

[0171]

[0172] Where wi is the search vector, wj is the vector in the vector database, ||wi||, ||wj|| are the moduli of vectors wi and wj;

[0173] Furthermore, the RAG module receives the results returned by the vector database, including the IDs and similarity scores of relevant documents, performs preliminary filtering, removes results that are irrelevant to the query or of low quality, sorts the results, and arranges them in descending order of relevance score to obtain similarity retrieval results.

[0174] In it, the construction process of the vector database is as follows:

[0175] (1) Data extraction: calling the preset data source for data collection;

[0176] (2) Data processing: The collected raw data enters the data processing process, performs data cleaning, and structures the processed data;

[0177] (3) Text segmentation: Based on the conditional random field (CRF), the text data is split into word chunks according to the word segmentation rules;

[0178] (4) Vectorization: Use the pre-trained M3E text embedding model and convert the text into vectors through the data vectorization network;

[0179] (5) Vector data storage: vector data is stored in the database to form the Chroma vector database.

[0180] Among them, the preset data sources include but are not limited to industry databases, public APIs and real-time data streams.

[0181] Among them, data cleaning includes but is not limited to deduplication, filtering irrelevant information, and processing missing values.

[0182] Furthermore, a deduplication algorithm is applied to remove duplicate entries from similarity search results to ensure the diversity of results, organize the processed search results into a structured format, including but not limited to document title, abstract, relevance score, keywords, link information, construct the parameters to be reviewed, generate a formatted search result group, and pass the formatted search results to the LLM module.

[0183] It should be noted that the formatted search results are as follows, taking the decision report on the current communication network operation status queried by the user as an example:

[0184]

[0185]

[0186] Step S503: The LLM module receives the formatted search result group, verifies its matching with the user's actual communication service request, and outputs the processing result after the verification is passed.

[0187] After confirming that the received search result group is generated based on the user's communication service request, the search results obtained from the search enhancement module include multiple information fragments, including but not limited to document titles, abstracts, relevance scores, keywords, link information, and the goal of generating text is clearly defined according to the user's communication service request, combining the user's original query with the retrieved information to form a complete context;

[0188] Furthermore, the LLM module calls the prompt template module to obtain the output format of the model, including: adding the beginning prompt and integrating the context information to make it coherent and logical;

[0189] The prompt template is as follows:

[0190] Your query was: "{user_query}".

[0191] Here is the relevant information:

[0192] 1.{retrieved_info[0].title}:{retrieved_info[0].content};

[0193] 2.{retrieved_info[1].title}:{retrieved_info[1].content}; ...

[0195] i.{retrieved_info[i].title}:{retrieved_info[i].content}; ...

[0197] m.{retrieved_info[m].title}:{retrieved_info[m].content};

[0198] Wherein, i.{retrieved_info[i].title represents the i-th output topic generated based on the user's communication service request, retrieved_info[i].content represents the i-th output content body;

[0199] Furthermore, based on the ChatGLM3 model, the LLM module performs text generation operations and, combined with the prompt template, performs basic cleaning of the generated text to remove extra spaces, punctuation, or incoherent sentences;

[0200] For example, the LLM module generates a decision report based on the prompt template and input data as follows:

[0201] “**Communication Network Operation and Maintenance Decision Report**

[0202] Based on your request, the following is a summary of the operation and maintenance decisions for the communication network:

[0203] **Equipment operating status**:

[0204] The average uptime of base station equipment in a certain area is 98%, and the overall performance is stable, but it should be noted that the operating status of individual equipment fluctuates slightly, and regular inspection and maintenance may be required to ensure continued normal operation;

[0205] **Network Performance Analysis**:

[0206] The average network latency in the last 7 days is 45ms, which is 10% higher than the standard threshold. This indicates that there may be bandwidth pressure or network congestion in the area. It is recommended to optimize traffic distribution or increase bandwidth resources.

[0207] **Suggestions for improvement**:

[0208] For the operating status of base station equipment, it is recommended to conduct a comprehensive inspection every month, especially for base stations in high-load areas;

[0209] To address the network latency issue, we recommend further analyzing traffic distribution and conducting traffic diversion, and increasing network capacity or performing hardware upgrades if necessary.

[0210] Furthermore, the LLM module returns the generated text results to the user interaction module for the user to view and use. At this time, the text generation results will be integrated into the corresponding part of the user interface and provided to the user;

[0211] Step S504: the user interaction module displays the text results output by the LLM module, and the user further interacts through the user interaction module, including but not limited to requesting more details and adjusting query parameters.

[0212] Step S505: The user evaluates or provides feedback on the results. The user interaction module provides a simple feedback mechanism (including but not limited to star ratings and text comments), and the collected feedback information is transmitted to the model training and fine-tuning module to ensure continuous improvement of the system.

[0213] Step S506: The model training and fine-tuning module regularly evaluates the large language model based on the collected user feedback and usage data, identifies areas where the language model performance is insufficient, and performs fine-tuning (including but not limited to retraining and adjusting hyperparameters); the fine-tuned large language model can be reloaded into the LLM module to improve its performance in actual applications.

[0214] Step S507: the log file database records the above user communication service request and feedback optimization process, and records the log files generated by each module.

[0215] like Fig.12 As shown, the embodiment of the present invention also provides a communication service request processing device, including:

[0216] at least one processor;

[0217] at least one memory for storing at least one program;

[0218] When the at least one program is executed by the at least one processor, the at least one processor implements the steps of a communication service request processing method described in the above method embodiment.

[0219] Among them, the memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. The memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a remote memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0220] It can be seen that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0221] The embodiment of the present invention further provides a computer-readable storage medium, which stores a program executable by a processor, and the program executable by the processor is used to implement the above method when executed by the processor. Similarly, the contents of the above method embodiment are all applicable to the present storage medium embodiment, and the functions specifically implemented by the present storage medium embodiment are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method embodiment.

[0222] It is understood that all or some steps and systems in the disclosed method above can be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, magnetic tape, disk storage or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0223] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for processing a communication service request, characterized in that: include: Obtaining a communication service request, and determining a query vector according to the communication service request; Performing a similarity search on the query vector based on a preset vector database to obtain a similarity search result; Generate a formatted search result group based on the similarity search result, the first preset condition and the second preset condition; The formatted search result group is verified according to the communication service request and the first preset condition, and a processing result is output.

2. The processing method according to claim 1, characterized in that: The determining a query vector according to the communication service request comprises: Based on the communication service request, obtaining key information; Parsing the communication service request based on the key information to generate a structured query statement; Semantic similarity matching is performed based on the structured query statement to generate the query vector.

3. The processing method according to claim 1, characterized in that: The performing similarity search on the query vector based on a preset vector database to obtain a similarity search result includes: Parsing the query vector based on the preset vector database; Perform similarity search based on the analysis result and return matching results; The matching results are screened based on a preset standard, and the similarity search results are obtained based on the screened results.

4. The processing method according to claim 1, characterized in that: The preset vector database is constructed by the following method: Collect data based on preset data sources to obtain original data; Cleaning the original data, and structuring the cleaned original data to obtain structured data; Based on the trained conditional random field, the text data in the structured data is split to obtain word block data; The word block data is converted into vector data based on the trained text embedding model to form a preset vector database.

5. The processing method according to claim 1, characterized in that: The step of generating a formatted search result group based on the similarity search result, the first preset condition and the second preset condition comprises: Calculating the parameters to be examined and the temporary search parameters based on the similarity search result and the first preset condition; If the parameter to be examined and the temporary search parameter meet a second preset condition, a formatted search result group is generated based on the similarity search result, the parameter to be examined and the temporary search parameter.

6. The processing method according to claim 5, characterized in that: The calculating of the to-be-examined parameter and the temporary search parameter based on the similarity search result and the first preset condition includes: Based on the first preset condition, obtaining a temporary token value; Calculate a first parameter based on the temporary token value and a first preset formula; Calculate the parameters to be reviewed based on the first parameter and the second preset formula; A temporary search parameter is calculated based on the similarity search result, the first parameter, the parameter to be examined and a third preset formula.

7. The processing method according to claim 5, characterized in that: The step of verifying the formatted search result group according to the communication service request and the first preset condition and outputting a processing result includes: Based on the first preset condition, checking the temporary search parameter and the parameter to be examined; The inspection result meets the first preset condition, and the audit factor is calculated based on a fourth preset formula and the first preset condition; Calculate a search result summary value based on the hash function and the similarity search result; Calculate a first auxiliary audit parameter based on a fifth preset formula, the search result summary value and the temporary search parameter; Calculate a second auxiliary review parameter based on a sixth preset formula, the parameter to be reviewed and the temporary search parameter; Calculate a third auxiliary audit parameter based on a seventh preset formula, the first auxiliary audit parameter, the second auxiliary audit parameter, and the audit factor; Calculating an audit parameter based on an eighth preset formula and the third auxiliary audit parameter; If the parameters to be examined are the same as the audit parameters, a processing result is output based on the formatted search result group.

8. A communication service request processing system, characterized in that: include: The first module is used to obtain a communication service request and determine a query vector according to the communication service request; The second module is used to perform similarity search on the query vector based on a preset vector database to obtain a similarity search result; A third module is used to generate a formatted search result group based on the similarity search result, the first preset condition and the second preset condition; The fourth module is used to verify the formatted search result group according to the communication service request and the first preset condition, and output a processing result.

9. A communication service request processing device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to execute the processing method according to any one of claims 1 to 7 when executed by the processor.

Images