Short message monitoring and early warning method, device, equipment, medium and program product

Abstract

The invention provides a short message monitoring and early warning method, device and equipment, a medium and a program product. The method comprises the following steps: regularly obtaining newly added complaint short message samples; based on the newly added complaint short message sample, performing fine tuning on the large language model by adopting a parameter efficient fine tuning mode to obtain a fine-tuned large model; obtaining a short message content service request; and based on a short message content service request, performing short message early warning through the fine-tuned large model. According to the short message monitoring and early warning method provided by the invention, the newly added complaint short message samples are obtained regularly, and the large language model is finely adjusted by adopting a parameter efficient fine adjustment mode, so that the large language model can regularly upgrade the ability to deal with short message monitoring and early warning, and after the short message content service request is received, the short message content service request is sent to the server. Real-time short message early warning can be carried out based on the large model after fine tuning, potential service specification problems are efficiently identified, an intelligent real-time short message monitoring and early warning process is realized, the manual judgment time is shortened, and the short message monitoring and early warning efficiency is integrally improved.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a text message monitoring and early warning method, device, equipment, medium and program product. Background Art

[0002] In recent years, the problem of operators sending commercial SMS through service numbers in basic telecommunications enterprises has become increasingly prominent, and higher requirements have been put forward in the field of network information for the identification of business SMS content types and monitoring and early warning. In this context, SMS type identification, SMS content configuration monitoring and early warning, and active feedback of irregular SMS business content have gradually become the primary issues to be solved.

[0003] In the prior art, when performing SMS type identification and monitoring and early warning, the main techniques used are matching keywords and blacklist and whitelist filtering to identify SMS types and intercept SMS sending. The keyword matching method includes accurately screening SMS content or identifying SMS types by feature vector similarity comparison method; the blacklist and whitelist filtering technology mainly achieves accurate filtering by accurately matching the sending request user with the list.

[0004] However, the use of matching keywords and blacklist and whitelist filtering technology is blind to SMS type identification, and is prone to misjudgment and omission. It requires a high degree of manual participation, and the accuracy of matching is highly dependent on the manual maintenance of the word library. Therefore, the existing SMS monitoring and early warning technology is not efficient. Summary of the invention

[0005] The present invention provides a short message monitoring and early warning method, device, equipment, medium and program product, which are used to solve the problem of low efficiency of the short message monitoring and early warning method in the prior art.

[0006] In a first aspect, the present invention provides a method for monitoring and warning short messages, comprising: Obtain new complaint SMS samples regularly; Based on the newly added complaint SMS samples, the large language model is fine-tuned by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model; Get SMS content service request; Based on the SMS content service request, SMS warning is performed through the fine-tuned large model.

[0007] In one embodiment, based on the newly added complaint SMS sample, the large language model is fine-tuned by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model, including: Performing data preprocessing on the newly added complaint SMS sample to obtain a target complaint SMS sample; The target complaint SMS sample is labeled to obtain a label of the target complaint SMS sample; the label includes a complaint SMS label and an SMS category label; Based on the target complaint SMS sample and its label, adjusting the fine-tuning parameters of the large language model; Based on the fine-tuning parameters, the large language model is fine-tuned to obtain a fine-tuned large model.

[0008] In one embodiment, the SMS warning is performed based on the SMS content service request by using the fine-tuned large model, including: Perform call intent analysis on the SMS content service request to obtain SMS service specification monitoring and early warning intent; Monitoring the early warning intention according to the SMS service specification, parsing the first target SMS content from the SMS content service request; Inputting the first target SMS content into the fine-tuned large model to obtain the SMS service standardization result output by the fine-tuned large model; the SMS service standardization result includes a complaint result and an SMS category result; If the SMS service specification result does not comply with the SMS service specification, an SMS warning is issued based on the first target SMS content.

[0009] In one embodiment, the sending of a text message warning based on the first target text message content includes: Generating warning information and optimization suggestions for the first target SMS content; The warning information and the optimization suggestion are sent to the business platform that sent the SMS content service request.

[0010] In one embodiment, after obtaining the SMS content service request, the method further includes: Perform call intent analysis on the SMS content service request to obtain the SMS specification solution acquisition intent; According to the SMS specification solution acquisition intent, the second target SMS content and the SMS service specification issues of the second target SMS content are parsed from the SMS content service request; the SMS service specification issues include SMS specification warning rules and SMS specification common problems; Extracting a vector of the SMS service specification question to obtain a query vector; Based on the second target SMS content and the query vector, a SMS standard solution for the second target SMS content is obtained.

[0011] In one embodiment, the acquiring a SMS standard solution for the second target SMS content based on the second target SMS content and the query vector includes: According to the query vector, query is performed in a vector database to obtain knowledge data for solving the SMS service standard problem; the vector database is used to store answer knowledge for different SMS service standard problems; Encapsulating the knowledge data into prompt words for generating a standard solution for short messages; The prompt word and the second target SMS content are input into the fine-tuned large model to obtain a SMS standard solution for the second target SMS content output by the fine-tuned large model.

[0012] In one embodiment, the large language model is pre-trained in the following manner: Based on Flink CDC's non-intrusive data collection capabilities, obtain the first historical SMS samples that were complained about, the second historical SMS samples of different categories, the third historical SMS samples with SMS standard warning rules, and the fourth historical SMS samples with common SMS standard problems; The first historical SMS sample, the second historical SMS sample, the third historical SMS sample, and the fourth historical SMS sample are respectively labeled to obtain a complaint label of the first historical SMS sample, a category label of the second historical SMS sample, an early warning rule solution label of the third historical SMS sample, and a common problem solution label of the fourth historical SMS sample; The large language model is obtained by performing model training based on the first historical SMS sample and its complaint label, the second historical SMS sample and its category label, the third historical SMS sample and its warning rule solution label, and the fourth historical SMS sample and its common problem solution label.

[0013] In a second aspect, the present invention further provides a short message monitoring and early warning device, comprising: The first acquisition module is used to periodically acquire new complaint SMS samples; A model fine-tuning module is used to fine-tune the large language model based on the newly added complaint SMS sample by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model; The second acquisition module is used to obtain the SMS content service request; The SMS warning module is used to issue SMS warnings based on the SMS content service request through the fine-tuned large model.

[0014] In a third aspect, the present invention provides a device, comprising an electronic device, wherein the electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned SMS monitoring and early warning methods are implemented.

[0015] In a fourth aspect, the present invention further provides a medium, comprising a non-transitory computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of any of the above-mentioned SMS monitoring and early warning methods are implemented.

[0016] In a fifth aspect, the present invention also provides a product, comprising a computer program product, wherein the computer program product comprises a computer program, wherein the computer program can be stored on a non-transitory computer-readable storage medium, and wherein when the computer program is executed by the processor, the steps of any of the above-mentioned SMS monitoring and early warning methods are implemented.

[0017] The SMS monitoring and early warning method, device, equipment, medium and program product provided by the present invention can regularly upgrade the large language model's ability to respond to SMS monitoring and early warning by regularly acquiring new complaint SMS samples and fine-tuning the large language model in a parameter-efficient fine-tuning manner, and the fine-tuning efficiency is relatively high. Therefore, after receiving an SMS content service request, it can perform real-time SMS early warning based on the fine-tuned large model, efficiently identify potential business specification issues, realize an intelligent real-time SMS monitoring and early warning process, reduce the time for manual judgment, and improve the efficiency of SMS monitoring and early warning as a whole, while comprehensively improving the business SMS intelligent configuration early warning and SMS content source management capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of the SMS monitoring and early warning method provided by the present invention.

[0020] Figure 2 It is a technical framework diagram of the AI ​​big model technology provided by the present invention.

[0021] Figure 3 This is a schematic diagram of Flink CDC non-intrusive data collection provided by the present invention.

[0022] Figure 4 This is a model fine-tuning flow chart provided by the present invention.

[0023] Figure 5 This is a large model real-time reasoning flow chart provided by the present invention.

[0024] Figure 6 It is the overall service flow chart of the SMS monitoring and early warning provided by the present invention.

[0025] Figure 7 It is an overall service flow chart of the solution for obtaining SMS specifications provided by the present invention.

[0026] Figure 8 It is a flow chart of the vector database knowledge plug-in provided by the present invention.

[0027] Fig. 9 It is a structural schematic diagram of the short message monitoring and early warning device provided by the present invention.

[0028] Fig.10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein.

[0031] Combine the following Figure 1-Figure 10 The text message monitoring and early warning method, device, equipment, medium and program product provided by the present invention are described.

[0032] Combination Figure 1 , Figure 1 It is a flow chart of the SMS monitoring and early warning method provided by the present invention.

[0033] like Figure 1 As shown, the method includes the following: Step 101, regularly obtain new complaint SMS samples; Step 102: Based on the newly added complaint SMS sample, fine-tune the large language model using an efficient parameter fine-tuning method to obtain a fine-tuned large model; Step 103: Obtain SMS content service request; Step 104: Based on the SMS content service request, SMS warning is performed using the fine-tuned large model.

[0034] It should be noted that the SMS monitoring and early warning method provided in the embodiment of the present invention is implemented based on the SMS monitoring and early warning device. The embodiment of the present invention takes the SMS monitoring and early warning device as an example to describe the SMS monitoring and early warning method.

[0035] Generally speaking, after the business personnel configure the SMS template or specific SMS content on the business system or business platform, when the SMS request is normally triggered, the request message is directly sent to the SMS gateway, and then forwarded to the target user by the SMS gateway. The SMS monitoring and early warning method provided by the present invention mainly uses artificial intelligence (AI) big model and real-time data stream processing technology, adds the automatic recognition capability of the AI ​​big model in the link before the SMS content service request is sent to the SMS gateway, and establishes a real-time interaction mechanism between the business platform and the underlying capability platform, adds the SMS monitoring and early warning link before the request is sent to the SMS gateway, and improves the business standardization and SMS content source governance capabilities of the SMS business at the source.

[0036] The present invention is based on the underlying AI framework and model fine-tuning reasoning technology to enable the entire SMS sending scenario, and can be used for automated SMS classification, SMS monitoring and warning, SMS standard warning rules, and SMS standard FAQ responses. First, supervised fine-tuning (SFT) of large models is performed on historical complaints from existing users, known SMS types, and new complaint data to achieve SMS type judgment and monitoring and warning model construction; second, knowledge vectorization, semantic storage, and RAG (Retrieval Augmented Generation) calls are performed to determine SMS standard warning rules and SMS standard FAQs. In the judgment and warning process, the management end can quickly obtain response knowledge and solve common problems when it does not understand the SMS standard warning rules, thereby empowering SMS management stakeholders in customer service scenarios. The overall AI technology framework consists of 4 layers, such as Figure 2 As shown, Figure 2 It is a technical framework diagram of the AI ​​big model technology provided by the present invention. The AI ​​big model technology includes an algorithm base, a storage and computing base, training and push technology, and external services.

[0037] (1) Algorithm base: including technical components related to large models. The first is the AI ​​model, including large language models, intent recognition models, and vectorized models. The basic large language model can perform intelligent interaction of general knowledge, but there are errors in reasoning decisions for scenarios such as SMS classification, detection and early warning, and the model needs to be fine-tuned to improve recognition accuracy; the intent recognition model is used to identify the calling intention of the external caller. For example, when the external system calls the interface in real time, it needs to be routed to different processing logic according to different intentions; the vectorized model assists the search engine, which not only realizes the vectorized calculation and query of knowledge corpus, but also supports the hybrid retrieval of semantic query and keyword query, expanding the query coverage and accuracy. Second, the framework includes PyTorch, Cuda, and vLLM. PyTorch is the programming framework for model training fine-tuning and real-time reasoning adopted in this solution. It performs calculations based on dynamic computational graphs to customize AI model architecture, optimizers, loss functions, and other contents. Cuda is the underlying parallel computing tool and GPU programming model, which enables efficient computing of the AI ​​model of this solution on the GPU. vLLM is a large model computing acceleration framework that effectively manages the Attention memory of large models based on the PagedAttention algorithm, thereby improving the reasoning throughput and memory utilization efficiency of large models.

[0038] (2) Storage and computing base: including offline and real-time storage and computing services. First, in terms of offline data, offline SMS rules and log files are collected through FTP, SFTP, and JDBC, business specification issues are aggregated, and delivered to the AI ​​model for regular model optimization and knowledge storage; second, in terms of real-time data, Kafka+Flink's data collection and computing architecture is used to achieve real-time aggregation and computing of complaint data, SMS content, SMS type, warning signs and other data in the center + edge dual-center deployment mode, and data preprocessing and feature extraction are achieved through data filtering, cleaning, completion, and association, and user large models are regularly fine-tuned and real-time inference to shorten data latency; in addition, data persistence storage is achieved through Hbase.

[0039] (3) Training and push technology: including corpus management, model fine-tuning, and model reasoning. The corpus management module is not only for training and fine-tuning data, but also for data that needs to be queried in real time during the reasoning and prediction process, such as SMS types, business specification warning rules, and business specification common problems. It is stored in the vector database through Embedding vectorization technology, and the HNSW index is used to improve the retrieval speed of the vector database. The fine-tuning training module is based on components such as Jupyter and WebSSH, encapsulates algorithms such as QLoRA and P-Tuning v2, and conducts data training on corpora such as complaints and SMS classification to generate SMS judgment and prediction models. The model reasoning module is based on prompt word engineering + RAG technology, and uses prompt word technologies such as Few Shot and Chain of Thought to enhance the semantic understanding of the big model for business specification warning rules and business specification common problems. The RAG technology of the vector database plug-in big model is used to enhance the big model's understanding of SMS content, so that the big model can generate reasoning responses related to SMS rules and common problems with generative technology.

[0040] (4) External services: Use technologies and frameworks such as Java, Python, and JavaScript to implement the model’s external services through an application programming interface (API) or a web terminal.

[0041] Therefore, under the technical framework of AI big model technology, after the big language model has completed pre-training, it will periodically obtain new complaint SMS samples to fine-tune the big language model, so that the model can be quickly adjusted according to the continuously generated sample data to improve prediction accuracy.

[0042] Specifically, based on Flink CDC's non-intrusive data collection technology, new complaint SMS samples are obtained regularly.

[0043] It should be noted that based on Flink CDC non-intrusive data collection technology, it adopts a highly available streaming computing framework, and uses lock-free, non-intrusive concurrent reading, breakpoint resumption and other technical features to monitor and capture complaints from the business platform and Binlog change logs of the SMS production library in real time, including data or data table insertion INSERT, update UPDATE, delete DELETE, etc., and writes them to the message middleware in chronological order, completing the visualization and millisecond-level real-time transmission of multi-source data such as complaints from the business platform and SMS, ensuring the continuous and real-time update of the data source for AI large model training. Figure 3 , Figure 3 This is a schematic diagram of Flink CDC non-intrusive data collection provided by the present invention.

[0044] Furthermore, based on the newly added complaint SMS samples, an efficient parameter fine-tuning method was used to fine-tune the large language model to obtain a fine-tuned large model. This allows the model performance to be improved through a small amount of parameter adjustments without changing all the parameters of the original model, which can significantly reduce computing resources and memory overhead.

[0045] Therefore, after the business personnel configure the specific SMS content on the business platform, they generate an SMS content service request and send it to the SMS monitoring and early warning system. The SMS monitoring and early warning system can determine whether the specific SMS content in the SMS content service request complies with the SMS business specifications based on the SMS content service request and the latest fine-tuned large model.

[0046] If it complies with the SMS business specifications, the SMS content service request is allowed to be sent to the gateway, and the gateway parses the SMS content and sends it to the target user; if it does not comply with the SMS business specifications, an early warning message is generated and sent to the business platform to inform business personnel that there is a business specification risk in the SMS content.

[0047] The SMS monitoring and early warning method provided by the present invention can periodically obtain new complaint SMS samples and fine-tune the large language model in a parameter-efficient fine-tuning manner, so that the large language model can be regularly upgraded to cope with SMS monitoring and early warning capabilities, and the fine-tuning efficiency is relatively high. Therefore, after receiving an SMS content service request, it can perform real-time SMS early warning based on the fine-tuned large model, efficiently identify potential business specification issues, realize an intelligent real-time SMS monitoring and early warning process, reduce the time for manual judgment, and improve the efficiency of SMS monitoring and early warning as a whole, while comprehensively improving the business SMS intelligent configuration early warning and SMS content source management capabilities.

[0048] In one embodiment, based on step 102, the large language model is fine-tuned by adopting an efficient parameter fine-tuning method based on the newly added complaint SMS sample to obtain a fine-tuned large model, including: Performing data preprocessing on the newly added complaint SMS sample to obtain a target complaint SMS sample; The target complaint SMS sample is labeled to obtain a label of the target complaint SMS sample; the label includes a complaint SMS label and an SMS category label; Based on the target complaint SMS sample and its label, adjusting the fine-tuning parameters of the large language model; Based on the fine-tuning parameters, the large language model is fine-tuned to obtain a fine-tuned large model.

[0049] Combination Figure 4 , Figure 4 This is the model fine-tuning flow chart provided by the present invention, and the following content describes its process.

[0050] Specifically, in order to better meet the input needs of the large language model, data preprocessing is performed on the newly added complaint samples to obtain the target complaint SMS samples after data preprocessing.

[0051] In the process of data preprocessing, it is necessary to clean, filter and classify the newly added complaint SMS samples, analyze the missing values ​​and outliers, and complete them in real time based on the mean of historical data to ensure the integrity of the data. The common sample processing method can achieve the established goal by deleting the missing value data. Since this method has great limitations, the Lagerrange interpolation method is used to complete the real-time data completion.

[0052] That is, given n+1 discrete point data pairs (xi, yi), where i=0,1,...,n, the goal of the Lagrange interpolation method is to construct an n-order polynomial L(x) such that L(xi)=yi. This polynomial can be expressed as:

[0053] Among them, each is a Lagrangian basis function, defined as follows:

[0054] Based on the given data points, calculate each Lagrangian basis function , and then substitute The final interpolation polynomial is obtained from the expression of .

[0055] At the same time, based on the cloud Flink streaming computing capabilities, the real-time data processing and processing method combining Flink Java API and SQL is adopted to realize real-time application development through visual operation of dragging controls. Adopting the dual-center deployment of center + edge, under the premise of avoiding data errors, shortening data delays, and ensuring data security, the data is filtered, cleaned, completed, associated and other feature extractions are carried out in real time to form the target complaint SMS samples as the input data source of the AI ​​large model. The newly added complaint SMS samples involved include SMS template identifiers, SMS business content, and SMS ports for sending SMS as inputs for the training model.

[0056] Furthermore, the target complaint SMS sample is labeled to obtain a label of the target complaint SMS sample, wherein the label specifically includes a complaint SMS label and an SMS category label. The complaint SMS label is used to reflect that the target complaint SMS sample has been complained by the user, and the SMS category label is used to reflect the category to which the target complaint SMS sample belongs.

[0057] Furthermore, based on the target complaint SMS samples and their labels, the fine-tuning parameters of the large language model are adjusted, which may include batch size, gradient accumulation steps, learning rate, and variation strategy.

[0058] Furthermore, the QLoRA method can be started to fine-tune the large language model based on the fine-tuning parameters. After the fine-tuning is completed, the adaptation weights generated by QLoRA are merged with the weights of the basic model to finally obtain the fine-tuned large model. The fine-tuned large model can be saved using an appropriate framework to generate the corresponding model file for subsequent use and deployment.

[0059] For model fine-tuning scenarios, efficient fine-tuning (partial parameter fine-tuning) is used. First, the model loads the pre-trained parameters. Initialize and then adjust the parameters by maximizing the conditional language model probability :

[0060] Among them, x represents the input sequence (feature) of the large model; y represents the output sequence of the large model; Z represents the input and output space; t represents the token sequence; is a multi-task learner based on full fine-tuning, indicating the full fine-tuning parameters Below, the conditional probability of the input predicted as the output.

[0061] Furthermore, fewer parameters are used to represent the parameter increments to be learned ,in, , originally looking for The optimization goal becomes to find :

[0062] Among them, x represents the input sequence (feature) of the large model; y represents the output sequence of the large model; Z represents the input and output space; t represents the token sequence; It is a multi-task learner based on efficient fine-tuning of small batch parameters, representing the initialization parameters Add small batches to fine-tune parameters Below, the conditional probability of the input predicted as the output.

[0063] The embodiment of the present invention obtains new complaint SMS samples at regular intervals and fine-tunes the large language model by means of efficient parameter fine-tuning, thereby significantly improving the model's understanding and response accuracy to complaint SMS. The large language model can be regularly upgraded to cope with SMS monitoring and early warning capabilities, and the fine-tuning efficiency is relatively high. Therefore, after receiving an SMS content service request, real-time SMS early warning can be performed based on the fine-tuned large model, potential business specification issues can be efficiently identified, and an intelligent real-time SMS monitoring and early warning process can be realized.

[0064] In some embodiments, based on step 104, the SMS warning is performed based on the SMS content service request by using the fine-tuned large model, including: Perform call intent analysis on the SMS content service request to obtain SMS service specification monitoring and early warning intent; Monitoring the early warning intention according to the SMS service specification, parsing the first target SMS content from the SMS content service request; Inputting the first target SMS content into the fine-tuned large model to obtain the SMS service standardization result output by the fine-tuned large model; the SMS service standardization result includes a complaint result and an SMS category result; If the SMS service specification result does not comply with the SMS service specification, an SMS warning is issued based on the first target SMS content.

[0065] It should be noted that, combined with Figure 5 , Figure 5 is a large model real-time reasoning flow chart provided by the present invention, consisting of Figure 5 It can be seen that the fine-tuned large language model can be used for real-time reasoning. After obtaining the SMS content service request in real time, the intent of the SMS content service request is first determined. When it is confirmed that the intent of the request is to monitor SMS service specifications and implement SMS monitoring and early warning, the specific SMS content is obtained, and the real-time reasoning is completed by calling the large model to generate the result.

[0066] Combination Figure 6 , Figure 6 This is the overall service flow chart of the SMS monitoring and early warning provided by the present invention, and the process is described below.

[0067] Specifically, the SMS content service request is subjected to call intent analysis, which can be performed by inputting an intent recognition model and performing AI intelligent analysis to obtain the intent in the SMS content service request.

[0068] If the intent in the SMS content service request is the SMS business specification monitoring and early warning intent, the first target SMS content is parsed from the SMS content service request according to the SMS business specification monitoring and early warning intent.

[0069] Furthermore, the first target SMS content is input into the fine-tuned large model to obtain the SMS service specification result output by the fine-tuned large model, and the SMS service specification result includes the complaint result and the SMS category result.

[0070] If the SMS service specification result does not comply with the SMS service specification, it means that the first target SMS content under this SMS category has SMS service irregularities, which is very likely to be complained by users, then an SMS warning will be issued based on the first target SMS content.

[0071] The embodiment of the present invention selects the processing logic process of SMS business specification monitoring and early warning through the intention analysis result, further parses the first target SMS content in the SMS content service request, and performs SMS early warning through the fine-tuned large language model, thereby effectively realizing automatic monitoring and early warning of potential SMS problems, improving the intelligence level of SMS processing, and at the same time improving response efficiency.

[0072] According to the above description, the SMS warning based on the first target SMS content includes: Generating warning information and optimization suggestions for the first target SMS content; The warning information and the optimization suggestion are sent to the business platform that sent the SMS content service request.

[0073] Specifically, after confirming that the SMS service specification result does not meet the SMS service specification, a warning message for the first target SMS content is generated. At the same time, optimization suggestions for the first target SMS content are also generated. The optimization suggestions for the SMS content can be generated by AI intelligence by inputting them into the fine-tuned large model.

[0074] Furthermore, a warning message and optimization suggestions for the first target SMS content are sent to the business platform that sent the SMS content service request. Therefore, after the business personnel receive the warning message on the business platform, they first cancel the request to send the SMS, and then change the SMS content again according to the obtained optimization suggestions.

[0075] When the embodiment of the present invention identifies that the content of a text message does not comply with the text message business specifications, it generates warning information and corresponding optimization suggestions for the text message content, and sends the warning information and optimization suggestions to the business platform, ensuring that business personnel can obtain feedback in a timely manner and take necessary adjustments. This not only improves the monitoring efficiency of potential problem text messages, but also provides data support for the optimization of business processes, thereby promoting smooth overall operations and improving customer satisfaction.

[0076] In some embodiments, after obtaining the SMS content service request, the method further includes: Perform call intent analysis on the SMS content service request to obtain the SMS specification solution acquisition intent; According to the SMS specification solution acquisition intent, the second target SMS content and the SMS service specification issues of the second target SMS content are parsed from the SMS content service request; the SMS service specification issues include SMS specification warning rules and SMS specification common problems; Extracting a vector of the SMS service specification question to obtain a query vector; Based on the second target SMS content and the query vector, a SMS standard solution for the second target SMS content is obtained.

[0077] It should be noted that Figure 5 It can be seen that after obtaining the SMS content service request in real time, the intention of the SMS content service request is first determined. When it is confirmed that the intention of the request is to obtain SMS business specification warnings or solutions to common problems, the specific SMS content and the SMS business specification problems therein are obtained, and the response knowledge of the vector database is matched based on semantic retrieval and keyword retrieval, and the knowledge is encapsulated as prompt words. The real-time reasoning is completed by calling the large model to generate results.

[0078] Combination Figure 7 , Figure 7 This is the overall service flow chart of the solution for obtaining SMS specifications provided by the present invention, and the process is described below.

[0079] Specifically, the SMS content service request is subjected to call intent analysis, which can be performed by inputting an intent recognition model and performing AI intelligent analysis to obtain the intent in the SMS content service request.

[0080] If the intent in the SMS content service request is the SMS specification solution acquisition intent, then the second target SMS content and the SMS business specification issues of the second target SMS content are parsed from the SMS content service request according to the SMS business specification monitoring and warning intent. The SMS business specification issues include SMS specification warning rules and SMS specification frequently asked questions.

[0081] Vectors are extracted for SMS service specification questions to obtain corresponding query vectors.

[0082] Further, based on the second target SMS content and the query vector, a SMS standard solution for the second target SMS content is obtained.

[0083] The embodiment of the present invention selects the SMS standard solution to obtain this processing logic process through the intention analysis result, further parses the second target SMS content and SMS business standard issues in the SMS content service request, extracts vectors from the data, and realizes the processability of data query in the vector database, so as to query the knowledge data that can solve the SMS business standard issues, and generates the corresponding SMS standard solution in combination with the second target SMS content, thereby providing business personnel with timely and effective solutions and improving the standardization of SMS business.

[0084] According to the above description, the method of acquiring a SMS standard solution for the second target SMS content based on the second target SMS content and the query vector includes: According to the query vector, query is performed in a vector database to obtain knowledge data for solving the SMS service standard problem; the vector database is used to store answer knowledge for different SMS service standard problems; Encapsulating the knowledge data into prompt words for generating a standard solution for short messages; The prompt word and the second target SMS content are input into the fine-tuned large model to obtain a SMS standard solution for the second target SMS content output by the fine-tuned large model.

[0085] Specifically, according to the query vector, a query is performed in the vector database to obtain knowledge data for solving the SMS service specification problem.

[0086] It should be noted that the vector database stores the answer knowledge of different SMS service standard questions, and connects to the big model in the form of knowledge plug-in. When the model is inferring, it triggers the RAG call, that is, to obtain the database knowledge, and then provide it to the big model as supplementary knowledge to assist in generating SMS standard solutions. Figure 8 , Figure 8 It is a flow chart of the vector database knowledge plug-in provided by the present invention. Specifically, it is as follows: obtain SMS standard warning rule samples and SMS standard common problem samples, split these samples to form Block data segments, generate corresponding responses for each Block data segment based on the summarization and summary capabilities of the large model, thereby forming multiple QA question-answer pairs, further, through the vectorization model, realize the data vectorization of these QA question-answer pairs, and at the same time build the HNSW index to realize the data entry into the vector database.

[0087] Furthermore, the acquired knowledge data is encapsulated into prompt words for generating a standard solution for SMS messages.

[0088] Furthermore, the prompt words and the second target SMS content are input into the fine-tuned large model, and the prompt words guide the fine-tuned large model to output a standard SMS solution for the second target SMS content, and the standard SMS solution is sent to the business platform so that business personnel can modify the SMS content according to the obtained standard SMS solution.

[0089] The embodiment of the present invention uses the generated query vector to perform precise query in the vector database, obtains relevant knowledge data on SMS business standardization issues, encapsulates the obtained knowledge data into prompt words, and performs intelligent analysis based on the prompt words and the second target SMS content through a fine-tuned large model, outputs corresponding SMS standard solutions, provides business personnel with timely and effective solutions, and improves the standardization of SMS services.

[0090] In some embodiments, the large language model is pre-trained in the following manner: Based on Flink CDC's non-intrusive data collection capabilities, obtain the first historical SMS samples that were complained about, the second historical SMS samples of different categories, the third historical SMS samples with SMS standard warning rules, and the fourth historical SMS samples with common SMS standard problems; The first historical SMS sample, the second historical SMS sample, the third historical SMS sample, and the fourth historical SMS sample are respectively labeled to obtain a complaint label of the first historical SMS sample, a category label of the second historical SMS sample, an early warning rule solution label of the third historical SMS sample, and a common problem solution label of the fourth historical SMS sample; The large language model is obtained by performing model training based on the first historical SMS sample and its complaint label, the second historical SMS sample and its category label, the third historical SMS sample and its warning rule solution label, and the fourth historical SMS sample and its common problem solution label.

[0091] Specifically, based on the non-intrusive data collection capability of Flink CDC, we obtain the first historical SMS samples that were complained about, the second historical SMS samples of different categories, the third historical SMS samples with SMS standard warning rules, and the fourth historical SMS samples with SMS standard common problems, and perform data preprocessing on the obtained sample data. The data preprocessing method is the same as above and will not be repeated here.

[0092] For example, being complained about means that the content of historical emails is not standardized and has been complained by the users to whom they were sent; different categories may include marketing, service, business reminder, account arrival reminder, public welfare notification, internal agency reminder, etc.; SMS standard warning rules may include service SMS mixed with marketing content, marketing content with links, marketing content without real-name signature, marketing content without clear SMS unsubscription method, SMS content is too long, etc.; Common SMS standard issues include "Please confirm the SMS type of this SMS template", "Please be careful not to include marketing content in service SMS", "Please confirm whether the link is mounted", "Do you want to add a real-name signature", "Is the SMS unsubscription method added normally", "Please pay attention to the word limit of the SMS content", etc.

[0093] The first historical SMS sample and the second historical SMS sample are manually labeled respectively to obtain a complaint label of the first historical SMS sample and a category label of the second historical SMS sample.

[0094] At the same time, the third historical SMS sample and the fourth historical SMS sample are respectively annotated with large model prompt words to obtain the warning rule solution label of the third historical SMS sample and the common problem solution label of the fourth historical SMS sample.

[0095] Furthermore, model training is performed based on the first historical SMS sample and its complaint label, the second historical SMS sample and its category label, the third historical SMS sample and its warning rule solution label, and the fourth historical SMS sample and its common problem solution label to obtain a pre-trained large language model.

[0096] The embodiment of the present invention uses Flink CDC technology to obtain a variety of SMS samples, including the first historical SMS samples that were complained about, the second historical SMS samples of different categories, the third historical SMS samples containing SMS standard warning rules, and the fourth historical SMS samples involving common problems. After accurately labeling these samples, complaint labels, category labels, warning rule solution labels, and common problem solution labels are generated respectively, providing rich and structured input data for model training, ensuring the accuracy and comprehensiveness of the model in understanding and processing SMS specification-related issues. Therefore, the pre-trained large language model has the ability to effectively identify and standardize SMS content, significantly improving the intelligence level and response quality of SMS services.

[0097] Based on the above, in the regular SMS delivery process, we have added SMS monitoring and early warning and SMS business specification question answering links to comprehensively improve the business SMS intelligent configuration early warning and SMS content source management capabilities. The advantages are as follows: 1. Intelligent service configuration: Based on the intent recognition model, the user's calling intention is judged, and reasoning and judgment are performed based on the fine-tuned large model to generate the final SMS service specification recognition results and reasoning responses to SMS service specification questions.

[0098] 2. Convenient configuration of business personnel: Through intelligent recognition and classification feedback of AI large models, SMS types and complaints can be intelligently identified based on the continuously updated large language model, and early warnings and optimization suggestions can be made based on the results to improve the standardization of SMS business.

[0099] 3. Improve service efficiency: Through real-time data stream processing technology, real-time data collection and service efficiency can be improved in the entire process from data collection to data service.

[0100] 4. Reduce labor costs: Through real-time intelligent SMS monitoring and early warning, the time for manual judgment and querying the knowledge base is reduced, which improves business configuration efficiency and reduces labor maintenance costs compared to existing technologies such as keywords and blacklists.

[0101] The text message monitoring and early warning device provided by the present invention is described below. The text message monitoring and early warning device described below and the text message monitoring and early warning method described above can be referenced to each other.

[0102] Reference Fig. 9 , Fig. 9It is a structural schematic diagram of the short message monitoring and early warning device provided by the present invention.

[0103] The short message monitoring and early warning device comprises: The first acquisition module 910 is used to periodically acquire new complaint SMS samples.

[0104] The model fine-tuning module 920 is used to fine-tune the large language model based on the newly added complaint SMS sample by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model.

[0105] The second acquisition module 930 is used to acquire a text message content service request.

[0106] The SMS warning module 940 is used to issue SMS warnings based on the SMS content service request using the fine-tuned large model.

[0107] The SMS monitoring and early warning device provided by the present invention can regularly upgrade the ability of the large language model to respond to SMS monitoring and early warning by periodically acquiring new complaint SMS samples and fine-tuning the large language model by means of efficient parameter fine-tuning. The fine-tuning efficiency is relatively high. Therefore, after receiving an SMS content service request, it can perform real-time SMS early warning based on the fine-tuned large model, efficiently identify potential business specification problems, realize intelligent real-time SMS monitoring and early warning process, reduce the time for manual judgment, improve the efficiency of SMS monitoring and early warning as a whole, and comprehensively improve the business SMS intelligent configuration early warning and SMS content source management capabilities.

[0108] Furthermore, the model fine-tuning module 920 is also used for: Performing data preprocessing on the newly added complaint SMS sample to obtain a target complaint SMS sample; The target complaint SMS sample is labeled to obtain a label of the target complaint SMS sample; the label includes a complaint SMS label and an SMS category label; Based on the target complaint SMS sample and its label, adjusting the fine-tuning parameters of the large language model; Based on the fine-tuning parameters, the large language model is fine-tuned to obtain a fine-tuned large model.

[0109] Furthermore, the SMS warning module 940 is also used for: Perform call intent analysis on the SMS content service request to obtain SMS service specification monitoring and early warning intent; Monitoring the early warning intention according to the SMS service specification, parsing the first target SMS content from the SMS content service request; Inputting the first target SMS content into the fine-tuned large model to obtain the SMS service standardization result output by the fine-tuned large model; the SMS service standardization result includes a complaint result and an SMS category result; If the SMS service specification result does not comply with the SMS service specification, an SMS warning is issued based on the first target SMS content.

[0110] Furthermore, the SMS warning module 940 is also used for: Generating warning information and optimization suggestions for the first target SMS content; The warning information and the optimization suggestion are sent to the business platform that sent the SMS content service request.

[0111] Furthermore, the SMS monitoring and early warning device is also used for: Perform call intent analysis on the SMS content service request to obtain the SMS specification solution acquisition intent; According to the SMS specification solution acquisition intent, the second target SMS content and the SMS service specification issues of the second target SMS content are parsed from the SMS content service request; the SMS service specification issues include SMS specification warning rules and SMS specification common problems; Extracting a vector of the SMS service specification question to obtain a query vector; Based on the second target SMS content and the query vector, a SMS standard solution for the second target SMS content is obtained.

[0112] Furthermore, the SMS monitoring and early warning device is also used for: According to the query vector, query is performed in a vector database to obtain knowledge data for solving the SMS service standard problem; the vector database is used to store answer knowledge for different SMS service standard problems; Encapsulating the knowledge data into prompt words for generating a standard solution for short messages; The prompt word and the second target SMS content are input into the fine-tuned large model to obtain a SMS standard solution for the second target SMS content output by the fine-tuned large model.

[0113] Furthermore, the SMS monitoring and early warning device is also used for: Based on Flink CDC's non-intrusive data collection capabilities, obtain the first historical SMS samples that were complained about, the second historical SMS samples of different categories, the third historical SMS samples with SMS standard warning rules, and the fourth historical SMS samples with common SMS standard problems; The first historical SMS sample, the second historical SMS sample, the third historical SMS sample, and the fourth historical SMS sample are respectively labeled to obtain a complaint label of the first historical SMS sample, a category label of the second historical SMS sample, an early warning rule solution label of the third historical SMS sample, and a common problem solution label of the fourth historical SMS sample; The large language model is obtained by performing model training based on the first historical SMS sample and its complaint label, the second historical SMS sample and its category label, the third historical SMS sample and its warning rule solution label, and the fourth historical SMS sample and its common problem solution label.

[0114] It should be noted that the SMS monitoring and early warning device provided by the present invention can execute the SMS monitoring and early warning method described in any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0115] Fig.10 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig.10 As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020 and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute the SMS monitoring and early warning method, which includes: regularly obtaining new complaint SMS samples; based on the new complaint SMS samples, fine-tuning the large language model using an efficient parameter fine-tuning method to obtain a fine-tuned large model; obtaining a SMS content service request; and based on the SMS content service request, performing SMS early warning using the fine-tuned large model.

[0116] In addition, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0117] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the SMS monitoring and early warning methods provided in the above-mentioned embodiments, and the method includes: regularly obtaining new complaint SMS samples; based on the new complaint SMS samples, fine-tuning the large language model using an efficient parameter fine-tuning method to obtain a fine-tuned large model; obtaining an SMS content service request; based on the SMS content service request, issuing an SMS early warning using the fine-tuned large model.

[0118] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the SMS monitoring and early warning method provided in the above-mentioned embodiments, the method comprising: regularly obtaining new complaint SMS samples; based on the new complaint SMS samples, fine-tuning the large language model using an efficient parameter fine-tuning method to obtain a fine-tuned large model; obtaining an SMS content service request; and based on the SMS content service request, issuing an SMS early warning using the fine-tuned large model.

[0119] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Those of ordinary skill in the art may understand and implement the present invention without creative effort.

[0120] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A text message monitoring and early warning method, characterized in that: include: Obtain new complaint SMS samples regularly; Based on the newly added complaint SMS samples, the large language model is fine-tuned by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model; Get SMS content service request; Based on the SMS content service request, SMS warning is performed through the fine-tuned large model.

2. The SMS monitoring and early warning method according to claim 1 is characterized in that: Based on the newly added complaint SMS sample, the large language model is fine-tuned by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model, including: Performing data preprocessing on the newly added complaint SMS sample to obtain a target complaint SMS sample; The target complaint SMS sample is labeled to obtain a label of the target complaint SMS sample; the label includes a complaint SMS label and an SMS category label; Based on the target complaint SMS sample and its label, adjusting the fine-tuning parameters of the large language model; Based on the fine-tuning parameters, the large language model is fine-tuned to obtain a fine-tuned large model.

3. The SMS monitoring and early warning method according to claim 1 is characterized in that: The SMS warning is performed based on the SMS content service request by using the fine-tuned large model, including: Perform call intent analysis on the SMS content service request to obtain SMS service specification monitoring and early warning intent; Monitoring the early warning intention according to the SMS service specification, parsing the first target SMS content from the SMS content service request; Inputting the first target SMS content into the fine-tuned large model to obtain the SMS service standardization result output by the fine-tuned large model; the SMS service standardization result includes a complaint result and an SMS category result; If the SMS service specification result does not comply with the SMS service specification, an SMS warning is issued based on the first target SMS content.

4. The SMS monitoring and early warning method according to claim 3 is characterized in that: The issuing of a text message warning based on the first target text message content includes: Generating warning information and optimization suggestions for the first target SMS content; The warning information and the optimization suggestion are sent to the business platform that sent the SMS content service request.

5. The SMS monitoring and early warning method according to claim 1 is characterized in that: After obtaining the SMS content service request, it also includes: Perform call intent analysis on the SMS content service request to obtain the SMS specification solution acquisition intent; According to the SMS specification solution acquisition intent, the second target SMS content and the SMS service specification issues of the second target SMS content are parsed from the SMS content service request; the SMS service specification issues include SMS specification warning rules and SMS specification common problems; Extracting a vector of the SMS service specification question to obtain a query vector; Based on the second target SMS content and the query vector, a SMS standard solution for the second target SMS content is obtained.

6. The SMS monitoring and early warning method according to claim 5 is characterized in that: The acquiring a SMS standard solution for the second target SMS content based on the second target SMS content and the query vector includes: According to the query vector, query is performed in a vector database to obtain knowledge data for solving the SMS service standard problem; the vector database is used to store answer knowledge for different SMS service standard problems; Encapsulating the knowledge data into prompt words for generating a standard solution for short messages; The prompt word and the second target SMS content are input into the fine-tuned large model to obtain a SMS standard solution for the second target SMS content output by the fine-tuned large model.

7. The SMS monitoring and early warning method according to any one of claims 1 to 6, characterized in that: The large language model is pre-trained in the following way: Based on Flink CDC's non-intrusive data collection capabilities, obtain the first historical SMS samples that were complained about, the second historical SMS samples of different categories, the third historical SMS samples with SMS standard warning rules, and the fourth historical SMS samples with common SMS standard problems; The first historical SMS sample, the second historical SMS sample, the third historical SMS sample, and the fourth historical SMS sample are respectively labeled to obtain a complaint label of the first historical SMS sample, a category label of the second historical SMS sample, an early warning rule solution label of the third historical SMS sample, and a common problem solution label of the fourth historical SMS sample; The large language model is obtained by performing model training based on the first historical SMS sample and its complaint label, the second historical SMS sample and its category label, the third historical SMS sample and its warning rule solution label, and the fourth historical SMS sample and its common problem solution label.

8. A short message monitoring and early warning device, characterized in that: include: The first acquisition module is used to periodically acquire new complaint SMS samples; A model fine-tuning module is used to fine-tune the large language model based on the newly added complaint SMS sample by adopting an efficient parameter fine-tuning method to obtain a fine-tuned large model; The second acquisition module is used to obtain the SMS content service request; The SMS warning module is used to issue SMS warnings based on the SMS content service request through the fine-tuned large model.

9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the SMS monitoring and early warning method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, the computer-readable storage medium comprising a non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the SMS monitoring and early warning method as described in any one of claims 1 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the steps of the SMS monitoring and early warning method as described in any one of claims 1 to 7 are implemented.

Images