Large model-based complaint information early warning method, apparatus and device, and storage medium

Abstract

The invention relates to the technical field of data analysis, and discloses a complaint information early warning method and device based on a large model, equipment and a storage medium, and the method comprises the steps: determining the voice semantic features of a user through ASR; constructing a target Prompt (Prompt Prompt) template; and generating a complaint information judgment result through the complaint risk model according to the target Prompt template and the user voice semantic features, and performing complaint information early warning according to the complaint information judgment result. Through the above mode, the voice semantic features in the call voice information are accurately extracted through the ASR, the Prompt template is constructed in combination with the user portrait, the service background and the historical behavior information, and the real intention of the user is comprehensively understood. Automatic complaint early warning is realized through a large model according to a Prompt template and voice semantic features, and the efficiency of recognizing complaint information and carrying out complaint information early warning by an intelligent customer service system is improved in the business fields of financial science and technology, medical health, old-age care and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and in particular to a complaint information early warning method and device based on a large model, equipment and a storage medium. BACKGROUND

[0002] With the acceleration of digital transformation, the interaction between customers and service providers generates a large amount of multi-modal complaint data, including voice, text, images, etc. Service providers face problems such as low efficiency in handling customer complaints and lagging risk early warning. The existing technology has obvious limitations in handling complaint risk identification. On the one hand, traditional methods rely on manual experience to judge complaint risks, which not only is inefficient but also is easily affected by subjective factors, resulting in a high misjudgment rate. On the other hand, some existing automated tools can only handle single data types, such as analyzing only text or voice, and cannot effectively integrate multi-modal data, thus failing to fully capture the real intentions and emotions of customers.

[0003] In the field of financial technology, although some institutions have begun to try to use natural language processing technology to analyze customer feedback, these applications mostly focus on text data analysis, ignoring the rich emotional and intonation information contained in voice calls. Similarly, in the field of medical health, although progress has been made in electronic medical records and medical image analysis, the use of voice conversations for complaint risk identification is still in its infancy. The existing technology lacks a comprehensive solution that can handle multiple data types such as text, voice, and images simultaneously, and cannot achieve comprehensive, real-time, and accurate identification of customer complaint risks. Therefore, in the fields of financial technology, medical health and aging care, how to improve the efficiency of intelligent customer service systems in identifying and early warning complaint information has become a technical problem that needs to be solved. SUMMARY

[0004] The present application provides a complaint information early warning method, device, equipment and storage medium based on a large model to improve the efficiency of intelligent customer service systems in identifying and early warning complaint information.

[0005] In a first aspect, the present application provides a complaint information early warning method based on a large model, the method comprising:

[0006] determining user voice semantic features in the call voice information through an automatic speech recognition engine ASR;

[0007] constructing a target prompt Prompt template according to user portrait information, business background information and historical behavior information of the target user;

[0008] The complaint information determination result of the call voice information is generated according to the target Prompt template and the user voice semantic feature through a preset complaint risk model, and complaint information early warning is performed according to the complaint information determination result.

[0009] In a second aspect, the application further provides a complaint information early warning device based on a large model, the device comprising:

[0010] A user voice semantic feature determination module is configured to determine user voice semantic features in call voice information through an automatic speech recognition engine ASR;

[0011] A target Prompt template construction module is configured to construct a target prompt symbol Prompt template according to user portrait information, business background information and historical behavior information of a target user;

[0012] A complaint information early warning module is configured to generate a complaint information determination result of the call voice information according to the target Prompt template and the user voice semantic feature through a preset complaint risk model, and perform complaint information early warning according to the complaint information determination result.

[0013] In a third aspect, the application further provides a computer device, comprising a memory and a processor; the memory is configured to store a computer program; the processor is configured to execute the computer program and realize the complaint information early warning method based on a large model as described above when executing the computer program.

[0014] In a fourth aspect, the application further provides a computer readable storage medium, which stores a computer program, and the computer program makes the processor realize the complaint information early warning method based on a large model as described above when executed by the processor.

[0015] The application discloses a complaint information early warning method and device based on a large model, equipment and a storage medium. The complaint information early warning method based on the large model comprises the following steps: determining user voice semantic features in call voice information through an automatic speech recognition engine ASR; constructing a target prompt Prompt template according to user portrait information, business background information and historical behavior information of a target user; generating a complaint information determination result of the call voice information according to the target Prompt template and the user voice semantic features through a preset complaint risk model, and performing complaint information early warning according to the complaint information determination result. In the foregoing manner, the application accurately extracts voice semantic features in call voice information through ASR, constructs a Prompt template in combination with user portrait, business background and historical behavior information, and comprehensively understands the real intention of the user. Through the large model, automatic complaint early warning is realized according to the Prompt template and the voice semantic features, and in the business fields of financial technology, medical health and old-age care, the efficiency of the intelligent customer service system in identifying complaint information and performing complaint information early warning is improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 FIG. 1 is a schematic flowchart of a complaint information early warning method based on a large model provided by an embodiment of the application;

[0018] Figure 2 FIG. 2 is a schematic block diagram of a complaint information early warning device based on a large model provided by an embodiment of the application;

[0019] Figure 3 FIG. 3 is a structural schematic block diagram of a computer device provided by an embodiment of the application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0021] The flowcharts shown in the drawings are merely illustrative and are not necessarily required to include all the contents and operations / steps, nor are they necessarily executed in the order described. For example, some operations / steps can be further divided, combined, or partially merged, so the actual execution order can be changed according to actual conditions.

[0022] It should be understood that the terms used in this specification of the present application are merely for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0023] It should also be understood that the term "and / or" used in this specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0024] Embodiments of the present application provide a large model-based complaint information early warning method, device, equipment and storage medium. Among them, the large model-based complaint information early warning method can be applied to an intelligent customer service system, accurately extracts voice semantic features in call voice information through ASR, constructs a Prompt template combining user portraits, business backgrounds and historical behavior information, and comprehensively understands the real intention of the user. Through the large model, automatic complaint early warning is realized according to the Prompt template and the voice semantic features, and in the business fields of financial technology, medical health and old-age care, the efficiency of the intelligent customer service system in identifying complaint information and performing complaint information early warning is improved.

[0025] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the case of no conflict, the embodiments described below and the features in the embodiments can be combined with each other.

[0026] Please refer to Figure 1 , Figure 1 is a schematic flowchart of a large model-based complaint information early warning method provided by an embodiment of the present application. The large model-based complaint information early warning method can be applied to an intelligent customer service system, and is used to improve the efficiency of the intelligent customer service system in identifying complaint information and performing complaint information early warning in the business fields of financial technology, medical health and old-age care.

[0027] As Figure 1 shown, the large model-based complaint information early warning method specifically includes steps S10 to S30.

[0028] Step S10, determining user voice semantic features in call voice information through an automatic speech recognition engine ASR;

[0029] In one embodiment, the financial technology business field is taken as an example to expand the description of the embodiment.

[0030] Extract the conversation recording between the customer and the customer service from the intelligent customer service system as the initial voice information, and perform noise reduction, editing and format standardization on the recording to improve voice clarity and facilitate subsequent processing.

[0031] Use a high-precision automatic speech recognition engine to convert the processed voice information into text content, analyze the text content, and extract voice semantic features, including keywords (such as "insurance claim", "repayment period", "handling fee", etc.), sentence structure, semantic unit, etc.

[0032] Step S20, constructing a target prompt template according to the user portrait information, business background information and historical behavior information of the target user;

[0033] Specifically, the user portrait information is constructed by integrating the basic information of the target user (age, gender, occupation, etc.), the financial product holding situation (such as the insurance type held, the type of financial product, etc.), and the risk tolerance assessment result. Extract the business order information corresponding to the conversation, including the business type (insurance claim application, loan consultation, etc.), the business handling progress, the customer's appeal, etc., to form the business background information.

[0034] Obtain the historical complaint records, customer service consultation times, business handling frequency, etc. of the target user from the financial customer relationship management system as historical behavior information, fill the user portrait information, business background information and historical behavior information collected above according to the preset template structure, construct the target Prompt (prompt) template, and clarify the focus and direction of the prompt model analysis.

[0035] Step S30, generating a complaint information determination result of the conversation voice information according to the target Prompt template and the user voice semantic features through a preset complaint risk model, and performing complaint information early warning according to the complaint information determination result.

[0036] Specifically, the constructed target Prompt template and the extracted user voice semantic features are taken as input data and submitted to the pre-trained complaint risk model. The model analyzes and calculates the input data to generate a complaint information determination result of the conversation voice information, including whether there is a complaint risk, the type of complaint (such as complaint on service quality, complaint on financial product terms, etc.) and risk level.

[0037] According to the determination result, if there is a complaint risk and the set early warning threshold is reached, complaint information early warning is triggered immediately, and early warning information is pushed to the management personnel and relevant business processing personnel of the intelligent customer service system, so as to take timely measures for intervention and processing, such as arranging special personnel for follow-up, giving priority to solving customer problems, etc.

[0038] The application discloses a complaint information early warning method and device based on a large model, equipment and a storage medium. The complaint information early warning method based on a large model comprises determining user voice semantic features in call voice information through an automatic speech recognition engine ASR; constructing a target prompt Prompt template according to user portrait information, business background information and historical behavior information of a target user; generating a complaint information determination result of the call voice information through a preset complaint risk model according to the target Prompt template and the user voice semantic features, and performing complaint information early warning according to the complaint information determination result. In the foregoing manner, the application accurately extracts voice semantic features in call voice information through ASR, constructs a Prompt template in combination with user portrait, business background and historical behavior information, and comprehensively understands the real intention of a user. Through a large model, automatic complaint early warning is realized according to a Prompt template and voice semantic features, and in the business fields of financial technology, medical health and old-age care, the efficiency of an intelligent customer service system in identifying complaint information and performing complaint information early warning is improved.

[0039] Based Figure 1 In the embodiment, step S10 comprises:

[0040] The target user voice information of the target user is separated from the call voice information through a pre-training voiceprint recognition model built in the ASR;

[0041] The target user voice information is subjected to multi-dimensional semantic analysis through the ASR, and the user voice semantic features are generated.

[0042] Specifically, the medical health and old-age care field is taken as an example to illustrate the embodiment.

[0043] The call recording between a patient and a medical institution customer service or medical staff is collected to ensure clear recording quality and facilitate subsequent processing. A pre-training voiceprint recognition model built in the ASR (Automatic Speech Recognition) is used to distinguish and separate the voice information of the target user, and remove the voice part of the medical staff or other non-patient voices.

[0044] Identify medical-related keywords in the text, such as "headache", "fever", "drug name", "surgery", "medical insurance reimbursement", etc. These keywords are directly related to the patient's medical needs or potential complaint points. Analyze the patient's expressed semantics to determine whether it is a consultation, complaint, or suggestion. For example, the patient says "I am very dissatisfied with the effect of this medicine", which clearly indicates a complaint about the effect of the medicine.

[0045] By analyzing the tone of voice and the content of the text, determine the emotional tendency of the patient, such as "very angry", "anxious", "calm", etc. When the patient says "How come you always make me wait so long, I'm very angry", it is recognized that the patient is in an angry state.

[0046] Compare the patient's text content with the standard semantic templates in the medical field, and calculate the similarity to better understand the patient's intention. For example, compare with the "complain about doctor's service attitude" template to determine whether the patient is complaining about the doctor.

[0047] Identify the core medical topics in the patient's voice information, such as "difficult to make an appointment", "high medical costs", "poor service attitude of medical staff", etc. Clearly identify the key issues that the patient is concerned about, extract key entity information from the text, such as specific time (appointment time), location (medical department), and person (names of medical staff involved), to more specifically locate the patient's medical experience and problem.

[0048] In specific embodiments, the target user voice information is analyzed by ASR in multiple dimensions to generate user voice semantic features, including:

[0049] The target user voice information is divided into at least one target semantic unit according to the preset semantic logic by the ASR;

[0050] Specifically, according to the characteristics of medical field dialogue, the preset semantic logic is divided into different semantic units, such as different paragraphs or sentences in the patient's expression that describe symptoms, discuss treatment plans, consult medical costs, and evaluate the service of medical staff. The separated target patient voice information is input into the ASR, and the voice information is analyzed based on the preset semantic logic to identify semantic turning points and key semantic paragraphs in the voice.

[0051] According to the recognition result, the target patient voice information is divided into at least one target semantic unit according to the preset semantic logic, and each unit represents a complete medical-related semantic topic.

[0052] Each target semantic unit is evaluated for emotional tendency by the emotion analysis algorithm built into the AST, generating target user emotion features;

[0053] Specifically, the emotion features in the patient's voice are identified by using the built-in emotion analysis algorithm of the ASR, such as happy, angry, anxious, calm, etc. Each target semantic unit is input into the emotion analysis algorithm one by one to evaluate the emotional tendency of the patient in each semantic unit and determine the emotional type and intensity. The emotional tendency evaluation results of each target semantic unit are summarized to generate an emotional feature vector of the target patient, which comprehensively represents the emotional state changes of the patient in the entire conversation.

[0054] According to the target semantic units and the target user emotional features, the user voice semantic features are generated.

[0055] Specifically, the text content of each target semantic unit and its corresponding emotional features are integrated to form a comprehensive data structure containing semantic and emotional information. The key semantic information in each semantic unit is extracted, combined with the emotional features of the patient, and a specific feature fusion algorithm is used to generate a vector that can completely represent the semantic features of the patient's voice. The final output of the user voice semantic feature vector contains the main semantic content in the patient's voice (such as symptom description, treatment requirements, etc.) and the corresponding emotional state, providing rich data support for subsequent complaint information recognition and processing.

[0056] In specific embodiments, the user voice semantic features are generated according to the target semantic units and the target user emotional features, including:

[0057] Converting each target semantic unit into a target semantic feature;

[0058] Specifically, key words are extracted for each target semantic unit to identify specific terms related to financial technology, such as "loan interest rate", "repayment period", "insurance claim", "income calculation", etc. The specific expressions of users on these financial terms are analyzed, such as whether the user is asking "how to calculate the loan interest rate" or complaining "the loan interest rate is too high", etc. to clarify the specific meaning and intent expressed by each semantic unit. The extracted keywords and corresponding semantic understanding results are quantitatively encoded to form a target semantic feature vector, which is used to represent the content characteristics of each semantic unit in the user's voice.

[0059] Obtain a historical database to determine the historical fusion weights of historical semantic features and historical user emotional features in the historical database;

[0060] Specifically, the historical database of the financial technology business is accessed to extract historical semantic features and historical user emotion features. The previously identified semantic features (such as consultation content about different financial products) and emotion features (such as customer satisfaction or dissatisfaction emotions about services) are obtained from historical records. Statistical analysis methods are used to calculate the contribution degree of the historical semantic features and the historical user emotion features in predicting customer complaint risks, and to determine their historical fusion weights. For example, if it is found that the "loan interest rate" related semantic features have a high relevance in triggering customer complaints, a greater weight is given to them.

[0061] The target semantic features and the target user emotion features are fused according to the historical fusion weights to generate the user voice semantic features.

[0062] Specifically, according to the determined historical fusion weights, a feature fusion model is established, and the target semantic features and the target user emotion features are input into the model for weighted fusion calculation. Through model calculation, a user voice semantic feature vector that comprehensively considers semantic content and emotional state is generated. The user voice semantic feature vector fully and accurately represents the user's appeal, focus and emotional state in the current voice call, and provides more powerful data support for complaint risk assessment, customer service optimization and the like in the financial technology business.

[0063] Based on Figure 1 In the embodiment shown in the figure, step S20 includes:

[0064] Obtaining a blank Prompt template;

[0065] Filling the user portrait information, the business background information and the historical behavior information into the blank Prompt template to generate an initial Prompt template;

[0066] Performing integrity check and logical check on the initial Prompt template based on a preset check rule;

[0067] Determining the initial Prompt template that passes the integrity check and the logical check as the target Prompt template.

[0068] Specifically, access the template management system or related database of the financial technology company, find a blank Prompt template for complaint risk identification, obtain the user portrait information such as age, gender, occupation, income level, and risk preference of the target user from the customer database of the financial technology system, and fill them into the user portrait part of the Prompt template; extract the background information such as the business type (such as loan application, credit card handling, insurance consultation, etc.), business handling channel (online APP, offline outlets, etc.), and business handling progress from the current business interaction record, and fill them into the business background part of the Prompt template; retrieve the historical behavior data of the target user on the financial technology platform, including the type, frequency, and processing result of past complaints, and fill them into the historical behavior part of the Prompt template.

[0069] If the initial Prompt template passes the integrity check and the logicality check, it is marked as a target Prompt template and stored in a designated template database for subsequent use by the complaint risk identification model. If it fails the check, an error report will be generated indicating the specific problems, such as the lack of certain key information or the existence of logical contradictions, and then the initial Prompt template will be returned to the relevant personnel for correction and improvement until it becomes a target Prompt template.

[0070] In specific embodiments, the initial Prompt template is subjected to integrity check and logicality check based on preset check rules, including:

[0071] Check whether the initial Prompt template contains the user portrait information, the business background information, and the historical behavior information;

[0072] In the case where the initial Prompt template contains the user portrait information, the business background information, and the historical behavior information, it is determined that the Prompt template passes the integrity check;

[0073] According to the initial Prompt template, predict the initial business logic;

[0074] In the case where the initial business logic matches the user portrait information, the business background information, and the historical behavior information, it is determined that the initial Prompt template passes the logicality check.

[0075] Specifically, check if the user portrait information is complete, including whether the basic information is complete, such as missing key information such as occupation or income level, then determine it is not complete; Verify whether the business background information contains key content such as business type, channel and progress, if missing, need to supplement; Confirm whether the historical behavior information covers past complaints, consultations and transaction records, etc., to ensure the time sequence and content coherence of historical behavior.

[0076] Check if the business background information matches the user portrait information, such as whether a high-income user may apply for a low-limit loan, etc., to determine if there is a logical contradiction; Analyze whether the historical behavior information is related to the current business background, such as the user has never had a loan business before, but is now consulting loan repayment problems, then further verify the accuracy of the information.

[0077] Based on any of the above embodiments, in this embodiment, step S30 includes:

[0078] Match the target Prompt template with the preset complaint risk Prompt template through the preset complaint risk model, and match the user voice semantic features with the preset complaint risk semantic features;

[0079] In the case where the target Prompt template matches the preset complaint risk Prompt template, and the user voice semantic features match the preset complaint risk semantic features, determine that the complaint information determination result is that there is complaint information in the call voice information, and perform complaint information warning based on the complaint information determination result.

[0080] Specifically, in the complaint risk management module of the financial technology system, load the pre-trained complaint risk model, extract key information from the generated target Prompt template, such as age, gender, occupation in user portrait, business type, business progress in business background, and complaint records in historical behavior. Compare the extracted key features with the corresponding features in the preset complaint risk Prompt template, calculate the similarity or matching degree, and check if there are consistent or similar complaint risk features.

[0081] Load a database containing various complaint risk semantic features, which are extracted from past complaint cases, such as "high interest rate on loans", "unreasonable repayment method", etc. Extract key elements such as keywords, semantic topics, and sentiment orientation from the generated user voice semantic features, compare the extracted key elements with the features in the preset complaint risk semantic feature library, calculate the similarity, and determine whether the user voice contains complaint-related semantic features.

[0082] If the target Prompt template matches the preset complaint risk Prompt template, and the user voice semantic features match the preset complaint risk semantic features, i.e., the similarity reaches a preset threshold (such as 80%), it is determined that there is complaint information in the call voice information, and complaint warning information is generated, including the basic information of the user, the complaint content summary, the business type involved, and other key contents.

[0083] The warning information is pushed to the customer service managers and the heads of the relevant business departments of the financial technology company, and can be pushed in the form of a system interface reminder, an email, a short message, and the like, so that they can learn about the complaint situation in a timely manner and take corresponding measures. After the relevant personnel receive the warning information, they quickly start the complaint handling process, contact the user to understand the specific situation, and provide a solution, so as to effectively handle the complaint and reduce the risk of complaint escalation.

[0084] Please refer to Figure 2 , Figure 2 An embodiment of the present application provides a schematic block diagram of a complaint information early warning device based on a large model, which is used for executing the aforementioned complaint information early warning method based on a large model. Wherein, the complaint information early warning device based on a large model can be configured in a server.

[0085] As shown in Figure 2 , the complaint information early warning device based on a large model 400 comprises:

[0086] A user voice semantic feature determination module 410 is configured to determine user voice semantic features in call voice information through an automatic speech recognition engine ASR;

[0087] A target Prompt template construction module 420 is configured to construct a target prompt symbol Prompt template according to user portrait information, business background information, and historical behavior information of a target user;

[0088] A complaint information early warning module 430 is configured to generate a complaint information determination result of the call voice information through a preset complaint risk model according to the target Prompt template and the user voice semantic features, and perform complaint information early warning according to the complaint information determination result.

[0089] Further, the user voice semantic feature determination module 410 comprises:

[0090] A target user voice information separation sub-module is configured to separate target user voice information of a target user from the call voice information through a pre-training voiceprint recognition model built in the ASR;

[0091] The user voice semantic feature generation sub-module is configured to perform multi-dimensional semantic analysis on the target user voice information through ASR to generate the user voice semantic feature.

[0092] Further, the user voice semantic feature generation sub-module includes:

[0093] The target semantic unit generation unit is configured to divide the target user voice information into at least one target semantic unit according to a preset semantic logic through the ASR.

[0094] The target user emotion feature generation unit is configured to perform emotion tendency evaluation on each target semantic unit through a sentiment analysis algorithm built in the AST to generate a target user emotion feature.

[0095] The user voice semantic feature generation unit is configured to generate the user voice semantic feature according to each target semantic unit and the target user emotion feature.

[0096] Further, the user voice semantic feature generation unit includes:

[0097] The target semantic feature conversion sub-unit is configured to convert each target semantic unit into a target semantic feature.

[0098] The historical database acquisition sub-unit is configured to acquire a historical database and determine a historical fusion weight of a historical semantic feature and a historical user emotion feature in the historical database.

[0099] The user voice semantic feature generation sub-unit is configured to perform feature fusion on the target semantic feature and the target user emotion feature according to the historical fusion weight to generate the user voice semantic feature.

[0100] Further, the target Prompt template construction module 420 includes:

[0101] The blank Prompt template acquisition sub-module is configured to acquire a blank Prompt template.

[0102] The initial Prompt template generation sub-module is configured to fill the user portrait information, the business background information, and the historical behavior information into the blank Prompt template to generate an initial Prompt template.

[0103] The verification sub-module is configured to perform integrity verification and logical verification on the initial Prompt template based on a preset verification rule.

[0104] The target Prompt template determination sub-module is configured to determine the initial Prompt template that passes the integrity verification and the logical verification as the target Prompt template.

[0105] Further, the check sub-module comprises:

[0106] The information check unit is configured to check whether the user portrait information, the business background information and the historical behavior information are contained in the initial Prompt template.

[0107] The integrity check unit is configured to determine that the Prompt template passes the integrity check in a case where the user portrait information, the business background information and the historical behavior information are contained in the initial Prompt template.

[0108] The initial business logic prediction unit is configured to predict initial business logic according to the initial Prompt template.

[0109] The logicality check unit is configured to determine that the initial Prompt template passes the logicality check in a case where the initial business logic matches the user portrait information, the business background information and the historical behavior information.

[0110] Further, the complaint information early warning module 430 comprises:

[0111] The matching sub-module is configured to match the target Prompt template with a preset complaint risk Prompt template and match the user voice semantic features with preset complaint risk semantic features through the preset complaint risk model.

[0112] The early warning sub-module is configured to determine that the complaint information determination result is that there is complaint information in the call voice information in a case where the target Prompt template matches the preset complaint risk Prompt template and the user voice semantic features match the preset complaint risk semantic features, and perform complaint information early warning based on the complaint information determination result.

[0113] It should be noted that, for the convenience and brevity of description, the specific working processes of the above-described device and each module can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein.

[0114] The above-described device can be implemented in the form of a computer program, which can run on a computer device as shown in Figure 3 .

[0115] Please refer to Figure 3 , Figure 3 is a structural schematic block diagram of a computer device provided by an embodiment of the present application. The computer device can be a server.

[0116] Referring to Figure 3 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0117] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions that, when executed, can cause the processor to perform any one of the complaint information early warning methods based on large models.

[0118] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0119] The internal memory provides an environment for the running of the computer program in the non-volatile storage medium, which, when executed by the processor, can cause the processor to perform any one of the complaint information early warning methods based on large models.

[0120] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0121] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0122] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0123] Determine the user voice semantic features in the call voice information through an automatic speech recognition engine ASR;

[0124] Construct a target prompt Prompt template according to the user portrait information, business background information and historical behavior information of the target user.

[0125] generate a complaint information determination result of the call voice information according to the target Prompt template and the user voice semantic features, and perform complaint information early warning according to the complaint information determination result.

[0126] In one embodiment, the user voice semantic features in the call voice information are determined by an automatic speech recognition engine ASR, which is used to achieve:

[0127] The target user voice information of the target user is separated from the call voice information by a pre-trained voiceprint recognition model built in the ASR;

[0128] The target user voice information is subjected to multi-dimensional semantic analysis by the ASR to generate the user voice semantic features.

[0129] In one embodiment, the target user voice information is subjected to multi-dimensional semantic analysis by the ASR to generate the user voice semantic features, which is used to achieve:

[0130] The target user voice information is divided into at least one target semantic unit according to a preset semantic logic by the ASR;

[0131] Each target semantic unit is subjected to emotion tendency evaluation by a sentiment analysis algorithm built in the AST to generate target user emotion features;

[0132] The user voice semantic features are generated according to each target semantic unit and the target user emotion features.

[0133] In one embodiment, the user voice semantic features are generated according to each target semantic unit and the target user emotion features, which is used to achieve:

[0134] Each target semantic unit is converted into a target semantic feature;

[0135] A historical database is obtained to determine a historical fusion weight of historical semantic features and historical user emotion features in the historical database;

[0136] The target semantic features and the target user emotion features are subjected to feature fusion according to the historical fusion weight to generate the user voice semantic features.

[0137] In one embodiment, a target Prompt template is constructed according to user portrait information, business background information and historical behavior information of a target user, which is used to achieve:

[0138] A blank Prompt template is obtained;

[0139] fill the user portrait information, the business background information and the historical behavior information into the blank Prompt template to generate an initial Prompt template;

[0140] perform integrity checking and logical checking on the initial Prompt template based on a preset checking rule;

[0141] determine the initial Prompt template that passes the integrity checking and the logical checking as the target Prompt template.

[0142] In one embodiment, the integrity checking and the logical checking on the initial Prompt template based on a preset checking rule are used to achieve:

[0143] check whether the initial Prompt template contains the user portrait information, the business background information and the historical behavior information;

[0144] if the initial Prompt template contains the user portrait information, the business background information and the historical behavior information, determine that the Prompt template passes the integrity checking;

[0145] predict an initial business logic according to the initial Prompt template;

[0146] if the initial business logic matches the user portrait information, the business background information and the historical behavior information, determine that the initial Prompt template passes the logical checking.

[0147] In one embodiment, a preset complaint risk model is used to generate a complaint information determination result of the call voice information according to the target Prompt template and the user voice semantic features, and perform complaint information early warning according to the complaint information determination result, which is used to achieve:

[0148] match the target Prompt template with a preset complaint risk Prompt template through the preset complaint risk model, and match the user voice semantic features with preset complaint risk semantic features;

[0149] if the target Prompt template matches the preset complaint risk Prompt template, and the user voice semantic features match the preset complaint risk semantic features, determine that the complaint information determination result is that there is complaint information in the call voice information, and perform complaint information early warning based on the complaint information determination result.

[0150] The embodiment of the present application also provides a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program comprises program instructions, and the processor executes the program instructions to realize any item of the complaint information early warning method based on a large model provided by the embodiment of the present application.

[0151] The computer readable storage medium can be an internal storage unit of the computer device, for example, a hard disk or a memory of the computer device. The computer readable storage medium can also be an external storage device of the computer device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like.

[0152] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A complaint information early warning method based on a large model, characterized in that: include: Determine the user's speech semantic features in the call voice information through the automatic speech recognition engine ASR; Build a target prompt template based on the target user's user profile, business background information, and historical behavior information; The preset complaint risk model is used to generate a complaint information determination result of the call voice information according to the target prompt template and the user voice semantic features, and a complaint information warning is issued according to the complaint information determination result.

2. The complaint information early warning method based on a large model according to claim 1 is characterized in that: Determining the user speech semantic features in the call voice information by using the automatic speech recognition engine ASR includes: Separating the target user's voice information from the call voice information through the pre-trained voiceprint recognition model built into the ASR; ASR is used to perform multi-dimensional semantic analysis on the target user's voice information to generate the user's voice semantic features.

3. The complaint information early warning method based on a large model according to claim 2 is characterized in that: The performing multi-dimensional semantic analysis on the target user's voice information by ASR to generate the user's voice semantic features includes: Segmenting the target user voice information into at least one target semantic unit according to a preset semantic logic through the ASR; Evaluate the emotional tendency of each target semantic unit using the sentiment analysis algorithm built into the AST to generate target user emotional features; The user speech semantic feature is generated according to each of the target semantic units and the target user emotional feature.

4. The complaint information early warning method based on a large model according to claim 3 is characterized in that: Generating the user speech semantic feature according to each of the target semantic units and the target user emotional feature includes: Converting each of the target semantic units into a target semantic feature; Acquire a historical database, and determine a historical fusion weight of historical semantic features and historical user emotional features in the historical database; The target semantic feature and the target user emotion feature are fused according to the historical fusion weight to generate the user speech semantic feature.

5. The complaint information early warning method based on a large model according to claim 1 is characterized in that: The target prompt template is constructed based on the target user's user profile information, business background information, and historical behavior information, including: Get a blank Prompt template; Fill the blank Prompt template with the user portrait information, the business background information, and the historical behavior information to generate an initial Prompt template; Performing integrity and logic checks on the initial Prompt template based on preset verification rules; The initial Prompt template that passes the integrity check and the logic check is determined as the target Prompt template.

6. The complaint information early warning method based on a large model according to claim 5 is characterized in that: The integrity check and logic check of the initial Prompt template based on the preset check rules include: Verify whether the initial Prompt template contains the user portrait information, the business background information, and the historical behavior information; In a case where the initial Prompt template includes the user portrait information, the business background information, and the historical behavior information, determining that the Prompt template passes the integrity check; Predicting initial business logic based on the initial Prompt template; In the case where the initial business logic matches the user portrait information, the business background information and the historical behavior information, it is determined that the initial Prompt template passes the logic check.

7. The complaint information early warning method based on a large model according to any one of claims 1 to 6, characterized in that: The method of generating a complaint information determination result of the call voice information according to the target prompt template and the user voice semantic features by using a preset complaint risk model, and issuing a complaint information warning according to the complaint information determination result, includes: Matching the target prompt template with the preset complaint risk prompt template through the preset complaint risk model, and matching the user voice semantic features with the preset complaint risk semantic features; When the target Prompt template matches the preset complaint risk Prompt template, and the user voice semantic feature matches the preset complaint risk semantic feature, the complaint information determination result is determined to be that complaint information exists in the call voice information, and a complaint information warning is issued based on the complaint information determination result.

8. A complaint information early warning device based on a large model, characterized in that: include: A user speech semantic feature determination module is used to determine the user speech semantic features in the call voice information through the automatic speech recognition engine ASR; The target prompt template construction module is used to construct the target prompt template based on the user profile information, business background information and historical behavior information of the target user; The complaint information warning module is used to generate a complaint information judgment result of the call voice information according to the target prompt template and the user voice semantic features through a preset complaint risk model, and to issue a complaint information warning based on the complaint information judgment result.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the complaint information early warning method based on a large model as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, enables the processor to implement the complaint information early warning method based on a large model as described in any one of claims 1 to 7.

Images