Full-amount speech analysis method, device and equipment based on large model, and storage medium

Abstract

The invention discloses a full-volume voice analysis method and device based on a large model, equipment and a storage medium. The method comprises the following steps: acquiring voice interaction data of an agent and a user; according to an artificial experience rule and a large language model, extracting attributes associated with a product service in the voice data as increment mark information; obtaining full-amount marking information after standardization of the voice interaction data; training the large language model by taking the voice interaction data, the incremental marking information and the full marking information as training data to generate expert model information; and matching a target seat, a target user and a corresponding target product according to the expert model, and feeding back to a calling system, so as to push a call strategy matched with the prediction information to the target seat. Through incremental marking and total analysis of the voice interaction data, the outbound strategy and the call strategy are accurately matched for the seat, and the service efficiency and the success rate of the seat and the user are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent agent outbound calls, and in particular to a full-volume speech analysis method, device, equipment and storage medium based on a large model. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models (LLMs) have shown strong capabilities in the field of natural language processing (NLP), especially in understanding, analyzing and generating human language. In the field of customer service, the application of these technologies is gradually increasing, aiming to improve service efficiency and quality through automation and intelligent means.

[0003] Traditional customer service systems usually rely on preset scripts and rules. These systems perform well when handling standardized inquiries, but are often inflexible and inaccurate when faced with complex or non-standardized user needs. In addition, when handling a large number of customer inquiries, human agents may suffer from information overload, emotional fatigue or knowledge limitations, which may affect service quality.

[0004] To solve these problems, some systems have begun to try to introduce machine learning algorithms to analyze customer interaction data to provide more personalized services. However, these early solutions are often limited by the complexity of the algorithm, the quality of the data, and the processing power, making it difficult to achieve large-scale application and real-time response.

[0005] In recent years, with the improvement of computing power and the increase of data volume, large language models have made significant progress in processing complex language tasks. These models can understand and generate natural language, providing new possibilities for automated customer service. However, how to effectively apply these models to actual customer service scenarios, especially in real-time voice interactions, remains a technical challenge. Summary of the invention

[0006] The purpose of this application is to provide a full-volume speech analysis method, device, equipment and storage medium based on a large model, so as to at least solve the problems of inaccurate outbound call strategies, low flexibility and low accuracy in the current outbound call industry.

[0007] In order to solve the above technical problems, the present application provides a full-volume speech analysis method based on a large model, including: Obtain voice interaction data between agents and users; Extracting attributes associated with product services from the voice interaction data as incremental tag information based on artificial experience rules and a large language model; Obtaining full amount of label information after normalization of the voice interaction data; Using the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; According to the expert model, the target agent, target user and corresponding target product are matched and fed back to the call system, so as to obtain the real-time communication data between the target agent and the target user from the call system, and predict the prediction information including user needs and user behavior patterns based on the real-time communication data according to the expert model, so as to push the call strategy matching the prediction information to the target agent.

[0008] Optionally, extracting attributes associated with product services from the voice interaction data according to artificial empirical rules and a large language model as incremental tag information includes: Extracting attributes associated with a preset prompt word from the voice interaction data; Marking the attributes associated with the prompt word for the first time according to the artificial experience rule; Performing a second labeling on attributes associated with product services in the voice interaction data according to the large language model; Incremental marking information is determined according to the first marking and the second marking.

[0009] Optionally, after using the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, the method further includes: Testing the expert model using test data to obtain a test result of the expert model, wherein the test result includes a prompt word tag of the test data; The test result is fed back to the prompt word project to update the prompt word associated attributes in the prompt word project according to the prompt word tags of the test data.

[0010] Optionally, after using the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, the method further includes: Obtain at least one update information of iterative update prompt words, product business, seat feedback, and user behavior pattern; Iteratively update the artificial experience rule according to the update information or make the large language model learn the update information; Incrementally mark the interaction data according to the updated manual experience rules or large language model to obtain updated incremental marking information; The expert model is updated based on the updated incremental labeling information.

[0011] Optionally, extracting attributes associated with product services from the voice interaction data according to artificial empirical rules and a large language model as incremental tag information includes: Determining an application field of the voice interaction data; Matching personalized analysis rules according to the application field, and extracting user needs and preferences in the voice interaction data based on the personalized analysis rules; The voice interaction data is marked according to the user needs and preferences to update the incremental marking information.

[0012] Optionally, the obtaining of full tag information of the normalized voice interaction data includes: Recognizing the voice interaction data and generating text information of the voice interaction data; Normalizing the text information to obtain normalized text information; Segmenting the normalized text information of the voice interaction data according to the minimum feature units of several classification rules to obtain voice data texts containing several minimum feature units; The minimum feature unit under different classification rules is extracted and the corresponding attribute information is matched, so that the proportion of the attribute information contained in the minimum feature unit of the voice data text under different classification rules reaches a preset value, and the full amount of labeling information after normalization of the voice interaction data is obtained.

[0013] Optionally, the pushing a call strategy matching the prediction information to the target agent includes: According to the user behavior pattern, push the corresponding speech strategy to the target agent; Determine whether the user demand is consistent with the target product. If so, push introduction information related to the target product to the target agent; if not, match the intended product according to the user demand, and push introduction information related to the intended product to the target agent.

[0014] In order to solve the above technical problems, the present application also provides a full-volume speech analysis device based on a large model, comprising: Data acquisition module, used to obtain voice interaction data between agents and users; An incremental marking module, used to extract attributes associated with product services from the voice interaction data according to manual experience rules and a large language model as incremental marking information; A full-quantity marking module, used to obtain full-quantity marking information of the voice interaction data after normalization; A model training module, used to train the large language model using the voice interaction data, the incremental tag information, and the full tag information as training data to generate an expert model, wherein the expert model establishes associated information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; The model application module is used to match the target agent, target user and corresponding target product according to the expert model and feed back to the call system, so as to obtain the real-time communication data between the target agent and the target user from the call system, and predict the prediction information including user needs and user behavior patterns based on the real-time communication data according to the expert model, so as to push the call strategy matching the prediction information to the target agent.

[0015] In order to solve the above technical problems, the present application also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the above-mentioned large model-based full-scale speech analysis method.

[0016] In order to solve the above technical problems, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the above-mentioned large model-based full-scale speech analysis method.

[0017] The beneficial effects of the embodiments of the present application are as follows: by acquiring voice interaction data between agents and users; extracting attributes related to product business in the voice interaction data as incremental tag information according to artificial experience rules and a large language model; acquiring full tag information after normalization of the voice interaction data; using the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; matching target agents, target users, and corresponding target products according to the expert model and feeding back to the call system to obtain the target agent from the call system. The real-time communication data between the agent and the target user is collected, and the expert model is used to predict the prediction information including user needs and user behavior patterns based on the real-time communication data, so as to push the call strategy matching the prediction information to the target agent; through the incremental marking and full analysis of voice interaction data, combined with the deep learning of the large language model and artificial experience rules, the efficient analysis and processing of voice interaction data is realized, the accuracy of the prediction of user needs and behavior patterns is improved, and the service personalization and response speed are enhanced; through the analysis of real-time interaction data, accurate call strategies are provided for the agents, and the customer service experience is optimized; at the same time, a comprehensive user, agent and product portrait is established, which provides data support for business decision-making and improves operational efficiency and market competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A schematic diagram of the basic flow of a full-volume speech analysis method based on a large model according to a specific embodiment of the present application; Figure 2 A schematic diagram of the basic structure of a full-volume speech analysis device based on a large model according to a specific embodiment of the present application; Figure 3 This is a basic structural block diagram of a computer device according to a specific embodiment of the present application. DETAILED DESCRIPTION

[0019] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0020] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0021] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0022] It will be understood by those skilled in the art that the term "terminal" as used herein includes both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capability, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of performing two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices, which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which may combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, a pager, Internet / intranet access, a web browser, a notepad, a calendar and / or a GPS (Global Positioning System) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with a radio frequency receiver. The "terminal" used here can be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable and / or configured to operate locally, and / or in a distributed form, at any other location on the earth and / or in space. The "terminal" used here can also be a communication terminal, an Internet terminal, a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device) and / or a mobile phone with music / video playback function, or a smart TV, a set-top box and other devices.

[0023] The hardware referred to by the names such as "server", "client", and "service node" in this application is essentially an electronic device with capabilities equivalent to those of a personal computer. It is a hardware device that has the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit calls the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.

[0024] It should be pointed out that the concept of "server" referred to in this application can also be extended to the case of server clusters. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or integrated into a physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility, and should not use it to restrict the implementation of the network deployment method of this application.

[0025] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for access.

[0026] The AI ​​models and large language models cited or may be cited in this application, unless explicitly specified, can be deployed on a remote server and remotely called on the client, or can be deployed and directly called by a client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning so as to reduce the requirements for the client hardware operating resources and avoid excessive occupation of the client hardware operating resources.

[0027] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as it is suitable for being called by the technical solution of this application.

[0028] Those skilled in the art should be aware that, although the various methods of the present application are described based on the same concept and thus present commonality to each other, unless otherwise specified, these methods can be independently executed. Similarly, for each embodiment disclosed in the present application, they are all proposed based on the same inventive concept, therefore, concepts with the same expression, and concepts that are appropriately changed for convenience despite different expressions, should be understood as equivalent.

[0029] Unless the mutually exclusive relationship between the embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct a new embodiment, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.

[0030] See also Figure 1 , Figure 1 The figure is a basic flow chart of the full-volume speech analysis method based on a large model in this embodiment.

[0031] like Figure 1 As shown, including: S1100, obtaining voice interaction data between the agent and the user; This embodiment is applied to outbound call scenarios in multiple fields such as finance, e-commerce, insurance, and education. In this embodiment, first, a voice recognition system is deployed in the call center to capture the voice communication between the agent and the user in real time. Furthermore, the voice interaction data will be stored in a secure database for subsequent processing and analysis. In order to ensure the integrity and quality of the data, a preset noise reduction technology is used to reduce background noise and improve voice clarity to avoid mixing in a lot of noise. In addition, during the voice interaction data collection process, corresponding reminders and measures are formulated in accordance with relevant privacy protection laws and regulations to ensure the security and privacy of user information.

[0032] It should be noted that, in addition to the basic voice content, this embodiment also collects other data types related to user interaction, such as call duration, user access channel, and agent response time, when acquiring the voice interaction data between the agent and the user. These multi-dimensional data can provide a more comprehensive perspective for subsequent analysis. For example, call duration may be related to user satisfaction, while a quick response may improve user experience. By combining these data, service efficiency and user satisfaction can be more accurately evaluated, thereby optimizing service processes and improving service quality.

[0033] It should be noted that when collecting and processing the voice interaction data between the agent and the user, this embodiment obtains the user's explicit consent before data collection in accordance with data protection regulations and privacy policies, and clearly explains the purpose and scope of data use to the user.

[0034] It should be noted that in order to improve the service response speed and quality, this embodiment deploys a real-time voice analysis function module, which can instantly identify key information and emotional changes during a call and feed this information back to the agent. For example, if the system detects that the user is emotionally excited or dissatisfied, the agent can be notified immediately to take appropriate mitigation measures. In addition, the real-time data analysis function module can also be used to dynamically adjust service strategies, such as automatically upgrading service priority when the user waits too long, to improve the overall service experience.

[0035] S1200, extracting attributes associated with product services from the voice interaction data according to manual experience rules and a large language model as incremental tag information; After acquiring the voice interaction data between agents and users, we use artificial experience rules, that is, experienced business analysts define key attributes according to business needs, such as business attributes, user attributes, problem attributes, problem-solving attributes, etc., and then extract relevant attributes from the voice interaction data. At the same time, we use a large language model to conduct in-depth analysis of the voice interaction data to identify and extract keywords and phrases related to the product's business attributes. For example, through natural language processing technology, the model can identify users' positive or negative comments on a product and use them as incremental tag information. This information will be used to train the large language model to improve the model's ability to identify and understand business-related attributes.

[0036] It should be pointed out that when this embodiment extracts attributes associated with product business based on artificial experience rules and large language models, it also deeply understands the context and situation of voice interaction data. In one embodiment, it not only identifies keywords and phrases in voice interaction data, but also analyzes their context and semantics in the conversation. For example, the same words may have different meanings or importance in different situations. Through situational analysis, the user's true intentions and needs can be more accurately identified, thereby providing richer training data for the large language model. In addition, situational analysis can also help the model learn how to apply rules in different business scenarios, improving the adaptability and accuracy of the model.

[0037] It should be pointed out that the extraction of attributes associated with product business in voice interaction data in this embodiment is a dynamic process, which needs to be continuously updated and optimized as the market changes and user behavior evolves. To this end, in one implementation method, a continuous learning mechanism is established to allow the large language model to regularly receive new training data and adjust itself according to the latest business rules and user feedback. In addition, the accuracy and effectiveness of the model can be evaluated through user feedback and agent feedback loops, the shortcomings of the model can be identified, and targeted optimization can be performed. This process of continuous learning and optimization helps to maintain the advancement and competitiveness of the model and ensure that it remains efficient and accurate in a changing market environment.

[0038] S1300, obtaining full amount of label information of the normalized voice interaction data; While extracting attributes related to product business from the voice interaction data as incremental tag information based on artificial experience rules and the large language model, the acquired voice interaction data is preprocessed, including voice-to-text, text cleaning, and word segmentation. Then, the text data is normalized according to predefined tagging rules to ensure the consistency of the data format. For example, all date formats are unified into YYYY-MM-DD, and the normalized full tag information will be used as part of the training data to help the large language model better learn and understand the business-related information in the voice interaction data.

[0039] It should be pointed out that this embodiment needs to take into account the multi-language and dialect differences of users in different regions when performing normalized processing of voice interaction data. Since the voice features and expressions of different languages ​​and dialects may differ significantly, the normalized processing system needs to have the ability to process multiple languages. This includes but is not limited to speech recognition, transcription, and text processing of different languages. For example, for Chinese, it may be necessary to distinguish between dialects such as Mandarin and Cantonese, and adjust the speech recognition algorithm according to regional characteristics to improve recognition accuracy. In addition, texts in different languages ​​also need to be normalized accordingly, such as unified punctuation use, vocabulary standardization, etc., to ensure the consistency and analyzability of the data.

[0040] It should be pointed out that in addition to basic speech-to-text and text cleaning, normalization processing also includes analysis of sentiment and intonation in voice interaction data. Sentiment analysis can help identify the user's emotional state, such as satisfaction, anger, or confusion, while intonation recognition can provide information about the user's tone and emphasis, which is essential for understanding the user's true intentions and needs. For example, even if the user's language expression is positive, if the tone is angry or disappointed, it may indicate that there are unresolved issues or dissatisfaction. By incorporating sentiment and intonation information into the normalized full amount of labeled information, richer contextual information can be provided to the large language model, thereby improving the model's understanding and prediction of user needs.

[0041] When emotion and intonation are recognized, the recognized information is the target state information. The text information converted from the voice interaction data needs to be converted into sentences according to the target state information. Sentence conversion refers to converting the declarative sentences in the text information into interrogative, rhetorical, and exclamatory sentences according to the sentence type corresponding to the target state information. After the sentence conversion, the real intention information in the text information is determined based on the generated conversion sentence, and the relevance of the text information is re-identified based on the real intention information, and the cause information whose relevance to the real intention information is greater than the set relevance threshold is highlighted to increase the weight of the cause information. The target state information is the emotion and intonation represented by negative emotions.

[0042] In the prior art, when identifying text information, the most easily overlooked information is modal particles, which are often identified as words without specific meanings and are screened out. However, in the real context interaction process, modal particles are often the words that can best express the user's true intentions. Through sentence conversion, the declarative sentences in the text information are converted into interrogative, rhetorical, and exclamatory sentences with modal particles. This makes it easier to identify the user's true intention information, and then identify the cause information in the text information that leads to the true intention information through the true intention information, and increase the probability of the cause information being identified and understood by increasing the weight of the cause information. Improve the accuracy of user intention recognition, and then improve the accuracy of subsequent processing results.

[0043] S1400, using the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; After extracting the attributes related to the product business in the voice interaction data according to artificial experience rules and the large language model as incremental label information and obtaining the full label information after normalization of the voice interaction data, the above collected and processed data is used as a training set and input into the large language model for training. During the training process, the model will learn how to identify key attributes from the voice interaction data and establish associations between user portraits, agent portraits, and product portraits. In addition, the model will also learn to predict user needs and behavior patterns. In order to improve the accuracy and generalization ability of the model, a variety of machine learning algorithms and techniques are used, such as deep learning, transfer learning, and cross-validation.

[0044] It should be pointed out that when training a large language model, this embodiment needs to ensure the balance and diversity of the training data, including avoiding bias in the data set and ensuring that all types of users, seats, and products are reasonably represented. In one embodiment, if certain types of user interactions in the data set are too frequent, it may cause the model to overfit these specific interaction patterns, thereby affecting the model's generalization ability for other types of interactions. To solve this problem, this embodiment uses data enhancement techniques, such as synthesizing samples of a minority category or balancing the number of samples of different categories through resampling techniques. In addition, this embodiment also introduces interaction data from different scenarios and backgrounds to enhance the adaptability of the model to different business scenarios.

[0045] It should be pointed out that in the process of training a large language model to generate an expert model, this embodiment can use ensemble learning technology to improve the prediction accuracy and robustness of the model. Ensemble learning combines the prediction results of multiple models to form a more comprehensive decision. For example, different algorithms or model architectures can be used to process the same training data set, and then the outputs of these models can be integrated through voting, weighted averaging, or stacking. This method can reduce the model's dependence on specific data features and improve its adaptability to new situations. In addition, model fusion can also help identify and solve possible deviations in a single model, thereby enhancing the performance of the expert model when processing complex and diverse voice interaction data.

[0046] It should be pointed out that the training of the expert model of this embodiment should not be limited to the initial stage, but should include a continuous learning and adaptive process. This means that the model needs to be able to adjust according to real-time feedback. For example, after the model generates prediction results in actual applications, the accuracy of these predictions can be evaluated through feedback from agents or users. If it is found that the prediction results deviate from the actual situation, the model should be able to use this feedback information to self-correct and optimize. This adaptive learning capability enables the expert model to evolve over time and better adapt to changes in user needs and behavior patterns, thereby providing more accurate and personalized services.

[0047] It should be pointed out that in the process of training the expert model in this embodiment, feature engineering is a key step, which involves extracting features from the original data that are helpful for model learning, including extracting acoustic features from voice interaction data, extracting language features from text data, and extracting behavioral features from interaction data. Through carefully designed features, the performance of the model can be significantly improved. At the same time, model tuning is also a continuous process, and hyperparameters such as model parameters and learning rate need to be adjusted according to the performance of the model on the validation set. In addition, this embodiment also uses regularization technology to prevent model overfitting, and evaluates the stability and generalization ability of the model through cross-validation. Through these methods, it is ensured that the expert model is more accurate and reliable when establishing user portraits, agent portraits, and product portraits.

[0048] S1500, matching the target agent, the target user, and the corresponding target product according to the expert model and feeding back to the call system, so as to obtain the real-time communication data between the target agent and the target user from the call system, and predicting the prediction information including the user demand and the user behavior pattern based on the real-time communication data according to the expert model, so as to push the call strategy matching the prediction information to the target agent.

[0049] After the expert model is generated, it will match the most suitable agent and product based on the correlation information of the user portrait, agent portrait and product portrait, that is, it will make intelligent recommendations based on real-time and historical data. For example, if a user shows a high interest in a specific product in the interaction data, the system will recommend an agent with corresponding product knowledge and high user evaluation to serve the user. In addition, when the agent communicates with the user in real time, the expert model will analyze the communication content in real time, extract key information, and predict the user's needs and behavior patterns, and feed back the prediction results to the agent in real time to help the agent better understand user needs and provide more personalized services. At the same time, based on the prediction results of the expert model, the system will automatically generate call strategies and push them to the agent in real time. These strategies may include how to better guide the conversation, solve user problems or promote specific products. The push of call strategies can be achieved through the user interface of the call system, and the agent can view and apply these strategies at any time during the communication process to improve service efficiency and user satisfaction.

[0050] In the above implementation, voice interaction data between agents and users is obtained; attributes related to product services in the voice interaction data are extracted as incremental tag information according to artificial experience rules and a large language model; full tag information after normalization of the voice interaction data is obtained; the voice interaction data, the incremental tag information, and the full tag information are used as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; target agents, target users, and corresponding target products are matched according to the expert model and fed back to the call system to obtain the target agents and target users from the call system. The real-time communication data of users is collected, and the expert model is used to predict the prediction information including user needs and user behavior patterns based on the real-time communication data, so as to push the call strategy matching the prediction information to the target agent; through the incremental marking and full analysis of voice interaction data, combined with the deep learning of the large language model and the artificial experience rules, the efficient analysis and processing of voice interaction data is realized, the accuracy of the prediction of user needs and behavior patterns is improved, and the personalization and response speed of services are enhanced; through the analysis of real-time interaction data, accurate call strategies are provided for agents, and the customer service experience is optimized; at the same time, a comprehensive user, agent and product portrait is established, which provides data support for business decision-making and improves operational efficiency and market competitiveness.

[0051] In some implementations, S1200 extracts attributes associated with product services from the voice interaction data according to artificial experience rules and a large language model as incremental tag information, including: S1211, extracting attributes associated with a preset prompt word in the voice interaction data; In this embodiment, when extracting attributes related to the product business in the voice interaction data according to artificial experience rules and a large language model as incremental tagging information, first, a set of prompt words closely related to the product business is defined. These prompt words may include product names, specific services, user common problems, etc., and the voice interaction data is converted into text data using speech recognition technology. Then, the prompt words and their contextual information appearing in the text are automatically identified through keyword matching technology. For example, if a user mentions the name of a product in a conversation, the system will capture the name and the conversation content around it, which may contain information such as the user's evaluation of the product, usage experience, or problems encountered. The extracted attributes will serve as the basis for further analysis and provide raw data for the subsequent tagging process.

[0052] S1212, marking the attributes associated with the prompt word for the first time according to the artificial experience rule; After extracting the attributes associated with the preset prompt words in the voice interaction data, experienced business analysts will develop a set of detailed manual experience rules based on business knowledge and historical data analysis results. These rules will guide the labeling process to ensure the consistency and accuracy of the labeling. For example, if a user mentions the product name and then expresses dissatisfaction, this may be marked as "negative feedback" according to manual experience rules. Business analysts will conduct a preliminary review of the extracted attributes and classify and label them according to the rules, such as "positive feedback", "technical issues", "service requests", etc., to provide guidance for subsequent large language model analysis.

[0053] S1213. Perform a second marking on the attributes associated with the product business in the voice interaction data according to the large language model; Based on the first labeling, the voice interaction data is deeply analyzed using the trained large language model, which uses natural language processing and machine learning techniques to identify and classify more subtle semantic information and emotional tendencies. In one embodiment, the large language model can identify subtle differences in the user's tone, such as disappointment, anger, or satisfaction, and label it more accurately accordingly. The second labeling will supplement and refine the results of the first labeling, improve the depth and breadth of the labeling, and enable the model to capture more complex user needs and behavior patterns.

[0054] S1214. Determine incremental marking information according to the first marking and the second marking.

[0055] Combining the results of the first and second labeling, the final incremental labeling information is determined by comparing and integrating the two types of labeling information. This step may involve manual review and model verification to ensure the accuracy and reliability of the labeling information. For example, if manual labeling identifies the user's need for a certain function, and the large language model labels the user's satisfaction with the same function, these two types of information can be integrated into a more comprehensive attribute, such as "functional requirement satisfaction". The incremental labeling information will be used as part of the training data to train the expert model, thereby improving the model's ability to predict user needs and behavior patterns. Through this multi-level and multi-angle labeling process, it can be ensured that the expert model is more accurate and comprehensive in understanding and predicting user behavior.

[0056] This implementation method can more accurately extract attributes related to product business from voice interaction data by combining artificial experience and intelligent analysis of a large language model, thereby ensuring the quality and relevance of the labeled information; artificial experience rules provide in-depth knowledge of the business field, while the large language model captures patterns and trends in the data through machine learning technology. The combination of the two helps to improve the model's adaptability and generalization capabilities to different situations; accurate incremental labeled information provides rich training data for the expert model, which helps to more accurately establish user portraits, agent portraits, and product portraits, as well as more effectively predict user needs and behavior patterns; based on more accurate predictions of user needs and behavior, more personalized services can be provided to users, improving user satisfaction and loyalty; with the continuous learning and optimization of the model, service processes and product quality can be continuously improved to adapt to market changes and the evolution of user needs.

[0057] In some implementations, after the S1400 uses the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, the step further includes: S1411, testing the expert model using test data to obtain a test result of the expert model, wherein the test result includes a prompt word tag of the test data; After the expert model training is completed, this embodiment selects a set of test data sets that are similar to the training data distribution but have not been used for model training to evaluate the performance of the model. The test data set contains a variety of voice interaction scenarios to ensure that the generalization ability of the model is fully verified. During the test, the expert model will analyze the voice interactions in the test data set, identify the prompt words therein, and mark them according to the knowledge learned in the previous training process. The test results will include key performance indicators such as the recognition accuracy and recall rate of the model for the prompt words in the test data, as well as the model's prediction accuracy of user needs and behavior patterns.

[0058] It should be noted that after the test is completed, the output results of the expert model are collected and analyzed. The test results record the model's recognition of each prompt word in the test data, including correct recognition, missed recognition, and incorrect recognition. In addition, the test results will also include the accuracy of the model's prediction of user needs and behavior patterns, such as predicting user satisfaction, product preferences, etc. Through in-depth analysis of the test results, the performance of the model in actual applications can be evaluated and possible deficiencies of the model can be identified.

[0059] S1412: Feedback the test result to the prompt word project, so as to update the attributes associated with the prompt words in the prompt word project according to the prompt word tags of the test data.

[0060] After testing the expert model with test data and obtaining the test results of the expert model, the test results of the expert model are fed back to the person responsible for the prompt word project. The prompt word project includes an automatic verification tool, as well as relevant business analysts and data analysts, who can review and adjust the prompt word project based on the test results. If the test results show that the recognition accuracy of certain prompt words is low, the relevant manual experience rules are reviewed, or the large language model is further trained and optimized. In addition, if the test results reveal new user needs or behavior patterns, this information can be used to expand or update the attribute associations in the prompt word project to improve the model's adaptability to new situations.

[0061] It should be noted that, in one embodiment, updating the prompt words and associated attributes in the prompt word engineering according to the test results includes adding new prompt words, deleting or revising the attribute association of existing prompt words, or adjusting the classification mode of attributes. The updating process combines the professional knowledge and user feedback of the business analyst to ensure that the updated prompt word engineering can more accurately reflect the real needs and behavior patterns of the user. In addition, the updated prompt word engineering will be used for a new round of model training to achieve continuous improvement and optimization of model performance. Through this iterative testing and feedback mechanism, it can be ensured that the expert model always maintains a high degree of accuracy and relevance.

[0062] This implementation method can timely discover and correct deviations in the expert model in actual application through the evaluation of test data, thereby ensuring the accuracy of the model output; the test results are fed back to the prompt word project, so that the model can continuously evolve according to the latest business needs and user behaviors, maintaining its advancement and adaptability; by optimizing the model's recognition of prompt words and prediction of user needs, it can provide users with more accurate and personalized services and enhance user satisfaction; in-depth analysis of test results provides enterprises with valuable insights, helping them better understand the market and customer needs, and guiding the improvement of products and services.

[0063] In some implementations, after the S1400 uses the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, the step further includes: S1421: Obtain at least one update information of iterative update prompt words, product business, seat feedback, and user behavior pattern; In this embodiment, after the expert model is generated and put into use, it continuously monitors and collects updated information related to prompt words, product business, agent feedback, and user behavior patterns, including new user feedback, market trends, product updates or service changes, etc. By analyzing customer service records, online surveys, social media, and other data collected through other channels, areas that need to be updated can be identified. For example, if a new feature is launched and users’ inquiries about the feature increase, this indicates that the prompt words need to be updated to include relevant discussions about the new feature.

[0064] S1422: Iteratively update the artificial experience rule according to the update information or enable the large language model to learn the update information; Furthermore, the collected update information is applied to the iterative update of the artificial experience rules, or is input into the large language model as new training data so that the model can learn this new information. For artificial experience rules, the rules are re-evaluated and adjusted based on the newly collected data to ensure that the rules can adapt to new business scenarios. For large language models, new data is integrated into the existing model through incremental learning, without the need to train from scratch, thereby improving learning efficiency and reducing the consumption of computing resources.

[0065] S1423: incrementally marking the interaction data according to the updated artificial experience rules or the large language model to obtain updated incremental marking information; Using updated manual experience rules or learned large language models, incrementally label new interaction data. This step will apply new rules or model capabilities to identify and label attributes or patterns that may not have been recognized before. For example, if the updated information contains new insights into specific user behaviors, the large language model will be able to more accurately label these behaviors as part of the incremental labeling information.

[0066] S1424: Update the expert model based on the updated incremental labeling information.

[0067] Integrate the new information obtained through incremental labeling into the expert model to update and optimize the model's predictive ability, including adjustments to model parameters, optimization of model structure, or introduction of new features. The updated expert model will be able to better reflect the latest user needs and behavior patterns, and provide more accurate predictions and recommendations. In addition, this update process can be automated, ensuring that the model can quickly adapt to new situations and maintain its effectiveness and relevance in practical applications. Through this continuous learning and updating, the expert model can keep its predictive ability up to date and provide users with the best service experience.

[0068] This implementation method ensures that the expert model can adapt to new business needs and market changes and improve the accuracy of prediction and analysis through continuous iterative updates; the integration of updated information enables the model to better understand and respond to the evolution of user behavior, and improve the personalization and timeliness of services; as the model performance improves, users can obtain more accurate and considerate services, enhancing user satisfaction and loyalty; through incremental learning and rule updates, the need to retrain the model is reduced, reducing the cost and complexity of maintenance and updates.

[0069] In some implementations, S1200 extracts attributes associated with product services from the voice interaction data according to artificial experience rules and a large language model as incremental tag information: S1221. Determine an application field of the voice interaction data; In this embodiment, first, the collected voice interaction data is classified to determine the application field to which it belongs. The application fields may include communication services, financial services, medical consultation, customer service, technical support, etc. This step involves a preliminary review of the data to identify key field-specific terms and contexts. For example, the communication service field may include terms such as communication packages, tariffs, and product strength, and interactions in the financial services field may include terms such as loans, interest rates, and investments, while medical consultations may involve words such as symptoms, drugs, and treatments. By determining the application field, contextual information can be provided for subsequent personalized analysis to ensure the accuracy and relevance of the analysis.

[0070] S1222. Matching personalized analysis rules according to the application field, and extracting user needs and preferences in the voice interaction data based on the personalized analysis rules; After determining the application field of voice interaction data, the corresponding personalized analysis rules are matched according to the characteristics and needs of the field. These rules are jointly formulated by business experts and data scientists, or formulated based on industry rules, aiming to capture field-specific user needs and preferences. For example, the communication service field can focus on communication packages, tariffs, product strength, etc. In the financial services field, the rules may focus on identifying users' financial goals, risk preferences, and product choices. By applying these rules, key information can be extracted from voice interaction data, such as users' interest in specific financial products or satisfaction with services. Using large language models and artificial experience rules, voice interaction data is deeply analyzed to identify and extract users' explicit and implicit needs and preferences. This includes analyzing users' language expressions, emotional tendencies, and interaction patterns. For example, through natural language processing technology, users' positive comments on a product or dissatisfaction with services can be identified. In addition, pattern recognition technology can be used to discover common problems or demand patterns of users in interaction.

[0071] S1223. Mark the voice interaction data according to the user needs and preferences to update the incremental marking information.

[0072] After matching personalized analysis rules according to the application field and extracting user needs and preferences in the voice interaction data based on the personalized analysis rules, the voice interaction data is marked according to the extracted user needs and preferences to generate updated incremental marking information. Specifically, the identified key information is associated with specific tags, such as marking a user's positive evaluation as "high satisfaction" or marking a user's question as "technical support required".

[0073] S1224. Update the expert model based on the updated incremental marking information.

[0074] After the voice interaction data is marked according to the user's needs and preferences to update the incremental marking information, the expert model is updated based on the updated incremental marking information. The incremental marking information will be used as part of the training data to train and optimize the large language model to improve the model's understanding and prediction capabilities of user needs. Through continuous iteration and updating, the incremental marking information will be richer and more accurate, thereby providing enterprises with more valuable insights and decision support.

[0075] This implementation method determines the application field of voice interaction data and implements customized analysis rules, thereby enhancing the adaptability and analysis depth of the model to specific business scenarios. The application of personalized analysis rules enables the model to more accurately identify and understand users' explicit and implicit needs and preferences, thereby improving the personalization level of services. The automated labeling process not only improves efficiency, but also optimizes the model's training data and enhances the model's prediction and analysis capabilities through continuously updated incremental labeling information. An in-depth understanding of user needs and preferences helps companies improve service and product design, better meet market demand, and enhance user satisfaction and corporate competitiveness.

[0076] In some implementations, S1300 obtains the full amount of tag information of the normalized voice interaction data, including: S1311, identifying the voice interaction data, and generating text information of the voice interaction data; In this embodiment, high-precision speech recognition technology (such as an automatic speech recognition ASR system) is used to process voice interaction data, converting voice signals into text information. This includes accurately capturing the voice characteristics of different speakers, including accents, speech rates, and background noise, etc. The converted text information will serve as the basis for subsequent processing. To improve the recognition accuracy, in one implementation, a multi-engine recognition strategy is adopted, that is, the same voice interaction data is processed in parallel by multiple speech recognition engines, and then the best result is selected through a voting mechanism or a post-processing algorithm.

[0077] S1312. Normalize the text information to obtain normalized text information; After recognizing the voice interaction data to generate the text information of the voice interaction data, the generated text information is normalized to eliminate inconsistencies and redundant information in the text. The normalization process includes, but is not limited to, text standardization (such as uniformly using all lowercase or all uppercase), removing stop words (common words that are not very meaningful for analysis, such as "de", "shi", etc.), stemming or lemmatization (还原词汇为基本形式), and synonym replacement, etc. These processes help reduce the complexity of the data and improve the efficiency and accuracy of subsequent analysis.

[0078] S1313. Segment the normalized text information of the voice interaction data according to the minimum feature units of several classification rules to obtain a voice data text containing several minimum feature units; After normalizing the text information to obtain normalized text information, according to several predefined classification rules, the normalized text information is segmented into minimum feature units. These minimum feature units can be words, phrases, or sentences, etc. The design of the classification rules should be based on business requirements and data characteristics. For example, the rules can be designed according to semantic relevance, grammatical structure, or the characteristics of the business domain. Through this segmentation, complex text information can be decomposed into basic units that are easy to manage and analyze.

[0079] S1314. Extract the minimum feature units under different classification rules and match the corresponding attribute information, so that the proportion of the minimum feature units of the voice data text under different classification rules containing the attribute information reaches a preset value, to obtain the full amount of labeled information after normalizing the voice interaction data.

[0080] Based on the smallest feature unit obtained by segmentation, key information is further extracted and matched with the corresponding attribute information. This step involves technologies such as entity recognition and relationship extraction in natural language processing to identify named entities in the text (such as names of people, places, product names, etc.) and the relationships between them. Through matching, the smallest feature unit can be associated with business-related attributes, such as associating user comments on products with product attributes.

[0081] Furthermore, in order to ensure the quality and relevance of the labeled information, it is necessary to set a preset value, that is, the proportion of attribute information contained in the minimum feature unit under different classification rules. Through statistical analysis, the proportion of attribute information of each minimum feature unit is evaluated and adjusted as needed. If the proportion of attribute information of some minimum feature units is lower than the preset value, the proportion can be increased by supplementing data, adjusting classification rules, or optimizing attribute matching algorithms. This process helps to ensure the richness and accuracy of the full amount of labeled information and provide high-quality input data for subsequent big data analysis and model training. This implementation ensures the accuracy and consistency of data through speech recognition and text normalization processing, providing high-quality input for subsequent analysis; adopts classification rules to segment text information, making the data more structured, facilitating in-depth semantic analysis and information extraction; by extracting the smallest feature unit and matching attribute information, the recognition rate of key information is improved, making the data richer and more insightful; by ensuring that the proportion of attribute information reaches the preset value, the comprehensiveness and coverage of the data are guaranteed, providing a solid foundation for building a comprehensive user and product portrait; the normalized full-volume labeled information provides high-quality data for the training of the large language model, which helps to improve the model's predictive ability and business application effect.

[0082] In some implementations, the step S1500 pushes a call strategy matching the prediction information to the target agent, including: S1511. Pushing corresponding speech strategies to target agents based on the user behavior pattern; In this embodiment, when pushing a call strategy that matches the predicted information to the target agent, first, the expert model will analyze the user's historical interaction data and current behavior pattern in real time to identify the user's personality traits, preferences, and communication style. Based on these analysis results, the system will select the most appropriate speech strategy from the pre-designed speech library. These speech strategies may include different communication skills such as open-ended questions, affirmative feedback, and product feature emphasis. For example, if the user's behavior pattern indicates that they tend to discuss product details in detail, the system will push a speech containing an in-depth description of product features. These speech will be displayed in real time through the agent's call interface to guide the agent on how to communicate with the user more effectively.

[0083] S1512: Determine whether the user demand is consistent with the target product. If so, push introduction information related to the target product to the target agent. If not, match the intended product according to the user demand, and push introduction information related to the intended product to the target agent.

[0084] Furthermore, the expert model will evaluate the degree of match between user needs and target products, and match them based on a deep understanding of user needs, including the user's specific problems, the problems they hope to solve, and their preferences for product features. The system analyzes the keywords and phrases expressed by the user in the interaction and compares them with the attributes of the target product. If the user's needs are highly consistent with the target product, the system will confirm this and push relevant information to the agent. When the user's needs are consistent with the target product, the system will push a detailed introduction of the product to the agent, including product advantages, user reviews, FAQs, etc. This information will help the agent introduce the product to the user in a more targeted manner and increase the success rate of sales or services. The pushed information will be customized according to the user's specific needs to ensure the relevance and attractiveness of the information.

[0085] If the user's needs do not fully match the target product, the system will match the most suitable intended product from the product library based on the user's detailed needs. This may involve further detailed analysis of user needs and in-depth exploration of product features. Once a suitable intended product is found, the system will push the product's introduction information to the agent, including how the product meets the user's specific needs and why it is an ideal choice for the user. This helps the agent quickly adjust the sales strategy and provide users with more personalized services.

[0086] This implementation method helps agents communicate with users more efficiently, shorten conversation time, and improve service efficiency by pushing speech strategies that match user behavior patterns in real time. At the same time, it provides personalized product introductions based on accurate judgment of user needs, increases user satisfaction and product matching, and optimizes user experience. It also pushes product information that meets user needs to agents in a timely manner, which helps to improve sales conversion rates and enhance marketing effectiveness. It dynamically adjusts service strategies based on user feedback and changes in demand, making services more flexible and responsive to market changes. Standardized and intelligent speech push reduces the need for agent training, reduces labor costs, and ensures service consistency.

[0087] Please refer to Figure 2 , Figure 2 Schematic diagram of the basic structure of the full-volume speech analysis device based on a large model in this embodiment.

[0088] like Figure 2As shown, a full-volume speech analysis device based on a large model includes: a data acquisition module 1100, used to acquire speech interaction data between an agent and a user; an incremental marking module 1200, used to extract attributes associated with product services in the speech interaction data as incremental marking information according to artificial experience rules and a large language model; a full-volume marking module 1300, used to acquire full-volume marking information after the speech interaction data is normalized; a model training module 1400, used to train the large language model with the speech interaction data, the incremental marking information, and the full-volume marking information as training data to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; a model application module 1500, used to match target agents, target users, and corresponding target products according to the expert model and feed them back to a call system, so as to acquire real-time communication data between the target agent and the target user from the call system, and predict prediction information including user needs and user behavior patterns based on the real-time communication data according to the expert model, so as to push a call strategy matching the prediction information to the target agent.

[0089] The above-mentioned full-volume speech analysis device based on a large model obtains speech interaction data between agents and users; extracts attributes related to product business in the speech data as incremental tag information according to artificial experience rules and a large language model; obtains full-volume tag information after normalization of the speech interaction data; uses the speech interaction data, the incremental tag information, and the full-volume tag information as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; matches target agents, target users, and corresponding target products according to the expert model and feeds back to the call system to obtain the target agent from the call system. The real-time communication data between the agent and the target user is collected, and the expert model is used to predict the prediction information including user needs and user behavior patterns based on the real-time communication data, so as to push the call strategy matching the prediction information to the target agent; through the incremental marking and full analysis of voice interaction data, combined with the deep learning of the large language model and artificial experience rules, the efficient analysis and processing of voice interaction data is realized, the accuracy of the prediction of user needs and behavior patterns is improved, and the service personalization and response speed are enhanced; through the analysis of real-time interaction data, accurate call strategies are provided for the agents, and the customer service experience is optimized; at the same time, a comprehensive user, agent and product portrait is established, which provides data support for business decision-making and improves operational efficiency and market competitiveness.

[0090] Optionally, the incremental marking module is further used for: Extracting attributes associated with a preset prompt word from the voice interaction data; Marking the attributes associated with the prompt word for the first time according to the artificial experience rule; Performing a second labeling on attributes associated with product services in the voice interaction data according to the large language model; Incremental marking information is determined according to the first marking and the second marking.

[0091] Optionally, the model training module is also used to: Testing the expert model using test data to obtain a test result of the expert model, wherein the test result includes a prompt word tag of the test data; The test result is fed back to the prompt word project to update the prompt word associated attributes in the prompt word project according to the prompt word tags of the test data.

[0092] Optionally, the model training module is also used to: Obtain at least one update information of iterative update prompt words, product business, seat feedback, and user behavior pattern; Iteratively update the artificial experience rule according to the update information or make the large language model learn the update information; Incrementally mark the interaction data according to the updated manual experience rules or large language model to obtain updated incremental marking information; The expert model is updated based on the updated incremental labeling information.

[0093] Optionally, the incremental marking module is further used for: Determining an application field of the voice interaction data; Matching personalized analysis rules according to the application field, and extracting user needs and preferences in the voice interaction data based on the personalized analysis rules; The voice interaction data is marked according to the user needs and preferences to update the incremental marking information.

[0094] Optionally, the full amount marking module is further used for: Recognizing the voice interaction data and generating text information of the voice interaction data; Normalizing the text information to obtain normalized text information; Segmenting the normalized text information of the voice interaction data according to the minimum feature units of several classification rules to obtain voice data texts containing several minimum feature units; The minimum feature unit under different classification rules is extracted and the corresponding attribute information is matched, so that the proportion of the attribute information contained in the minimum feature unit of the voice data text under different classification rules reaches a preset value, and the full amount of labeling information after normalization of the voice interaction data is obtained.

[0095] Optionally, the model application module is further used to: According to the user behavior pattern, push the corresponding speech strategy to the target agent; Determine whether the user demand is consistent with the target product. If so, push introduction information related to the target product to the target agent; if not, match the intended product according to the user demand, and push introduction information related to the intended product to the target agent.

[0096] To solve the above technical problems, the present application also provides a computer device. Figure 3 , Figure 3 This is a basic structural block diagram of the computer device in this embodiment.

[0097] like Figure 3 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected via a system bus. Among them, the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence, and the computer-readable instructions are executed by the processor. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor can execute a full-volume speech analysis method based on a large model. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0098] In this embodiment, the processor is used to execute Figure 2The memory stores the program codes and various data required to execute the above modules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program codes and data required to execute all sub-modules in the full-volume speech analysis device based on the large model, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

[0099] The computer device obtains voice interaction data between agents and users; extracts attributes related to product services from the voice interaction data as incremental tag information based on artificial experience rules and a large language model; obtains full tag information after normalization of the voice interaction data; uses the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; matches target agents, target users, and corresponding target products according to the expert model and feeds back to the call system to obtain the target agents and target users from the call system. The real-time communication data of users is collected, and the expert model is used to predict the prediction information including user needs and user behavior patterns based on the real-time communication data, so as to push the call strategy matching the prediction information to the target agent; through the incremental marking and full analysis of voice interaction data, combined with the deep learning of the large language model and the artificial experience rules, the efficient analysis and processing of voice interaction data is realized, the accuracy of the prediction of user needs and behavior patterns is improved, and the personalization and response speed of services are enhanced; through the analysis of real-time interaction data, accurate call strategies are provided for agents, and the customer service experience is optimized; at the same time, a comprehensive user, agent and product portrait is established, which provides data support for business decision-making and improves operational efficiency and market competitiveness.

[0100] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the full-scale speech analysis method based on a large model as described in any of the above embodiments.

[0101] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0102] Those skilled in the art will appreciate that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

[0103] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for full-volume speech analysis based on a large model, characterized in that: include: Obtain voice interaction data between agents and users; Extracting attributes associated with product services from the voice interaction data as incremental tag information based on artificial experience rules and a large language model; Obtaining full amount of label information after normalization of the voice interaction data; Using the voice interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, wherein the expert model establishes association information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; According to the expert model, the target agent, target user and corresponding target product are matched and fed back to the call system, so as to obtain the real-time communication data between the target agent and the target user from the call system, and predict the prediction information including user needs and user behavior patterns based on the real-time communication data according to the expert model, so as to push the call strategy matching the prediction information to the target agent.

2. The method for full-volume speech analysis based on a large model according to claim 1, characterized in that: The extracting, according to artificial experience rules and a large language model, attributes associated with product services in the voice interaction data as incremental tag information includes: Extracting attributes associated with a preset prompt word from the voice interaction data; Marking the attributes associated with the prompt word for the first time according to the artificial experience rule; Performing a second labeling on attributes associated with product services in the voice interaction data according to the large language model; Incremental marking information is determined according to the first marking and the second marking.

3. The method for full-volume speech analysis based on a large model according to claim 2 is characterized in that: After using the speech interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, the method further includes: Testing the expert model using test data to obtain a test result of the expert model, wherein the test result includes a prompt word tag of the test data; The test result is fed back to the prompt word project to update the prompt word associated attributes in the prompt word project according to the prompt word tags of the test data.

4. The method for full-volume speech analysis based on a large model according to claim 1, characterized in that: After using the speech interaction data, the incremental tag information, and the full tag information as training data to train the large language model to generate an expert model, the method further includes: Obtain at least one update information of iterative update prompt words, product business, seat feedback, and user behavior pattern; Iteratively update the artificial experience rule according to the update information or make the large language model learn the update information; Incrementally mark the interaction data according to the updated manual experience rules or large language model to obtain updated incremental marking information; The expert model is updated based on the updated incremental labeling information.

5. The method for full-volume speech analysis based on a large model according to claim 1, characterized in that: The extracting, according to artificial experience rules and a large language model, attributes associated with product services in the voice interaction data as incremental tag information includes: Determining an application field of the voice interaction data; Matching personalized analysis rules according to the application field, and extracting user needs and preferences in the voice interaction data based on the personalized analysis rules; The voice interaction data is marked according to the user needs and preferences to update the incremental marking information.

6. The method for full-volume speech analysis based on a large model according to claim 1, characterized in that: The obtaining of the full amount of label information of the normalized voice interaction data includes: Recognizing the voice interaction data and generating text information of the voice interaction data; Normalizing the text information to obtain normalized text information; Segmenting the normalized text information of the voice interaction data according to the minimum feature units of several classification rules to obtain voice data texts containing several minimum feature units; The minimum feature unit under different classification rules is extracted and the corresponding attribute information is matched, so that the proportion of the attribute information contained in the minimum feature unit of the voice data text under different classification rules reaches a preset value, and the full amount of labeling information after normalization of the voice interaction data is obtained.

7. The method for full-volume speech analysis based on a large model according to claim 1, characterized in that: The pushing of the call strategy matching the prediction information to the target agent includes: According to the user behavior pattern, push the corresponding speech strategy to the target agent; Determine whether the user demand is consistent with the target product. If so, push introduction information related to the target product to the target agent; if not, match the intended product according to the user demand, and push introduction information related to the intended product to the target agent.

8. A full-volume speech analysis device based on a large model, characterized in that: include: Data acquisition module, used to obtain voice interaction data between agents and users; An incremental marking module, used to extract attributes associated with product services from the voice interaction data according to manual experience rules and a large language model as incremental marking information; A full-quantity marking module, used to obtain full-quantity marking information of the voice interaction data after normalization; A model training module, used to train the large language model using the voice interaction data, the incremental tag information, and the full tag information as training data to generate an expert model, wherein the expert model establishes associated information of user portraits, agent portraits, and product portraits, as well as prediction information of user needs and user behavior patterns; The model application module is used to match the target agent, target user and corresponding target product according to the expert model and feed back to the call system, so as to obtain the real-time communication data between the target agent and the target user from the call system, and predict the prediction information including user needs and user behavior patterns based on the real-time communication data according to the expert model, so as to push the call strategy matching the prediction information to the target agent.

9. A computer device comprising a memory and a processor, characterized in that: The memory stores computer-readable instructions, which, when executed by the processor, enable the processor to perform the steps of the large-model-based full-volume speech analysis method as described in any one of claims 1 to 7.

10. A storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the large model-based full-scale speech analysis method as described in any one of claims 1 to 7.

Images