Potential user demand identification analysis method and device, equipment and medium

Abstract

The invention discloses an analysis method and device for potential user demand identification based on a large model, equipment and a medium. The method comprises the following steps: translating a user incoming call into target text data; inputting the target text data into the large model through a first prompt template to obtain first structured data; extracting a summarizing intention list from the first structured data; on the basis of the summing intention list, marketing interaction data are obtained through interaction between marketing personnel and the large model; generating a second prompt template according to the marketing interaction data; inputting the first structured data into the large model through a second prompt template to obtain second structured data; extracting an intention list subset from the second structured data; and if the intention list subset can be matched with the marketing scene, identifying potential user requirements according to the intention list subset. According to the method, accurate logic judgment can be performed through zero sample prediction of a large model in combination with text input and training data.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an analysis method, device, equipment and medium for identifying potential user needs based on a large model. Background Art

[0002] In the modern communications industry, operators collect user data through various channels to explore potential package and network needs. Among them, the demands expressed by users who actively call customer service hotlines contain rich information and are more direct and real. Compared with indirect data mining through the network domain (O domain) and billing domain (B domain), users' active calls reflect their current actual concerns and should theoretically have higher priority and operability.

[0003] However, the processing of user-initiated calls in the prior art has the following problems:

[0004] (1) Inefficiency of annotating corpus:

[0005] Existing technologies usually rely on annotating corpora or providing relevant keywords for each specific mining scenario. In the process of annotating corpora, it is necessary to annotate the positive and negative samples corresponding to the scenario. However, for scenarios with small prior probabilities, it is very difficult to randomly select positive samples, resulting in extremely low annotation efficiency. On the other hand, the keyword method is highly subjective, and a single semantic word may not be able to fully express the meaning of the entire sample, especially in the case of negative sentences, which may lead to errors in sample mining. These problems make the deployment of specific scenarios difficult, costly and time-consuming.

[0006] (2) Reliance on expert knowledge:

[0007] Existing technologies often require expert knowledge support so that specific algorithm engineers can help business personnel with scenario deployment. Such a high threshold means that information asymmetry may occur during the communication process, further increasing the manpower and time required for deployment.

[0008] (3) Complexity of scene modification:

[0009] For existing scenarios, once the logic changes, it is necessary to retrain the model or change the regular expression of the keyword. This adjustment method is not only time-consuming but also inefficient. Summary of the invention

[0010] The technical problem to be solved by the present invention is to address the above-mentioned deficiencies in the prior art and propose an analysis method, device, equipment and medium for identifying potential user needs based on a large model. The method can automatically generate prompt templates so that front-line personnel can efficiently deploy and update scenarios, thereby efficiently and accurately identifying potential user needs.

[0011] In a first aspect, the present invention provides an analysis method for identifying potential user needs based on a large model, the method comprising the following steps:

[0012] Step S1: Translate the user's incoming speech into target text data;

[0013] Step S2: inputting the target text data into a large model to obtain first structured data; and extracting a first overall intent from the first structured data;

[0014] Step S3: Based on the first general intention, obtain marketing interaction data through the interaction between the marketing personnel and the big model; and generate a second prompt template according to the marketing interaction data;

[0015] Step S4: inputting the first structured data into the large model through a second prompt template to obtain second structured data; and extracting a second overall intent from the second structured data;

[0016] Step S5: Determine whether the second general intention can match the marketing scenario:

[0017] If the second general intent cannot match the marketing scenario, repeat steps S3 to S5 until the second general intent can match the marketing scenario;

[0018] Step S6: Identify potential user needs based on the second general intention.

[0019] Furthermore, in step S2, the target text data is input into the large model to obtain the first structured data, which specifically includes the following steps:

[0020] Design the first prompt template;

[0021] Inserting the target text data into the first prompt template to obtain the target text data with the template;

[0022] The target text data with template is input into the large model to obtain the first structured data outputted in the large model.

[0023] Further, the first general intention is stored in a list form;

[0024] Based on the first general intention, marketing interaction data is obtained through the interaction between the marketing personnel and the big model, which specifically includes the following steps:

[0025] Design targeted questions based on the first general intention;

[0026] Based on the targeted issues, marketing personnel interact with the big model to obtain marketing interaction data;

[0027] Wherein, the marketing interaction data includes video data, and / or voice data, and / or text data.

[0028] Further, the second general intention is stored in a list form;

[0029] Determining whether the second general intention can match the marketing scenario specifically includes the following steps:

[0030] Determine a sub-intention according to the second general intention; the sub-intention is N, where N is a natural number greater than or equal to 1; and determine a marketing goal according to the marketing scenario;

[0031] Evaluate the relevance of the sub-intent to the marketing goal:

[0032] If the correlation between any of the sub-intentions and the marketing goal is greater than a preset threshold, it is determined that the second overall intention can match the marketing scenario; otherwise, it is determined that the second overall intention cannot match the marketing scenario.

[0033] In a second aspect, the present invention provides an analysis device for identifying potential user needs based on a large model, the device comprising:

[0034] A translation unit, used for translating user's incoming speech into target text data;

[0035] A first input unit, connected to the translation unit, for inputting the target text data into the large model to obtain first structured data;

[0036] a first extraction unit, connected to the first input unit, and configured to extract a first general intent from the first structured data;

[0037] an acquisition unit, connected to the first extraction unit, and configured to acquire marketing interaction data based on the first general intent through interaction between the marketing personnel and the big model;

[0038] a generating unit, connected to the acquiring unit, and configured to generate a second prompt template according to the marketing interaction data;

[0039] a second input unit connected to the generating unit, and configured to input the first structured data into the large model through a second prompt template to obtain second structured data;

[0040] a second extraction unit, connected to the second input unit, and configured to extract a second general intent from the second structured data;

[0041] a determination unit, connected to the second extraction unit, configured to determine whether the second general intention can match a marketing scenario;

[0042] The acquisition unit is connected to the determination unit and is used to re-acquire the marketing interaction data when the determination unit determines that the second general intention cannot match the marketing scenario;

[0043] An identification unit is connected to the determination unit and is used to identify potential user needs according to the second general intention when the determination unit determines that the second general intention can match the marketing scenario.

[0044] Furthermore, the first input unit includes:

[0045] A first design module, used for designing a first prompt template;

[0046] An inserting module, connected to the first designing module, and used for inserting the target text data into the first prompting template to obtain the target text data with the template;

[0047] An input module is connected to the insertion module and is used to input the target text data with template into the large model to obtain the first structured data output in the large model.

[0048] Furthermore, the acquisition unit includes:

[0049] A second design module is used to design targeted questions according to the first general intention;

[0050] The first general intention is stored in a list form;

[0051] An interaction module, connected to the second design module, for obtaining marketing interaction data by having marketing personnel interact with the big model based on the targeted problems;

[0052] Wherein, the marketing interaction data includes video data, and / or voice data, and / or text data.

[0053] Furthermore, the determination unit includes:

[0054] a determination module, configured to determine, according to the second general intent, a sub-intent, wherein the sub-intent is N, and N is a natural number greater than or equal to 1; and to determine a marketing target according to the marketing scenario;

[0055] The second general intention is stored in a list form;

[0056] An evaluation module, connected to the determination module, for evaluating the degree of relevance between the sub-intention and the marketing goal;

[0057] A determination module, connected to the evaluation module, is used to determine that the second overall intention can match the marketing scenario when the evaluation result of the evaluation module shows that the correlation between the sub-intention and the marketing target is greater than a preset threshold; and is also used to determine that the second overall intention cannot match the marketing scenario when the evaluation result of the evaluation module shows that the correlation between the sub-intention and the marketing target is less than or equal to a preset threshold.

[0058] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the analysis method for identifying potential user needs based on a large model according to the first aspect.

[0059] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the analysis method for identifying potential user needs based on a large model as described in the first aspect is implemented.

[0060] The present invention uses zero-sample prediction of a large model, combines text input and training data to make accurate logical judgments, and automatically generates prompt templates through interactive questions and summaries with users, so that front-line personnel can efficiently deploy and update scenarios, thereby efficiently and accurately identifying potential user needs. The specific beneficial effects are as follows:

[0061] 1. Zero-sample prediction capability:

[0062] The present invention can identify potential user needs without relying on samples, thereby greatly improving efficiency, especially when samples are scarce, and ensuring the flexibility of the analysis process.

[0063] 2. Improve the accuracy of logical judgment:

[0064] The present invention is based on pre-trained data of a large model and can perform precise logical judgments to ensure more accurate understanding and identification of user needs, thereby enhancing the operability of actual business.

[0065] 3. Self-service and efficient deployment:

[0066] The present invention enables front-line personnel to autonomously generate and modify demand prompt templates by interacting with the large model. This design significantly reduces the dependence on professional technicians and makes scenario deployment simple and intuitive.

[0067] 4. Improve dynamic update capabilities:

[0068] The present invention can allow simple questions to modify and update templates through a large model, adapt to changing business needs, achieve rapid response, and thus improve service quality and user satisfaction.

[0069] 5. Simplify the operation process:

[0070] The entire operation process of the present invention does not require complicated technical support, and relevant personnel can focus on demand identification rather than technical implementation, thereby improving work efficiency.

[0071] 6. Enhance user participation:

[0072] The present invention uses interactive questioning to directly feedback the real needs of users, which helps to explore deeper potential needs and enhance the user's sense of participation and satisfaction.

[0073] 7. Improve cost-effectiveness:

[0074] The present invention reduces the need for manual annotation and professional knowledge, thereby achieving high efficiency in resource investment and reducing manpower and time costs.

[0075] 8. Increase scalability and availability:

[0076] The present invention has good scalability and usability, can be quickly adapted and adjusted to different business scenarios, and is suitable for a variety of application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A schematic diagram of an analysis method for identifying potential user needs based on a large model in an embodiment of the present invention;

[0078] Figure 2 The following is an analysis flow chart of identifying potential user needs based on a large model in an embodiment of the present invention;

[0079] Figure 3 A schematic diagram of generating a prompt in an embodiment of the present invention;

[0080] Figure 4 This is a Prompt display diagram in an embodiment of the present invention;

[0081] Figure 5 A schematic diagram of an analysis device for identifying potential user needs based on a large model in an embodiment of the present invention;

[0082] Figure 6 is a system framework diagram in an embodiment of the present invention;

[0083] Among them, the figure marks are: 10, translation unit, 20, first input unit, 30, first extraction unit, 40, acquisition unit, 50, generation unit, 60, second input unit, 70, second extraction unit, 80, determination unit, 90, recognition unit. DETAILED DESCRIPTION

[0084] In order to enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0085] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0086] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.

[0087] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.

[0088] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0089] It can be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0090] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0091] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0092] Operators face multiple challenges when handling a large number of customer service calls and online customer service data:

[0093] (1) Limitations of manual sampling:

[0094] Operators usually rely on manual spot checks of customer service recordings to explore user needs, but this method is not only time-consuming and laborious, but also fails to fully listen to the real voice of users due to the randomness of sampling. Artificially set focus points may lead to the omission of important information. In addition, the professional knowledge requirements of analysts are high, which limits the sustainability of manual analysis.

[0095] (2) High cost of machine learning:

[0096] Existing common machine learning models (such as neural networks) require a large amount of labeled data during the training process. However, these labeling processes rely on professional customer service personnel, which is time-consuming and costly, and it is difficult to maintain continuous analysis capabilities. When processing millions of conversation data, it is almost impossible to rely on manual processing, especially in a high-pressure environment, which is more prone to errors.

[0097] (3) Challenges of intent recognition:

[0098] Although existing machine learning clustering methods can identify and cluster call intent, it is often difficult to continuously and accurately output classification standards that are understandable to the business. Many classification recognition or neural network learning methods (such as patent CN107679234B) require setting user intent in advance and rely on professionals to perform a large amount of annotation, which puts huge pressure on human resources, thereby affecting the efficiency and stability of the analysis process.

[0099] (4) Limitations of generative models:

[0100] In patent CN 116775870 A, although the preset answer module can solve some user questions, the remaining questions still need to rely on the generative large model to answer. Since the answers of the model are random, it cannot be guaranteed that each output meets the business requirements, so there is a large uncertainty in the intention recognition and response process.

[0101] (5) High cost of annotation and keyword mining:

[0102] Existing technologies need to annotate corpora or provide relevant keywords when mining specific scenario requirements. For methods of annotating corpora, it is necessary to annotate scene samples with positive and negative samples, but for scenarios with small prior probabilities, it is difficult to randomly select positive samples, resulting in extremely low annotation efficiency. When using keywords, the provided keywords are often highly subjective, and a single semantic word may lose the semantics of the overall sample, especially in the case of negative sentences, which will cause the sample mining to be completely wrong. These problems make deployment in specific scenarios difficult, costly, and time-consuming.

[0103] (6) Reliance on expert knowledge:

[0104] Many existing technologies require expert knowledge, especially specific algorithm engineers to assist business personnel in scenario deployment. The deployment threshold is high, and information asymmetry during communication may lead to higher manpower and time consumption.

[0105] (7) Difficulty in scene modification:

[0106] For existing scenarios, once the logic changes, it is necessary to retrain the model or change the keyword regularization, and these changes are often time-consuming and labor-intensive.

[0107] In summary, the rich demands contained in the active calls made by users to customer service reflect their directness and authenticity. However, the existing user demand identification methods and related analysis methods fail to make full use of the calls made by users. Based on the above problems, the inventors have proposed a method, device, equipment and medium for identifying and analyzing potential user needs based on a large model to better capture and analyze user demands.

[0108] Embodiment 1:

[0109] like Figure 1 and Figure 2 As shown, this embodiment provides an analysis method for identifying potential user needs based on a large model, and the method includes the following steps:

[0110] Step S1: Translate the user's incoming speech into target text data.

[0111] Specifically, the process of translating user calls into target text data includes multiple steps. First, collect user calls, and then convert them into an editable text format. Then, preprocess the acquired text to remove irrelevant information and noise to ensure the purity of the data. Next, use natural language processing technology to analyze the text, deeply understand the user's intentions and emotions, and perform keyword extraction or semantic classification as needed to convert user needs into structured target text data. The processed data is output in a unified format (such as JSON or CSV) to facilitate subsequent data analysis and application. This process effectively improves the ability to understand and respond to user needs.

[0112] Step S2: input the target text data into a large model to obtain first structured data; and extract a first general intent (summary intent list) from the first structured data.

[0113] In this embodiment, the target text data is input into the macro model through a first prompt template to obtain the first structured data, wherein the first prompt template is a pre-designed prompt template.

[0114] The design requirements for the first prompt template include: clear task instructions to ensure that the model understands the output target; format user input and use placeholders to insert actual text; define the output structure, clearly specify the fields that should be included and the expected format; provide example guidance to help the model generate expected results through specific input-output examples; add contextual information to enhance the model's understanding of the input; and ensure that the template is extensible to adapt to different application scenarios. These requirements ensure that the prompt template can effectively guide large models to generate high-quality structured data.

[0115] As a specific implementation method, the target text data is input into the large model through the first prompt template to obtain the first structured data, which specifically includes the following steps:

[0116] Design the first prompt template;

[0117] Inserting the target text data into the first prompt template to obtain the target text data with the template;

[0118] The target text data with template is input into the large model to obtain the first structured data outputted in the large model.

[0119] The process of extracting the first general intent (summary intent list) from the first structured data is as follows:

[0120] (1) Data analysis:

[0121] First parse the first structured data to ensure that all information has been organized into a readable format, such as JSON, CSV, etc.

[0122] (2) Identify intent:

[0123] Leveraging natural language processing techniques, analyze the text in each data item to identify potential user intent. This may involve techniques such as keyword extraction, topic modeling, or sentiment analysis.

[0124] (3) Classification and sorting:

[0125] The identified intents are classified into a list of intents. Each intent can be marked as a different category for subsequent analysis and use.

[0126] (4) Deduplication and optimization:

[0127] De-duplicate the extracted intents to ensure that there are no duplicates in the list. If necessary, similar intents can be merged to simplify the list.

[0128] (5) Output results:

[0129] The finally generated first total intent (summary intent list) can be output in the form of a table or a list, including a description of each intent and a corresponding frequency or importance score.

[0130] The first prompt template is a pre-designed framework for guiding interactions with users and helping to clearly capture user needs and feedback. The template contains specific keywords and question forms to ensure that key information can be efficiently extracted and structured output can be provided when communicating with users. By adopting this carefully designed prompt template, the accuracy and actionability of user feedback can be significantly improved, providing a solid foundation for further data analysis and decision-making.

[0131] Step S3: Based on the first general intent (summary intent list), obtain marketing interaction data through the interaction between the marketing personnel and the big model; and generate a second prompt template according to the marketing interaction data.

[0132] As a specific implementation method, based on the first general intent (summary intent list), marketing interaction data is obtained through the interaction between the marketing personnel and the big model, which specifically includes the following steps:

[0133] Design targeted questions based on the first general intention (summary intention list);

[0134] Based on the targeted issues, marketing personnel interact with the big model to obtain marketing interaction data;

[0135] Wherein, the marketing interaction data includes video data, and / or voice data, and / or text data.

[0136] Step S4: input the first structured data into the large model through a second prompt template to obtain second structured data; and extract a second total intent (a subset of the intent list) from the second structured data.

[0137] The first general intent and the second general intent are both stored in the form of lists. The first general intent can also be called a summary intent list, and the second general intent can also be called an intent list subset.

[0138] Step S5: Determine whether the second total intent (intent list subset) can match the marketing scenario:

[0139] If the second total intent (subset of the intent list) cannot match the marketing scenario, repeat steps S3 to S5 until the second total intent (subset of the intent list) can match the marketing scenario.

[0140] As a specific implementation, determining whether the second total intent (intent list subset) can match a marketing scenario specifically includes the following steps:

[0141] According to the second total intent (subset of the intent list), determine a sub-intent; the sub-intent is N, where N is a natural number greater than or equal to 1; and determine a marketing goal according to the marketing scenario;

[0142] Evaluate how relevant the sub-intent is to your marketing goals:

[0143] If the correlation between the sub-intention and the marketing goal is greater than a preset threshold, it is determined that the second total intention (subset of the intention list) can match the marketing scenario, otherwise it is determined that the second total intention (subset of the intention list) cannot match the marketing scenario.

[0144] Step S6: Identify potential user needs based on the second overall intent (intent list subset).

[0145] As a specific implementation method, Figure 3 As shown, the process of generating a prompt template (or generating a Prompt) is as follows:

[0146] Q: Please describe the scenarios that need to be explored, including user needs, product descriptions that can be pushed, etc.

[0147] Answer: If the user is dependent on traffic and needs to use more traffic to meet his daily mobile phone usage needs, a traffic package will be pushed to the user, which includes short-term supplementary traffic.

[0148] Q: Keywords or possible conversation sentences that the user may mention in the call

[0149] Answer: If the traffic is not enough, you need to apply for a traffic package or more traffic packages.

[0150] Q: If the customer service has handled the issue, how will the user be informed?

[0151] Answer: Customer service will ask the user for the SMS verification code they received.

[0152] Based on the above questions, generate a prompt.

[0153] like Figure 4 As shown, the steps prompted by Prompt are as follows:

[0154] 1. Demonstrate the intention to be explored

[0155] The intentions include: requesting to process or order products, inquiring about packages, checking the remaining mobile phone bills, checking the details of mobile phone packages, checking the remaining mobile phone data, checking the usage of mobile phone packages, canceling data caps, directional data disputes, etc.

[0156] 2. Determine the prompt that matches the scenario

[0157] Please output "Conform to the scenario" if the following conditions are met:

[0158] 2.1. The user indicates that the traffic has been used up or the remaining balance is insufficient.

[0159] 2.2. The user indicates that the flow is restricted.

[0160] 2.3. The user asked whether there were any packages with more data.

[0161] 2.4. The user asks whether it is possible to apply for a data package.

[0162] 2.5. Users can check their remaining traffic.

[0163] 3. Has the prompt been processed?

[0164] Please output "Completed" if the following conditions are met:

[0165] 3.1. The user indicates that no application is required

[0166] 3.2. Users want to downgrade their plans

[0167] 3.3. Users want to reduce their rates

[0168] 3.4. Customer service asks the user for the verification code number they received

[0169] Overall Prompt

[0170] Please analyze the following content through the summary of the conversation between the user and the customer service:

[0171] Please output "Conform to the scenario" if the following conditions are met:

[0172] 1. The user indicates that the data traffic has been used up or the remaining traffic is insufficient.

[0173] 2. The user indicates that the flow is restricted.

[0174] 3. The user asks whether there are packages with more data.

[0175] 4. The user asks whether it is possible to apply for a data package.

[0176] 5. Users check their remaining traffic status.

[0177] Please output "Completed" if the following conditions are met:

[0178] 1. The user indicated that no processing is necessary.

[0179] 2. The user expressed the desire to downgrade the package.

[0180] 3. The user expressed the desire to reduce the tariff.

[0181] 4. Customer service asks the user for the verification code number they received.

[0182] The implementation process of this embodiment is described below with a specific example:

[0183] Step S1: Translate the user's incoming call into target text data

[0184] The user said on the phone: "I would like to know about your data packages, especially whether there are any discounts?"

[0185] Translate this sentence into target text data:

[0186] {

[0187] "User feedback":"I would like to know about your data packages, especially if there are any discounts"

[0188] }

[0189] Step S2: Input the target text data into the large model through the first prompt template

[0190] Use a pre-designed first prompt template, such as: "What packages and related offers do users want to know?" to input the target text data into the big model.

[0191] The first structured data returned by the model might be:

[0192] {

[0193] "Intent":"Ask about data plans and discount information"

[0194] }

[0195] Extract the first total intent (summary intent list) from the first structured data:

[0196] The intent list might look like this: [

[0198] "Ask about data plans and discounts" ]

[0200] Step S3: Based on the first general intent (summary intent list), obtain marketing interaction data through the interaction between marketing personnel and the big model

[0201] The marketer looks at the list of intents and asks the big model, “Can you recommend some offers for people asking about data plans?”

[0202] The marketing interaction data returned by the model may include:

[0203] {

[0204] "Recommended Package":["Package A","Package B","Package C"],

[0205] "User demand analysis": "Users are very concerned about traffic and discounts"

[0206] }

[0207] Generate a second prompt template according to the marketing interaction data:

[0208] For example, the generated second prompt template is: "Recommend suitable packages and discounts for users who ask about data packages."

[0209] Step S4: Input the first structured data into the large model through the second prompt template

[0210] The first structured data is input into the large model again, using the second prompt template.

[0211] The second structured data obtained may be:

[0212] {

[0213] "Recommended Package":["Package A","Package B"],

[0214] "Offer details": "Package A enjoys a 20% discount, and Package B offers 10GB of data"

[0215] }

[0216] Extract the second total intent (intent list subset) from the second structured data:

[0217] The second overall intent (a subset of the intent list) might be: [

[0219] "Follow the discount information of Package A",

[0220] "Pay attention to the free data for Package B" ]

[0222] Step S5: Determine whether the second total intent (intent list subset) can match the marketing scenario

[0223] Assume that the current marketing scenario is a promotional activity targeting traffic users.

[0224] Now check the second total intent (a subset of the intent list):

[0225] ["Pay attention to the discount information of Package A","Pay attention to the free data of Package B"]

[0226] Both can match the current marketing scenarios because they both involve traffic and discounts.

[0227] If the second overall intent (intent list subset) cannot match the marketing scenario, repeat steps S3 to S5.

[0228] Step S6: Identify potential user needs based on the second general intent (intent list subset)

[0229] Finally, potential user needs are identified based on the second overall intent (a subset of the intent list):

[0230] "Users want to know about the benefits of data packages, and are particularly interested in the discounts on package A and the free data on package B."

[0231] Through this process, operators can effectively identify the specific needs of users and provide them with tailored package recommendations, thereby improving user satisfaction and retention. At the same time, this systematic identification process can provide data support for future marketing strategies.

[0232] like Figure 2 As shown, to ensure efficient handling of user calls and matching of relevant marketing scenarios, the entire process is carried out in the following steps:

[0233] 1. User translation of calls

[0234] When a user receives an incoming call, the operator's customer service representative needs to convert the call content into structured information. During this process, the customer service representative will identify the customer's main needs, such as bill inquiries, network failures, package changes, etc.

[0235] 2. Formatting definition description A

[0236] Based on the identified needs, customer service develops "Formatted Definition Description A". This step ensures consistency and clarity of information by standardizing the user's language to the terminology used within the company. For example, if a user asks for package details, Formatted Definition Description A may be:

[0237] Request Type: Query

[0238] Request content: Package information

[0239] Specific needs: Users want to know the current package fees and usage.

[0240] 3. Splice to get the prompt word template A

[0241] Convert the formatting definition A into a "prompt word template A" so that it can be input by the system. Assume that the definition is:

[0242] "Please provide information about {request type} for user {user ID}, involving {request content}."

[0243] The actual template might be formatted like this:

[0244] "Please provide the package information for user 12345."

[0245] 4. Input the large model of prompt word template A

[0246] The prompt word template A is input into the big model, and the model analyzes the user request based on its database and algorithm and generates a response.

[0247] 5. Get data A

[0248] After processing, the big model returns "Data A", which contains detailed information about the user's current package, fee structure, usage, etc.

[0249] 6. Get the summary intention list A from data A

[0250] Extract "summary intent list A" from data A. This list sorts out the key points that users care about, such as:

[0251] Current plan costs

[0252] Remaining data and call duration

[0253] Changeable package options

[0254] 7. Marketers talk to the big model

[0255] Based on current customer information, marketers communicate with the big model and provide "formatted definition description B". This can be about current promotional activities, package adjustment suggestions or customer preference analysis.

[0256] 8. Obtain prompt word template B through the large model

[0257] The big model will parse "Formatting Definition Description B" and generate "Prompt Word Template B" for specific queries. For example:

[0258] "Please recommend suitable packages or value-added services based on customer needs."

[0259] 9. Splice to get the prompt word template C

[0260] Combine "prompt word template B" with "summary intention list A" to form "prompt word template C":

[0261] "{prompt word template B}, user attention intentions include: {summary intention list A}."

[0262] This could become:

[0263] "Please recommend suitable packages or value-added services based on customer needs. User concerns include: package fees, remaining traffic, and package change options."

[0264] 10. Obtain matching intent subsets through large models

[0265] The prompt word template C is input into the big model, and the model generates a "matching intent subset", that is, all marketing or service suggestions related to the customer's intent.

[0266] 11. Filter data A by matching the intent subset to obtain data B

[0267] Filter "Data A" using "matching intent subset" to obtain "Data B", which may include upgrade options for the current package, promotional information, or value-added services suitable for customers.

[0268] 12. Splice to get the formatting definition description D

[0269] Combine "Formatting Definition C", "Data B" and "Prompt Word Template B" to generate "Formatting Definition D". An example may be:

[0270] "Based on the user's needs and concerns, the provided data B includes: {data B}, and the recommended related service is: {prompt word template B}."

[0271] 13. Data B and formatting definition D are both entered into the large model

[0272] "Data B" and "Formatting Definition Instructions D" are input into the big model together, and the system makes a judgment to determine whether the incoming call matches the current marketing scenario and promotion activities.

[0273] 14. Determine whether the incoming call matches the marketing scenario

[0274] If a match occurs: The system confirms that the content of the call matches the current marketing campaign, ends the process, and may prompt customer service for follow-up services or recommendations.

[0275] Otherwise: If there is no match, the system can go back and continue matching other possible intents or service options to ensure that customer needs can be effectively responded to.

[0276] Finally, if the analysis results show that the user meets the conditions, he / she will be added to the candidate set of potential customers. Through the above steps, this implementation plan successfully integrates user interaction and intelligent analysis, thereby improving the accurate identification of marketing scenarios and the efficiency of potential customer mining.

[0277] Figure 6 Shown is an overview of the potential user demand identification and analysis system framework based on a large model. The analysis system framework is divided into three layers: data storage layer, model layer and visualization layer.

[0278] The data storage layer is mainly responsible for the storage and management of data required by the system, and includes two important components: call analysis data storage and prompt word template storage. Call analysis data storage uses MongoDB and Redis. MongoDB is responsible for storing structured and unstructured data such as user voice records and transcribed texts to meet flexible data query requirements; while Redis, as a cache database, quickly provides hot data and real-time analysis results to ensure efficient system response. Prompt word template storage also uses MongoDB to save a variety of predefined prompt word templates so that the model can dynamically select the corresponding template based on the user's contextual needs.

[0279] The model layer is the core of the system, including the voice model and Langchain. The voice model is responsible for processing the user's voice input, identifying and analyzing potential needs, and ensuring a high-accuracy response. In addition, the Langchain framework simplifies the development of language model applications, making the model more flexible and efficient in processing data, thereby improving the user's interactive experience.

[0280] The visualization layer presents the analysis results to users in a friendly way, including three entrances: APP, H5 and WEB (React / Vue). Users can interact at any time on the mobile terminal through the APP to view real-time feedback and data analysis; the H5 interface provides the convenience of cross-device access, allowing users to understand the system functions without downloading the application; and the WEB terminal using React or Vue provides users with a smooth network interaction experience, supporting dynamic data display and real-time feedback. This multi-level framework design ensures the flexibility, efficiency and availability of the system in meeting user demand identification.

[0281] This embodiment uses zero-sample prediction of a large model, combines text input and training data to make accurate logical judgments, and automatically generates prompt templates through interactive questions and summaries with users, so that front-line personnel can efficiently deploy and update scenarios, thereby efficiently and accurately identifying potential user needs, including the following:

[0282] (1) Zero-sample prediction capability:

[0283] This embodiment can identify potential user needs without relying on samples, thereby greatly improving efficiency, especially when samples are scarce, and ensuring the flexibility of the analysis process.

[0284] (2) Improve the accuracy of logical judgment:

[0285] This embodiment is based on the pre-trained data of the large model and can perform accurate logical judgments to ensure more accurate understanding and identification of user needs, thereby enhancing the operability of actual business.

[0286] (3) Self-service and efficient deployment:

[0287] This embodiment enables front-line personnel to independently generate and modify demand prompt templates through interaction with the big model. This design significantly reduces the dependence on professional technicians and makes scenario deployment simple and intuitive.

[0288] (4) Improve dynamic update capabilities:

[0289] This embodiment can allow simple questions to modify and update templates through a large model, adapt to changing business needs, achieve rapid response, and thus improve service quality and user satisfaction.

[0290] (5) Simplify the operation process:

[0291] The entire operation process of this embodiment does not require complicated technical support, and relevant personnel can focus on demand identification rather than technical implementation, thereby improving work efficiency.

[0292] (6) Enhance user participation:

[0293] In this embodiment, through the interactive questioning method, the user's real needs can be directly fed back, which helps to explore deeper potential needs and enhance the user's sense of participation and satisfaction.

[0294] (7) Improve cost-effectiveness:

[0295] This embodiment reduces the need for manual annotation and professional knowledge, thereby achieving high efficiency in resource investment and reducing manpower and time costs.

[0296] (8) Increase scalability and availability:

[0297] This embodiment has good scalability and availability, can quickly adapt and adjust to different business scenarios, and is suitable for a variety of application scenarios.

[0298] Embodiment 2:

[0299] like Figure 5 As shown, this embodiment provides an analysis device for identifying potential user needs based on a large model, and the device includes:

[0300] The translation unit 10 is used to translate the user's incoming speech into target text data;

[0301] A first input unit 20, connected to the translation unit 10, is used to input the target text data into the large model through a first prompt template to obtain first structured data;

[0302] Wherein, the first prompt template is a pre-designed prompt template;

[0303] A first extraction unit 30, connected to the first input unit 20, configured to extract a first general intent (a summary intent list) from the first structured data;

[0304] An acquisition unit 40, connected to the first extraction unit 30, is used to acquire marketing interaction data based on the first general intent (summarized intent list) through the interaction between the marketing personnel and the big model;

[0305] A generating unit 50, connected to the acquiring unit 40, and configured to generate a second prompt template according to the marketing interaction data;

[0306] A second input unit 60, connected to the generating unit 50, for inputting the first structured data into the large model through a second prompt template to obtain second structured data;

[0307] A second extraction unit 70, connected to the second input unit 60, for extracting a second total intent (intent list subset) from the second structured data;

[0308] A determination unit 80, connected to the second extraction unit 70, for determining whether the second total intent (intent list subset) can match a marketing scenario;

[0309] The acquisition unit 40 is connected to the determination unit 80, and is used to re-acquire the marketing interaction data when the determination unit 80 determines that the second total intent (intent list subset) cannot match the marketing scenario;

[0310] The identification unit 90 is connected to the determination unit 80, and is used to identify potential user needs based on the second general intent (subset of the intent list) when the determination unit determines that the second general intent (subset of the intent list) can match the marketing scenario.

[0311] As a specific implementation, the first input unit 20 includes:

[0312] A first design module, used for designing a first prompt template;

[0313] An inserting module, connected to the first designing module, and used for inserting the target text data into the first prompting template to obtain the target text data with the template;

[0314] An input module is connected to the insertion module and is used to input the target text data with template into the large model to obtain the first structured data output in the large model.

[0315] As a specific implementation, the acquisition unit 40 includes:

[0316] A second design module is used to design targeted questions according to the first general intention (summary intention list);

[0317] An interaction module, connected to the second design module, for obtaining marketing interaction data by having marketing personnel interact with the big model based on the targeted problems;

[0318] Wherein, the marketing interaction data includes video data, and / or voice data, and / or text data.

[0319] As a specific implementation, the determination unit 80 includes:

[0320] a determination module, configured to determine a sub-intent according to the second total intent (subset of the intent list); the sub-intent is N, where N is a natural number greater than or equal to 1; and to determine a marketing target according to the marketing scenario;

[0321] An evaluation module, connected to the determination module, for evaluating the degree of relevance between the sub-intention and the marketing goal;

[0322] A determination module, connected to the evaluation module, is used to determine that the second total intent (intent list subset) can match the marketing scenario when the evaluation result of the evaluation module is that the correlation between the sub-intention and the marketing goal is greater than a preset threshold; and is also used to determine that the second total intent (intent list subset) cannot match the marketing scenario when the evaluation result of the evaluation module is that the correlation between the sub-intention and the marketing goal is less than or equal to a preset threshold.

[0323] Embodiment 3:

[0324] This embodiment provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the analysis method for identifying potential user needs based on a large model according to Embodiment 1.

[0325] Embodiment 4:

[0326] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the analysis method for identifying potential user needs based on a large model as described in Example 1 is implemented.

[0327] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. An analytical method for identifying potential user needs based on a large model, characterized in that: The method comprises the following steps: Step S1: Translate the user's incoming call into target text data; Step S2: inputting the target text data into a large model to obtain first structured data; and extracting a first overall intent from the first structured data; Step S3: Based on the first general intention, obtain marketing interaction data through the interaction between the marketing personnel and the big model; and generate a second prompt template according to the marketing interaction data; Step S4: inputting the first structured data into the large model through a second prompt template to obtain second structured data; and extracting a second overall intent from the second structured data; Step S5: Determine whether the second general intention can match the marketing scenario: If the second general intent cannot match the marketing scenario, repeat steps S3 to S5 until the second general intent can match the marketing scenario; Step S6: Identify potential user needs based on the second general intention.

2. The analysis method for identifying potential user needs based on a large model according to claim 1 is characterized in that: In the step S2, the target text data is input into the large model to obtain the first structured data, which specifically includes the following steps: Design the first prompt template; Inserting the target text data into the first prompt template to obtain the target text data with the template; The target text data with template is input into the large model to obtain the first structured data outputted in the large model.

3. The analysis method for identifying potential user needs based on a large model according to claim 1 is characterized in that: The first general intention is stored in a list form; Based on the first general intention, marketing interaction data is obtained through the interaction between the marketing personnel and the big model, which specifically includes the following steps: Design targeted questions based on the first general intention; Based on the targeted issues, marketing personnel interact with the big model to obtain marketing interaction data; Wherein, the marketing interaction data includes video data, and / or voice data, and / or text data.

4. The analysis method for identifying potential user needs based on a large model according to any one of claims 1 to 3, characterized in that: The second general intention is stored in a list form; Determining whether the second general intention can match the marketing scenario specifically includes the following steps: Determine sub-intentions according to the second general intention; The number of sub-intentions is N, where N is a natural number greater than or equal to 1; and determining a marketing target according to the marketing scenario; Evaluate the relevance of the sub-intent to the marketing goal: If the correlation between any of the sub-intentions and the marketing goal is greater than a preset threshold, it is determined that the second overall intention can match the marketing scenario; otherwise, it is determined that the second overall intention cannot match the marketing scenario.

5. An analysis device for identifying potential user needs based on a large model, characterized in that: The device comprises: A translation unit, used for translating user's incoming speech into target text data; A first input unit, connected to the translation unit, for inputting the target text data into the large model to obtain first structured data; a first extraction unit, connected to the first input unit, and configured to extract a first general intent from the first structured data; an acquisition unit, connected to the first extraction unit, and configured to acquire marketing interaction data based on the first general intent through interaction between the marketing personnel and the big model; a generating unit, connected to the acquiring unit, and configured to generate a second prompt template according to the marketing interaction data; a second input unit connected to the generating unit, and configured to input the first structured data into the large model through a second prompt template to obtain second structured data; a second extraction unit, connected to the second input unit, and configured to extract a second general intent from the second structured data; a determination unit, connected to the second extraction unit, configured to determine whether the second general intention can match a marketing scenario; The acquisition unit is connected to the determination unit and is used to re-acquire the marketing interaction data when the determination unit determines that the second general intention cannot match the marketing scenario; An identification unit is connected to the determination unit and is used to identify potential user needs according to the second general intention when the determination unit determines that the second general intention can match the marketing scenario.

6. The analysis device for identifying potential user needs based on a large model according to claim 5 is characterized in that: The first input unit comprises: A first design module, used for designing a first prompt template; An inserting module, connected to the first designing module, and used for inserting the target text data into the first prompting template to obtain the target text data with the template; An input module is connected to the insertion module and is used to input the target text data with template into the large model to obtain the first structured data output in the large model.

7. The analysis device for identifying potential user needs based on a large model according to claim 5 is characterized in that: The acquisition unit comprises: A second design module is used to design targeted questions according to the first general intention; The first general intention is stored in a list form; An interaction module, connected to the second design module, for obtaining marketing interaction data by having marketing personnel interact with the big model based on the targeted problems; Wherein, the marketing interaction data includes video data, and / or voice data, and / or text data.

8. The analysis device for identifying potential user needs based on a large model according to any one of claims 5 to 7, characterized in that: The determination unit comprises: a determination module, configured to determine, according to the second general intent, a sub-intent, wherein the sub-intent is N, and N is a natural number greater than or equal to 1; and to determine a marketing target according to the marketing scenario; The second general intention is stored in a list form; An evaluation module, connected to the determination module, for evaluating the degree of relevance between the sub-intention and the marketing goal; A determination module, connected to the evaluation module, is used to determine that the second overall intention can match the marketing scenario when the evaluation result of the evaluation module shows that the correlation between the sub-intention and the marketing target is greater than a preset threshold; and is also used to determine that the second overall intention cannot match the marketing scenario when the evaluation result of the evaluation module shows that the correlation between the sub-intention and the marketing target is less than or equal to a preset threshold.

9. An electronic device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the processor executes the analysis method for identifying potential user needs based on a large model according to any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the analysis method for identifying potential user needs based on a large model as described in any one of claims 1 to 4.

Images