User clue mining method and device, computer equipment and storage medium

Abstract

The embodiment of the invention belongs to the field of artificial intelligence and big data, and relates to a user clue mining method, which comprises the following steps of: obtaining a user call text, performing data cleaning on the user call text, and inputting a preset tag processing big model to perform text keyword extraction processing to obtain a first user data tag, carrying out data recall processing on the potential user call text through the first user data label to obtain weak label data, taking the weak label data as a clue training data set, training the general model to obtain a clue model, inputting the potential user call text into the clue model to carry out clue search processing, generating a second user data label, and uploading the first user data tag and the second user data tag to a tag information base, and carrying out classification processing on the transaction keyword and the clue keyword to obtain user clue information. According to the invention, a closed-loop system for clue mining is established, so that the clue mining efficiency and the clue amount mined from stock customers can be improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and big data technology, and in particular to user clue mining methods, devices, computer equipment and storage media. Background Art

[0002] In the field of insurance customer service, how to encourage customers to repurchase new categories of insurance, that is, to tap potential new customers from existing customers, is an important research direction. Compared with tapping potential new users externally, the cost of tapping leads for existing customers is lower and the transaction success rate is higher.

[0003] The traditional lead mining process mainly involves collecting keywords from the business side, recalling data based on keywords, manually annotating them, and finally training a lead model to use the lead model for lead mining.

[0004] However, the limitation of this method is that the data labeling is completely dependent on manual work, and it is impossible to obtain clues directly from the call text, and it is impossible to actively discover new clues or keywords. The clue mining efficiency is low, and it is unable to process a large amount of historical call data. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose a user clue mining method, device, computer equipment and storage medium to solve the technical problems that the traditional clue mining process cannot actively discover new clues, the clue mining efficiency is low, and it cannot process a large amount of historical data.

[0006] In order to solve the above technical problems, the present application embodiment provides a user clue mining method, which adopts the following technical solution:

[0007] Acquire user call texts, wherein the user call texts include potential user call texts and transaction user call texts;

[0008] Performing data cleaning on the call text of the transaction user to obtain target text data;

[0009] Input the target text data into a preset label processing model to perform text keyword extraction processing to obtain a first user data label, wherein the first user data label includes a plurality of transaction keywords;

[0010] Performing data recall processing on the potential user's call text using the first user data label to obtain weak label data;

[0011] Using the weakly labeled data as a clue training data set, training the general model to obtain a clue model;

[0012] Inputting the potential user's call text into the clue model for clue search processing to generate a second user data tag, wherein the second user data tag includes a plurality of clue keywords;

[0013] The first user data tag and the second user data tag are uploaded to a tag information library, and the transaction keywords and the clue keywords are classified to obtain user clue information.

[0014] Furthermore, the label processing large model includes an LLM model, a BCE embedding model and a text clustering model. The step of inputting the target text data into the preset label processing large model for text keyword extraction processing to obtain the first user data label specifically includes:

[0015] Execute prompt instruction on the target text data through the LLM model to obtain target text keywords;

[0016] Performing embedding vectorization processing on the target text keywords through the BCE embedding model to obtain a target text vector;

[0017] The target text vector is clustered by the text clustering model to obtain a cluster category, wherein the cluster category includes a plurality of transaction keywords;

[0018] The transaction keywords are subjected to category interpretation processing by the LLM model to generate the first user data label.

[0019] Furthermore, the step of performing data recall processing on the potential user's call text by using the first user data label to obtain weak label data specifically includes:

[0020] Randomly screening all transaction keywords in the first user data tag to obtain target transaction keywords;

[0021] Contextually locate the potential user's call text based on the target transaction keyword to obtain text position information of the target transaction keyword in the potential user's call text;

[0022] Performing context field interception processing on the potential user's call text based on the text position information to obtain text recall data;

[0023] The text recall data is input into the LLM model for data annotation processing to obtain the weak label data.

[0024] Furthermore, the step of performing data labeling processing on the text recall data through the LLM model to obtain the weak label data specifically includes:

[0025] The text recall data is processed by the text clustering model to divide the intent and obtain category samples;

[0026] Perform label classification processing on the category samples through the LLM model to generate target label data;

[0027] Calculating the text confidence of the target label data;

[0028] If the text confidence is greater than or equal to a preset confidence threshold, the target label data is annotated to obtain the weak label data, and the target label data is saved in the label information library;

[0029] If the text confidence is greater than or equal to a preset confidence threshold, the target label data is directly saved to the label information library.

[0030] Furthermore, the step of using the weak label data as a clue training data set to train the general model to obtain the clue model specifically includes:

[0031] Performing text feature extraction on the clue training data set to obtain text label features of the clue training data set;

[0032] estimating model parameters of the clue model through text label features of the clue training data set;

[0033] The model parameters of the clue model are imported into the general model, and the general model is trained using the clue training data set to obtain the clue model.

[0034] Furthermore, the step of uploading the first user data tag and the second user data tag to the tag information library, and classifying the transaction keywords and the clue keywords to obtain user clue information specifically includes:

[0035] Traversing all transaction keywords in the first user data tag and all clue keywords in the second user data tag to obtain a target clue data set;

[0036] The text confidences of all keywords in the target clue data set are obtained, and classification is performed based on the text confidences to obtain the user clue information.

[0037] Furthermore, after the step of uploading the first user data tag and the second user data tag to the tag information library to obtain user clue information, the method further includes:

[0038] The tag information library is subscribed and connected with the monitoring and scheduling platform through a message queue, the monitoring and scheduling platform includes a monitoring sub-platform and a scheduling sub-platform, and the monitoring and scheduling platform interacts with the clue model, the preset tag processing large model and the tag information library through the message queue;

[0039] Performing data monitoring on the user clue information through the monitoring sub-platform to obtain clue monitoring data, wherein the clue monitoring data includes the amount of valid clues, the amount of valid clues of the tags, and the confidence of the tags;

[0040] Based on the clue monitoring data, data scheduling is performed on the tag information library through the routing rules of the scheduling sub-platform to obtain clue scheduling data.

[0041] In order to solve the above technical problems, the embodiment of the present application further provides a user clue mining device, which adopts the following technical solution:

[0042] An acquisition module, used to acquire user call texts, wherein the user call texts include potential user call texts and transaction user call texts;

[0043] A data cleaning module, used for performing data cleaning on the call text of the transaction user to obtain target text data;

[0044] A processing module, used for inputting the target text data into a preset label processing model to perform text keyword extraction processing to obtain a first user data label, wherein the first user data label includes a plurality of transaction keywords;

[0045] A data recall module, configured to perform data recall processing on the potential user's call text using the first user data label to obtain weakly labeled data;

[0046] A training module, used to train a general model using the weak label data as a clue training data set to obtain a pre-trained clue model;

[0047] An input module, used for inputting the potential user's call text into the clue model for clue search processing to generate a second user data tag, wherein the second user data tag includes a plurality of clue keywords;

[0048] The classification module is used to upload the first user data tag and the second user data tag to the tag information library, and classify the transaction keywords and the clue keywords to obtain user clue information.

[0049] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0050] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the user clue mining method when executing the computer-readable instructions.

[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0052] A computer-readable storage medium, characterized in that computer-readable instructions are stored on the computer-readable storage medium, and when the computer-readable instructions are executed by a processor, the steps of the user clue mining method are implemented.

[0053] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0054] The embodiment of the present application obtains user call texts, pre-processes the call texts of transaction users, and then inputs them into a preset label processing large model for text keyword extraction processing to obtain a first user data label, performs data recall processing on the call texts of potential users through the first user data label to obtain weak label data, uses the weak label data as a clue training data set, trains a general model to obtain a clue model, inputs the call texts of potential users into the clue model for clue search processing to generate a second user data label, uploads the first user data label and the second user data label to the label information library, classifies the transaction keywords and clue keywords to obtain user clue information, and realizes a closed loop of a complete clue mining process, fully utilizes the capabilities of the large model, maximizes its advantages in various sub-tasks of clue mining, effectively reduces the labeling cost, improves the efficiency of clue mining, and can also actively discover new clues or keywords, reduce the omission of clues, and increase the number of leads mined from existing customers. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0057] Figure 2A flowchart of an embodiment of a user clue mining method according to the present application;

[0058] Figure 3 is a structural schematic diagram of an embodiment of a user clue mining device according to the present application;

[0059] Figure 4 is a schematic structural diagram of an embodiment of a computer device according to the present application;

[0060] Figure 5 A flowchart of an embodiment of a user clue mining method according to the present application. DETAILED DESCRIPTION

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0062] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0063] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0064] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0065] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0066] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers and desktop computers, etc.

[0067] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0068] It should be noted that the user clue mining method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the user clue mining device is generally set in the server / terminal device.

[0069] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0070] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for mining user clues according to the present application. The method for mining user clues comprises the following steps:

[0071] Step S201, obtaining the user's call text.

[0072] In this embodiment, the user clue mining method can be deployed in a user clue mining platform. The user clue mining platform can be constructed by the server or server cluster. The server or server cluster can be any electronic device with functions such as text recognition, text processing, data transmission, and data storage. The user clue mining method runs on the electronic device (for example, Figure 1The server / terminal device shown in the figure) can receive the above-mentioned user call text through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultrawideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0073] In this embodiment, the user call texts include potential user call texts and transaction user call texts.

[0074] Specifically, the above-mentioned transaction user call texts may be call records of customers who have already purchased the company's products or services, and the above-mentioned transaction user call texts may include text information such as customer satisfaction, product feedback, and repurchase intention. The above-mentioned potential user call texts may be call records covering customers who have not yet purchased but have shown interest, and the above-mentioned potential user call texts may include clues to customers' needs, questions, or purchasing intentions.

[0075] In a possible embodiment, taking the promotion of non-auto insurance products for existing auto insurance customers as an example, the above-mentioned call texts of the users who have completed the transaction may be the call texts of some existing auto insurance customers who have completed the transaction of non-auto insurance, and the above-mentioned call texts of potential users may be the historical call texts of all auto insurance customers (who have not completed the transaction of non-auto insurance). The customer relationship management (CRM) system and call center technology integrated in the insurance company platform can automatically record the call records between customer service personnel and customers or the chat records of users on the auto insurance consultation platform, and convert the sound into text information (i.e., the above-mentioned user call texts).

[0076] Step S202, performing data cleaning on the call text of the transaction user to obtain target text data;

[0077] In this embodiment, the data cleaning process may be text processing such as format standardization, stop word removal, and stem extraction.

[0078] By performing data cleaning on the transaction user call texts, erroneous and inconsistent data such as duplicate records, spelling errors or format errors can be deleted or corrected in the transaction user call texts. This helps to improve the accuracy of the transaction user call texts, better support the decision-making process, and reduce problems caused by erroneous data.

[0079] Step S203: input the target text data into a preset label processing large model for text keyword extraction processing, and input the preset label processing large model for text keyword extraction processing to obtain a first user data label.

[0080] In this embodiment, the above-mentioned preset label processing large model may include an LLM model, a BCE embedding model and a text clustering model.

[0081] Specifically, the above-mentioned LLM model can be a large pre-trained language model, which can include large models such as OpenAI's GPT series, Google's BERT or T5. The above-mentioned LLM model can learn the wide range of uses of language by processing a large amount of text data, and can perform tasks such as text generation, translation, summarization, and keyword extraction. The above-mentioned BCE embedding model can be a text processing model based on transformer structure training for generating numerical representations of text data, so that the machine learning model can process text data. The above-mentioned text clustering model can be a data processing model trained by a clustering algorithm based on a neural network. The above-mentioned clustering method can include clustering algorithms such as K-means and DBSCAN. The above-mentioned text clustering model can cluster text vectors into different categories and group them according to the similarity of the text. The above-mentioned preprocessing can be a series of techniques for cleaning and preparing data, and the above-mentioned preprocessing can include data cleaning processes such as standardized format, removal of stop words, and stem extraction.

[0082] In this embodiment, within the above-mentioned preset label processing large model, the LLM model, the BCE embedding model and the text clustering model can be connected through a message queue to realize data exchange processing. The above-mentioned message queue can be middleware such as rocketmq, redis and IBM MQ.

[0083] In a possible embodiment, Figure 5 As shown, the pre-processed transaction user call text is stored in the server, and keywords are first extracted through the LLM model, and the extracted keywords are used as input data of the BCE embedding model. The above keywords are vectorized through the BCE embedding model to obtain vector data of the keywords, and the vector data of the above keywords are used as input data of the text clustering model. Clustering is performed through the text clustering model to obtain text categories and keywords after clustering (the above transaction keywords), and finally the above text categories are input into the above LLM model again for category interpretation processing to obtain the above first user data label, and finally the first user data label is stored in the label information library.

[0084] Specifically, the first user data tag may be a tag obtained through a series of processing and model analysis based on the call text of the transaction user, the transaction keyword may be a keyword or phrase extracted from the call text of the transaction user, and the first user data tag contains multiple transaction keywords. For example, taking the non-auto insurance product that promotes the existing customers of auto insurance as an example, the first user data tag is the tag of the target customer of the non-auto product, and the transaction keyword is the keyword under the first user data tag. Taking school insurance as an example, the target customer is the customer with school-age children, the first user data tag is student, child, and its transaction keywords are baby / child / child, after school / primary school / junior high school / school, etc.

[0085] Specifically, regarding the above-mentioned specific implementation process of preprocessing the call text of the transaction user, inputting the preset label processing model to perform text keyword extraction processing, and obtaining the first user data label, this application will provide further detailed descriptions of this in subsequent specific embodiments, and will not be elaborated in detail here.

[0086] Step S204: performing data recall processing on the potential user's call text using the first user data label to obtain weak label data.

[0087] In this embodiment, the weakly labeled data can be a reliable labeled data for quickly generating and expanding a training data set. The training data set can be used to train a machine model, and can also be used for actual predictions in iterative optimization of the machine model. The accuracy and robustness of the model can be improved through gradual correction and optimization through a feedback loop.

[0088] Specifically, regarding the specific implementation process of performing data recall processing on potential user call texts through the first user data label to obtain weak label data, this application will provide further detailed descriptions of this in subsequent specific embodiments, and will not be elaborated on here.

[0089] Step S205 , using the weak label data as a clue training data set to train the general model to obtain a clue model.

[0090] In this embodiment, the general model can be a basic deep learning model with wide application capabilities, which can be used for pre-training on large-scale data sets to learn a wide range of data representations and patterns. It is not optimized for specific fields or specific tasks, but maintains sufficient versatility to adapt to a variety of different data and tasks. The general model can be trained in a specific way to become a deep learning model that can efficiently implement specific tasks in specific fields.

[0091] The above clue model can be used to mine user tags that match the target customer group of the product, and can realize clue mining of call texts of a large number of potential users.

[0092] Specifically, regarding the specific implementation process of using weak label data as a clue training data set to train the general model and obtain the clue model, this application will provide further details in subsequent specific embodiments and will not be elaborated on here.

[0093] Step S206: Input the potential user's call text into the clue model for clue search processing to generate a second user data tag.

[0094] In this embodiment, the second user data tag may be a user tag with a target customer group for the product. The second user data tag includes a plurality of clue keywords. The clue keywords may be elements constituting the second user tag. The clue keywords may reflect specific needs, problems or interests of potential customers.

[0095] In a possible embodiment, taking the promotion of non-auto insurance products for existing auto insurance customers as an example, the above-mentioned clue keywords may include demand clue keywords such as "insurance coverage", "long-term investment return", "emergency service needs", "worry", "dissatisfaction", "excitement", and other emotional clue keywords, as well as behavioral characteristic clue keywords such as "ask for details", "compare prices", etc.

[0096] Specifically, the clue search process may be to identify the tag category according to the call text of the potential user.

[0097] In a possible embodiment, taking the promotion of non-auto insurance products for existing auto insurance customers as an example, the clue model can also have a built-in self-attention mechanism. The initial feature vector of the text is extracted from the sequence information of the potential user's call text through the input layer of the clue model, and these initial feature vectors are standardized to obtain standard feature vectors. The standard feature vectors are input into the hidden layer of the clue model. The hidden layer enhances the sensitivity to user needs, user emotions and user behavior characteristics, thereby mapping them to the corresponding latent space. In this latent space, the clue model focuses on extracting clue keywords related to the potential needs of existing auto insurance customers for non-auto insurance products from the standard feature vectors to obtain a clue keyword matrix. The above clue keyword matrix can be used as a second user data label. Finally, the clue model outputs the second user data label of the potential user's call text through the output layer.

[0098] Step S207: Upload the first user data tag and the second user data tag to the tag information library, classify the transaction keywords and the clue keywords, and obtain user clue information.

[0099] In this embodiment, the user clue information may be a new message data set obtained by union processing the transaction keyword and the clue keyword, and the new message data set includes all data information in the first user data tag and the second user data tag.

[0100] Specifically, regarding the above-mentioned specific implementation process of uploading the first user data label and the second user data label to the label information library, classifying and processing the transaction keywords and clue keywords, and obtaining user clue information, this application will provide further detailed descriptions of this in subsequent specific embodiments, and will not be elaborated in detail here.

[0101] The present application obtains user call texts, pre-processes the call texts of transaction users, and then inputs them into a preset label processing large model for text keyword extraction to obtain a first user data label, performs data recall processing on potential user call texts through the first user data label to obtain weak label data, uses the weak label data as a clue training data set, trains a general model to obtain a clue model, inputs the potential user call texts into the clue model for clue search processing, generates a second user data label, uploads the first user data label and the second user data label to a label information library, classifies the transaction keywords and clue keywords, and obtains user clue information, thereby realizing a closed loop of a complete clue mining process, making full use of the capabilities of the large model to maximize its advantages in each sub-task of clue mining, effectively reducing the labeling cost, and improving the efficiency of clue mining. It can also actively discover new clues or keywords, reduce the omission of clues, and increase the amount of leads mined from existing customers.

[0102] In some optional implementations, step S203 includes the following steps:

[0103] Execute prompt instruction on the target text data through LLM model to obtain target text keywords;

[0104] Use the BCE embedding model to embed the target text keywords and obtain the target text vector.

[0105] Clustering the target text vector using a text clustering model to obtain a cluster category, wherein the cluster category includes a plurality of transaction keywords;

[0106] The LLM model is used to interpret the transaction keywords into categories and generate the first user data label.

[0107] In this embodiment, the prompt instruction may refer to an input text built into the LLM model. The purpose of the input text is to guide or instruct the model to give a specific answer or output. The target text keyword may be the transaction keyword.

[0108] Specifically, by proposing an instruction text of "listing all product features mentioned in the target text data and user-expressed needs" as the prompt instruction, the LLM model can be guided to focus on the most important information points in the text and extract the keywords in the call text of the transaction user that are most helpful for understanding the user's intention and transaction behavior.

[0109] In this embodiment, the above-mentioned embedding vectorization processing can be to use a pre-trained word embedding model such as BERT, Word2Vec or GloVe to convert the target text keywords into numerical vectors. The above-mentioned numerical vectors can capture the semantic characteristics of the vocabulary and enable the model to process text data based on mathematical methods.

[0110] In this embodiment, the above-mentioned clustering processing can be to use Faiss mean clustering, K-means, hierarchical clustering, DBSCAN and other clustering algorithms to process the above-mentioned target text vector, obtain the keywords of each clustering category, and finally randomly extract 20 keywords for each category. The above-mentioned clustering categories may include "insurance needs", "product questions", "price sensitivity" and other categories. Each clustering category can show different user concerns based on the transaction keywords contained.

[0111] In this embodiment, the above-mentioned category explanation processing can be performed by defining a prompt task, which is intended to guide LLM to deeply explain the meaning of each category. Specifically, taking the promotion of non-auto insurance products for existing auto insurance customers as an example, if the clustering category contains transaction keywords such as "emergency service needs" and "24-hour response", the above-mentioned prompt may be defined as "explaining the customer needs represented by this group of keywords", and the prompt task can be automatically generated, and the prompt content can be adjusted according to the dynamic changes of the clustering category. Then, the clustering category and the corresponding prompt are combined to form the input data of the LLM model, and one or more user tags (the above-mentioned first user data tag) are output through the LLM model.

[0112] In a possible embodiment, taking automobile insurance business as an example, the above-mentioned data cleaning processes such as standardized format, stop word removal, and stem extraction are performed on the call text of the transaction user, and the prompt instruction of "listing all product features mentioned in the text data and user-expressed needs" is executed through the LLM model, and the extracted transaction keywords are used as input data of the BCE embedding model. The pre-trained word embedding models such as BERT, Word2Vec or GloVe built into the BCE embedding model are used to convert the transaction keywords into vector data, and the above vector data is used as input data of the text clustering model. The text clustering model is used to perform Faiss mean clustering processing to obtain text categories such as "insurance needs", "product questions", and "price sensitivity". Finally, the prompt task is automatically generated, the prompt content is adjusted according to the dynamic changes of the text category, and then the text category and the corresponding prompt are combined to form the input data of the LLM model, and one or more user tags are output through the LLM model.

[0113] This application combines automated data processing with a variety of neural network models for analysis, which can accurately extract and cluster customer call keywords and generate detailed user tags, thereby improving the efficiency of lead mining.

[0114] In some optional implementations, step S204 includes the following steps:

[0115] Randomly screen all transaction keywords in the first user data tag to obtain target transaction keywords;

[0116] Contextually locate the potential user's call text based on the target transaction keyword to obtain the text position information of the target transaction keyword in the potential user's call text;

[0117] Based on the text location information, context field interception processing is performed on the call text of the potential user to obtain text recall data;

[0118] The text recall data is input into the LLM model for data annotation processing to obtain weakly labeled data.

[0119] In this embodiment, the random screening may be a simple random sampling method for all transaction keywords in the first user data tag and a weighted random selection method based on keyword frequency, relevance score, etc.

[0120] In this embodiment, the above-mentioned context positioning can be to identify and locate the position of the selected target transaction keywords in the potential user's call text. Specifically, it can be achieved through natural language processing technologies such as regular expression search, keyword matching algorithm, or more advanced text analysis tools to ensure that the keyword location can be accurately found.

[0121] In this embodiment, the above-mentioned context field interception processing can be centered on the above-mentioned target transaction keyword, and intercepted into 10 context sentences as fragments. The above-mentioned text position information can be the specific word position of the target transaction keyword in the potential user's call text or the sentence or paragraph number where the above-mentioned target transaction keyword is located.

[0122] In this embodiment, the above text recall data can be a relevant text segment intercepted from the potential user's call text according to the context location information. Specifically, regarding the specific implementation process of inputting the text recall data into the LLM model for data annotation processing to obtain weak label data, this application will further describe this in detail in subsequent specific embodiments, and will not be elaborated on here.

[0123] In this embodiment, text location information is obtained through context positioning and interception, data is recalled using the text location information, and the recalled data is then annotated, thereby improving the automation and efficiency of data annotation and being able to quickly discover potential customers' interests and needs in large-scale data.

[0124] In some optional implementations, the step of inputting the text recall data into the LLM model for data annotation processing to obtain weakly labeled data further includes the following steps:

[0125] The text recall data is processed by the text clustering model to divide the intent and obtain category samples;

[0126] Use the LLM model to classify the category samples and generate target label data;

[0127] Calculate the text confidence of the target label data;

[0128] If the text confidence is greater than or equal to the preset confidence threshold, the target label data is annotated to obtain weak label data, and the target label data is saved in the label information library;

[0129] If the text confidence is greater than or equal to the preset confidence threshold, the target label data is directly saved to the label information library.

[0130] In this embodiment, the above-mentioned intent division processing can be to first vectorize the text recall data through the BCE embedding model, and then perform Faiss mean clustering processing through the above-mentioned text clustering model to divide the text recall data of a single sentence into several category samples. The above-mentioned category samples can be labeled as samples. If a category sample is labeled as a positive sample, the sample of this category will be given priority to enter the large model for label classification. The above-mentioned positive sample can be a category sample with clear intent.

[0131] In a possible embodiment, whether a category sample is a positive sample can be evaluated by calculating two clustering quality evaluation indicators, Davies-Bouldin Index (DBI) and Calinski-Harabasz Index (CHI). The above-mentioned DBI can be used to evaluate the internal compactness of the category sample and the separation between the category samples to reflect the quality of the category samples. The similarity of the internal elements of each category sample and the distinction from other category samples can be determined to evaluate whether the category sample is a positive sample. The above-mentioned CHI can measure the quality of the category sample by calculating the ratio of the inter-class discreteness to the intra-class discreteness of the category sample.

[0132] Specifically, the average distance from all points in each category sample to the category sample center can be obtained, the distance between the centers of different category samples can be calculated, the DBI formula can be used to evaluate the relative quality of each category sample and the nearest category sample, and the average value can be taken to determine the overall category sample quality (DBI value of the category sample); at the same time, the inter-class dispersion of each category sample (the sum of the distances from the centers of different category samples to the global center) and the intra-class dispersion (the sum of the distances from the internal points of each category sample to the category sample center) can be calculated, and the CHI formula can be used to obtain the CHI value of the category sample by comparing the ratio of the inter-class dispersion to the intra-class dispersion. Finally, the DBI value and CHI value of each category sample can be collected and compared. When the DBI value of the category sample is less than the preset DBI threshold and the CHI value of the category sample is greater than the preset CHI threshold, the above category sample is marked as a positive sample.

[0133] In this embodiment, the label classification process can be to design a specific prompt to guide the LLM model to generate relevant user labels for each category sample. Specifically, taking the field of vehicle insurance as an example, for the category of "auto insurance product inquiry", the prompt may be "mark which type of auto insurance product the user is interested in", and for the category of "service complaint", the prompt may be "identify the specific problem of the user's complaint".

[0134] In this embodiment, the above-mentioned text confidence can be used to describe the confidence level of the LLM model in its prediction results. Specifically, the text confidence can be a numerical measure predicted by the LLM model, usually expressed as a probability value ranging from 0 to 1, where a higher value indicates a higher degree of confidence. The text confidence of the target label data can be calculated by means of the built-in Softmax layer of the LLM model. The text confidence information can also be obtained by calculating the relative size of the difference between the highest value and the second highest value of the logits before the softmax layer output by the LLM model (the original output before the softmax layer). The above-mentioned Logits can be the output of the last linear layer in the LLM model.

[0135] In a possible embodiment, the logits output by the LLM model can be processed according to the confidence level. If the confidence of the target label data is greater than or equal to the preset confidence threshold, the data is considered to be reliable, and the target label data is marked as weak label data and saved in the label information library. If the text confidence is greater than or equal to the preset confidence threshold, the target label data is directly saved in the label information library.

[0136] This application realizes the automatic intent segmentation and label classification of text recall data by combining the text clustering model and the LLM model, and can screen out high-confidence target label data as weak label data, reducing the reliance on manual annotation and reducing costs.

[0137] In some optional implementations, step S205 includes the following steps:

[0138] Performing text feature extraction on the clue training data set to obtain text label features of the clue training data set;

[0139] estimating model parameters of the cue model through text label features of the cue training dataset;

[0140] The model parameters of the clue model are imported into the general model, and the general model is trained using the clue training data set to obtain the clue model.

[0141] In the present application, the above-mentioned text label features may include word frequency (TF), inverse document frequency (IDF), word embedding (such as Word2Vec, GloVe, FastText), syntactic features (such as dependency syntax tree information), and sentiment analysis scores. Specifically, high-dimensional features in the clue training data set can be extracted through natural language processing toolkits such as NLTK, SpaCy or more advanced pre-trained models such as BERT, and the model parameters of the clue model can be estimated through the text label features of the clue training data set through parameter estimation methods such as gradient descent, regularization and cross-validation.

[0142] In a possible embodiment, the clue training data set can be input into the clue model for processing to obtain the processing result corresponding to the clue training data, and the error loss function between the processing result corresponding to the clue training data and the label data corresponding to the clue training data is calculated. The error loss function is minimized as the optimization goal, and the parameters of the clue model are adjusted through the back propagation algorithm. The parameter adjustment process is iterated until the error loss function converges to the minimum, or the number of iterations reaches a preset value, and the training is stopped to obtain a trained clue model.

[0143] In some optional implementations, step S207 includes the following steps:

[0144] Traverse all transaction keywords in the first user data tag and all clue keywords in the second user data tag to obtain a target clue data set;

[0145] The text confidence of all keywords in the target clue data set is obtained, and classification is performed based on the text confidence to obtain the user clue information.

[0146] In this embodiment, the target clue data set may be a new message data set obtained by union processing the transaction keyword and the clue keyword, and the target clue data set includes all data information in the first user data tag and the second user data tag.

[0147] In a possible embodiment, the user clue information includes efficient user clue information and backup user clue information. Specifically, the text confidence of all keywords in the target clue data set can be calculated, the keywords can be sorted in order from low to high according to the text confidence, and the median of the text confidence can be taken. If the text confidence of the keyword is greater than or equal to the median of the text confidence, it will be classified as efficient user clue information, and if the text confidence of the keyword is less than the median of the text confidence, it will be classified as backup user clue information.

[0148] This application classifies keywords based on the text confidence of each keyword in the data set, thereby accurately acquiring and organizing user clue information, extracting key information closely related to user behavior and intention, and improving the accuracy and value of user clue information.

[0149] In some optional implementations, the following steps are further included after step S207:

[0150] Subscribe and connect the tag information library with the monitoring and scheduling platform through the message queue;

[0151] Performing data monitoring on the user clue information through the monitoring sub-platform to obtain clue monitoring data;

[0152] Based on the clue monitoring data, the tag information library is scheduled through the routing rules of the scheduling sub-platform to obtain the clue scheduling data.

[0153] In this embodiment, the above-mentioned monitoring and scheduling platform includes a monitoring sub-platform and a scheduling sub-platform. The above-mentioned monitoring sub-platform and scheduling sub-platform can build a back-end platform through Django and a front-end web page through Vue3.0 to visualize the clue monitoring data. The above-mentioned visualization processing can be to display the clue monitoring data in the form of bar charts, line charts and pie charts on the front-end web page.

[0154] In this embodiment, the clue monitoring data includes the effective clue quantity, the effective clue quantity of the tag, and the tag confidence. Specifically, the effective clue quantity can be the number of transaction keywords contained in the current first user data tag and the number of clue keywords contained in the current second user data tag; the effective clue quantity of the tag can be the number of weakly labeled data in the first user data tag; the tag confidence can be the sum of the text confidences of all transaction keywords in the first user data tag and the sum of the text confidences of all clue keywords in the second user data tag.

[0155] In this embodiment, the above-mentioned subscription connection can be achieved by binding the data input and output interface of the label information library with the data input and output interface of the monitoring and scheduling platform and the message queue through middleware such as rocketmq, redis and IBM MQ to realize data exchange. The above-mentioned monitoring and scheduling platform can also interact with the clue model, the preset label processing large model and the label information library through the message queue.

[0156] In this embodiment, the scheduling sub-platform can decide when to obtain data, call the tag processing model and the clue model according to the following routing rules, so as to efficiently use data and improve the output efficiency of clue samples. The main routing rules include:

[0157] Rule 1: give priority to selecting the first user data tag with the most valid clues;

[0158] Rule 2: Prioritize the first user data label with high label confidence;

[0159] Rule 3: If the number of valid clues of the first user data label reaches N, switch to sample mining of another label;

[0160] Rule 4: If the output ratio of the second user data label of the current potential user's call text is lower than the threshold P, switch to another potential user's call text for mining;

[0161] Among them, rule1, rule2 and rule3 can be routing rules for calling the tag processing large model, and rule4 can be a routing rule for calling the clue model.

[0162] This application implements the subscription connection between the tag information library and the monitoring and scheduling platform through a message queue, and builds an efficient data interaction mechanism. The monitoring sub-platform in the monitoring and scheduling platform can monitor user clue information in real time, providing accurate data support for subsequent clue mining methods. At the same time, the scheduling sub-platform, based on these monitoring data, intelligently schedules the tag information library through preset routing rules and generates clue scheduling data, thereby realizing the optimal configuration and efficient utilization of clue resources.

[0163] It should be emphasized that in order to further ensure the privacy and security of the above user clue information, the above user clue information can also be stored in a blockchain node.

[0164] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0165] 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 computer-readable instructions, and the computer-readable instructions 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).

[0166] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0167] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a user clue mining device, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0168] like Figure 3 As shown, the user clue mining device 300 described in this embodiment includes: an acquisition module 301, a data cleaning module 302, a processing module 303, a data recall module 304, a training module 305, an input module 306 and a classification module 307. Among them:

[0169] An acquisition module 301 is used to acquire user call texts, wherein the user call texts include potential user call texts and transaction user call texts;

[0170] A data cleaning module 302 is used to perform data cleaning on the transaction user's call text to obtain target text data;

[0171] The processing module 303 is used to input the target text data into a preset tag processing model to perform text keyword extraction processing to obtain a first user data tag, wherein the first user data tag includes a plurality of transaction keywords;

[0172] A data recall module 304 is used to perform data recall processing on the potential user's call text using the first user data label to obtain weak label data;

[0173] A training module 305 is used to train a general model using the weak label data as a clue training data set to obtain a pre-trained clue model;

[0174] An input module 306 is used to input the potential user's call text into the clue model for clue search processing to generate a second user data tag, wherein the second user data tag includes a plurality of clue keywords;

[0175] The classification module 307 is used to upload the first user data tag and the second user data tag to the tag information library, and classify the transaction keywords and the clue keywords to obtain user clue information.

[0176] Wherein, the processing module 303 includes:

[0177] An execution submodule, used for executing a prompt instruction on the target text data through the LLM model to obtain target text keywords;

[0178] A vectorization submodule, used for performing embedding vectorization processing on the target text keywords through the BCE embedding model to obtain a target text vector;

[0179] A clustering submodule, used for clustering the target text vector by using the text clustering model to obtain a cluster category, wherein the cluster category includes a number of transaction keywords;

[0180] The category interpretation submodule is used to perform category interpretation processing on the transaction keyword through the LLM model to generate the first user data label.

[0181] The data recall module 304 includes:

[0182] A random screening submodule, used to randomly screen all transaction keywords in the first user data tag to obtain target transaction keywords;

[0183] A context positioning submodule, configured to perform context positioning on the potential user's call text based on the target transaction keyword, and obtain text position information of the target transaction keyword in the potential user's call text;

[0184] A field interception submodule, configured to perform context field interception processing on the potential user's call text based on the text position information to obtain text recall data;

[0185] The data annotation submodule is used to input the text recall data into the LLM model for data annotation processing to obtain the weak label data.

[0186] Among them, the above data annotation submodule includes:

[0187] An intention division unit, used for performing intention division processing on the text recall data through the text clustering model to obtain category samples;

[0188] A label classification unit, used to perform label classification processing on the category samples through the LLM model to generate target label data;

[0189] A calculation unit, used to calculate the text confidence of the target label data;

[0190] A first judgment unit is used to mark the target label data to obtain the weak label data if the text confidence is greater than or equal to a preset confidence threshold, and save the target label data to the label information library;

[0191] The second judgment unit is used to directly save the target label data to the label information library if the text confidence is greater than or equal to a preset confidence threshold.

[0192] Wherein, the training module 305 includes:

[0193] An extraction submodule, used for performing text feature extraction on the clue training data set to obtain text label features of the clue training data set;

[0194] An estimation submodule, used for estimating model parameters of the clue model through text label features of the clue training data set;

[0195] The import submodule is used to import the model parameters of the clue model into the general model, and train the general model using the clue training data set to obtain the clue model.

[0196] Wherein, the classification module 307 includes:

[0197] A traversal submodule, used to traverse all transaction keywords in the first user data tag and all clue keywords in the second user data tag to obtain a target clue data set;

[0198] The classification submodule obtains the text confidence of all keywords in the target clue data set, and classifies based on the text confidence to obtain the user clue information.

[0199] Among them, after the classification module 307, it also includes:

[0200] A connection module, used to subscribe and connect the tag information library with the monitoring and scheduling platform through a message queue, wherein the monitoring and scheduling platform includes a monitoring sub-platform and a scheduling sub-platform, and the monitoring and scheduling platform exchanges data with the clue model, the preset tag processing large model and the tag information library through the message queue;

[0201] A monitoring module, used to perform data monitoring on the user clue information through the monitoring sub-platform to obtain clue monitoring data, wherein the clue monitoring data includes the amount of valid clues, the amount of valid clues of tags, and the confidence of tags;

[0202] The scheduling module is used to schedule data on the tag information library based on the clue monitoring data and through the routing rules of the scheduling sub-platform to obtain clue scheduling data.

[0203] In this embodiment, by obtaining user call texts, the transaction user call texts are preprocessed and then input into a preset label processing large model for text keyword extraction processing to obtain a first user data label, and the potential user call texts are processed for data recall through the first user data label to obtain weak label data, and the weak label data is used as a clue training data set to train the general model to obtain a clue model, and the potential user call texts are input into the clue model for clue search processing to generate a second user data label, and the first user data label and the second user data label are uploaded to the label information library, and the transaction keywords and clue keywords are classified to obtain user clue information, thereby realizing a closed loop of a complete clue mining process, making full use of the capabilities of the large model to maximize its advantages in each subtask of clue mining, effectively reducing the labeling cost, and improving the efficiency of clue mining. It can also actively discover new clues or keywords, reduce the omission of clues, and increase the amount of clues mined from existing customers.

[0204] In this embodiment, the operations performed by the above units or modules correspond one-to-one to the steps of the user clue mining method in the above implementation mode, and will not be repeated here.

[0205] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0206] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0207] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0208] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the user clue mining method, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0209] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the user clue mining method.

[0210] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0211] The present embodiment provides a computer device, which obtains user call texts, pre-processes the call texts of transaction users, and then inputs them into a preset label processing large model for text keyword extraction processing to obtain a first user data label, performs data recall processing on potential user call texts through the first user data label to obtain weak label data, uses the weak label data as a clue training data set, trains a general model to obtain a clue model, inputs the potential user call texts into the clue model for clue search processing to generate a second user data label, uploads the first user data label and the second user data label to a label information library, classifies the transaction keywords and clue keywords to obtain user clue information, and realizes a closed loop of a complete clue mining process, fully utilizes the capabilities of the large model, maximizes its advantages in various sub-tasks of clue mining, effectively reduces the labeling cost, improves the efficiency of clue mining, and can also actively discover new clues or keywords, reduce the omission of clues, and increase the number of clues mined from existing customers.

[0212] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the user clue mining method as described above.

[0213] The present embodiment provides a computer-readable storage medium, which obtains user call texts, pre-processes the call texts of transaction users, and then inputs them into a preset label processing large model for text keyword extraction processing to obtain a first user data label, performs data recall processing on the call texts of potential users through the first user data label to obtain weak label data, uses the weak label data as a clue training data set, trains a general model to obtain a clue model, inputs the call texts of potential users into the clue model for clue search processing to generate a second user data label, uploads the first user data label and the second user data label to a label information library, classifies the transaction keywords and clue keywords to obtain user clue information, and realizes a closed loop of a complete clue mining process, fully utilizes the capabilities of the large model, maximizes its advantages in each sub-task of clue mining, effectively reduces the labeling cost, improves the efficiency of clue mining, and can also actively discover new clues or keywords, reduce the omission of clues, and increase the number of clues mined from existing customers.

[0214] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0215] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A user clue mining method, characterized in that: The steps include: Acquire user call texts, wherein the user call texts include potential user call texts and transaction user call texts; Performing data cleaning on the call text of the transaction user to obtain target text data; Input the target text data into a preset label processing model to perform text keyword extraction processing to obtain a first user data label, wherein the first user data label includes a plurality of transaction keywords; Performing data recall processing on the potential user's call text using the first user data label to obtain weak label data; Using the weakly labeled data as a clue training data set, training the general model to obtain a clue model; Inputting the potential user's call text into the clue model for clue search processing to generate a second user data tag, wherein the second user data tag includes a plurality of clue keywords; The first user data tag and the second user data tag are uploaded to a tag information library, and the transaction keywords and the clue keywords are classified to obtain user clue information.

2. The user clue mining method according to claim 1, characterized in that: The label processing model includes an LLM model, a BCE embedding model and a text clustering model. The step of inputting the target text data into the preset label processing model for text keyword extraction processing to obtain the first user data label specifically includes: Execute prompt instruction on the target text data through the LLM model to obtain target text keywords; Performing embedding vectorization processing on the target text keywords through the BCE embedding model to obtain a target text vector; The target text vector is clustered by the text clustering model to obtain a cluster category, wherein the cluster category includes a plurality of transaction keywords; The transaction keywords are subjected to category interpretation processing by the LLM model to generate the first user data label.

3. The user clue mining method according to claim 2, characterized in that: The step of performing data recall processing on the potential user's call text by using the first user data label to obtain weak label data specifically includes: Randomly screening all transaction keywords in the first user data tag to obtain target transaction keywords; Contextually locate the potential user's call text based on the target transaction keyword to obtain text position information of the target transaction keyword in the potential user's call text; Performing context field interception processing on the potential user's call text based on the text position information to obtain text recall data; The text recall data is input into the LLM model for data annotation processing to obtain the weak label data.

4. The user clue mining method according to claim 3, characterized in that: The step of performing data labeling processing on the text recall data through the LLM model to obtain the weak label data specifically includes: The text recall data is processed by the text clustering model to divide the intent and obtain category samples; Perform label classification processing on the category samples through the LLM model to generate target label data; Calculating the text confidence of the target label data; If the text confidence is greater than or equal to a preset confidence threshold, the target label data is annotated to obtain the weak label data, and the target label data is saved in the label information library; If the text confidence is greater than or equal to a preset confidence threshold, the target label data is directly saved to the label information library.

5. The user clue mining method according to claim 1, characterized in that: The step of using the weak label data as a clue training data set to train the general model to obtain the clue model specifically includes: Performing text feature extraction on the clue training data set to obtain text label features of the clue training data set; estimating model parameters of the clue model through text label features of the clue training data set; The model parameters of the clue model are imported into the general model, and the general model is trained using the clue training data set to obtain the clue model.

6. The user clue mining method according to claim 1, characterized in that: The step of uploading the first user data tag and the second user data tag to the tag information library, and classifying the transaction keywords and the clue keywords to obtain user clue information specifically includes: Traversing all transaction keywords in the first user data tag and all clue keywords in the second user data tag to obtain a target clue data set; The text confidences of all keywords in the target clue data set are obtained, and classification is performed based on the text confidences to obtain the user clue information.

7. The user clue mining method according to claim 1, characterized in that: After the step of uploading the first user data tag and the second user data tag to the tag information library to obtain user clue information, the method further includes: The tag information library is subscribed and connected with the monitoring and scheduling platform through a message queue, the monitoring and scheduling platform includes a monitoring sub-platform and a scheduling sub-platform, and the monitoring and scheduling platform interacts with the clue model, the preset tag processing large model and the tag information library through the message queue; Performing data monitoring on the user clue information through the monitoring sub-platform to obtain clue monitoring data, wherein the clue monitoring data includes the amount of valid clues, the amount of valid clues of the tags, and the confidence of the tags; Based on the clue monitoring data, data scheduling is performed on the tag information library through the routing rules of the scheduling sub-platform to obtain clue scheduling data.

8. A user clue mining device, characterized in that: include: An acquisition module, used to acquire user call texts, wherein the user call texts include potential user call texts and transaction user call texts; A data cleaning module, used for performing data cleaning on the call text of the transaction user to obtain target text data; A processing module, used for inputting the target text data into a preset label processing model to perform text keyword extraction processing to obtain a first user data label, wherein the first user data label includes a plurality of transaction keywords; A data recall module, configured to perform data recall processing on the potential user's call text using the first user data label to obtain weakly labeled data; A training module, used to train a general model using the weak label data as a clue training data set to obtain a pre-trained clue model; An input module, used for inputting the potential user's call text into the clue model for clue search processing to generate a second user data tag, wherein the second user data tag includes a plurality of clue keywords; The classification module is used to upload the first user data tag and the second user data tag to the tag information library, and classify the transaction keywords and the clue keywords to obtain user clue information.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the user clue mining method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the user clue mining method according to any one of claims 1 to 7 are implemented.

Images