Intelligent friend adding method and device based on large model, equipment and medium

Abstract

The invention relates to the field of artificial intelligence, can be applied to business system platforms of financial science and technology, medical health and the like, and discloses an intelligent friend adding method and device based on a large model, equipment and a medium. Performing feature engineering processing on the basic information and the behavior information according to the target business scene, and extracting key features in the target business scene; inputting the key features into the large model, and predicting the optimal adding time, the optimal adding mode and the optimal adding verbal skill according to the key features; and generating an optimal contact strategy of the target object according to the predicted optimal adding time, optimal adding mode and optimal adding verbal skill, and pushing the optimal contact strategy to the user. Intelligent friend adding prediction is performed through the relevant information of the target object in combination with the service scene, so that the best contact strategy can be obtained in combination with the demand and behavior preference of the target object, and the success rate and efficiency of friend adding are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, equipment and medium for adding intelligent friends based on a large model. Background Art

[0002] With the development of computer technology, digital transformation is playing an increasingly important role in various fields. For example, institutions in many fields will adopt multi-channel online and offline traffic diversion methods, such as video accounts, live broadcasts, official accounts, mini-programs and other channels within the WeChat ecosystem, as well as online traffic diversion methods of public domain platforms such as Douyin, Xiaohongshu, and Kuaishou. These channels provide various fields with abundant customer contact opportunities, but also bring new challenges in how to establish connections with customers efficiently and accurately.

[0003] In the financial sector, frontline salespeople often proactively add clients as corporate WeChat friends. While this approach is direct, it ignores the client's subjective wishes and actual needs. Forcibly adding clients to WeChat at inappropriate times or circumstances can not only cause customer resentment but also reduce customer conversion rates.

[0004] In the healthcare sector, especially in the management and services for patients with chronic diseases, hospitals and medical staff are increasingly relying on online channels to communicate with patients in order to provide more convenient medical services and health management advice. Medical staff also face similar difficulties when trying to establish contact with patients through WeChat for Business. Patients may feel disturbed due to physical discomfort or bad mood.

[0005] Therefore, the problem of how to add customer friends in a more intelligent and personalized way so as to respect customer preferences and improve the success rate of adding friends still needs to be solved urgently. Summary of the Invention

[0006] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a large-scale model-based intelligent friend adding method, device, equipment and medium that can be applied to the medical field, financial technology or other related fields. Its main purpose is to improve the success rate and efficiency of friend adding.

[0007] The technical solutions of the present invention are as follows:

[0008] A first aspect of the present invention provides a large model-based intelligent friend adding method, comprising:

[0009] Get the basic information and behavior information of the target object to be added;

[0010] Perform feature engineering on the basic information and behavior information of the target object according to the target business scenario, and extract key features of the target business scenario from the basic information and behavior information;

[0011] Input the key features into a pre-fine-tuned large model, and predict the optimal time, method, and words for adding the key features based on the key features;

[0012] An optimal contact strategy for the target object is generated according to the predicted optimal adding time, optimal adding method, and optimal adding words, and the optimal contact strategy is pushed to the user.

[0013] A second aspect of the present invention provides an intelligent friend adding device based on a large model, comprising:

[0014] The acquisition module is used to obtain the basic information and behavior information of the target object to be added;

[0015] A feature engineering module is used to perform feature engineering on the basic information and behavior information of the target object according to the target business scenario, and extract key features of the target business scenario from the basic information and behavior information;

[0016] A prediction module is used to input the key features into a pre-fine-tuned large model and predict the optimal time, method, and words for adding the key features based on the key features;

[0017] The strategy push module is used to generate the optimal contact strategy for the target object according to the predicted optimal adding time, optimal adding method and optimal adding words, and push the optimal contact strategy to the user.

[0018] A third aspect of the present invention provides a computer device comprising at least one processor; and

[0019] a memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned large model-based intelligent friend adding method.

[0021] A fourth aspect of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the above-mentioned large model-based intelligent friend adding method.

[0022] Beneficial effects: The present invention discloses a method, apparatus, device and medium for adding intelligent friends based on a large model. Compared with the prior art, the embodiments of the present invention use a method, apparatus, device and medium for adding intelligent friends based on a large model, including: obtaining basic information and behavioral information of the target object to be added; performing feature engineering processing on the basic information and behavioral information of the target object according to the target business scenario, and extracting key features under the target business scenario from the basic information and behavioral information; inputting the key features into a large model that has been fine-tuned and trained in advance, and predicting the optimal adding time, optimal adding method and optimal adding words based on the key features; generating the optimal contact strategy for the target object based on the predicted optimal adding time, optimal adding method and optimal adding words, and pushing the optimal contact strategy to the user. By collecting relevant information of the target object and performing intelligent friend adding prediction in combination with the business scenario, the optimal contact strategy can be obtained in combination with the needs and behavioral preferences of the target object, thereby improving the success rate and efficiency of adding friends. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the solutions in the present invention, a brief introduction is given below to the drawings required for use in describing the embodiments of the present invention. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1 A schematic diagram of an application environment for the large-model-based intelligent friend adding method provided by an embodiment of the present invention;

[0025] Figure 2 A flow chart of a large-model-based intelligent friend adding method provided by an embodiment of the present invention;

[0026] Figure 3 A flowchart of step S202 in the large model-based intelligent friend adding method provided in an embodiment of the present invention;

[0027] Figure 4 A flowchart of step S203 in the large model-based intelligent friend adding method provided in an embodiment of the present invention;

[0028] Figure 5 Another flow chart of the large model-based intelligent friend adding method provided by an embodiment of the present invention;

[0029] Figure 6 A schematic diagram of the functional modules of a large-model-based intelligent friend adding device provided by an embodiment of the present invention;

[0030] Figure 7A schematic diagram of the hardware structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] To make the objectives, technical solutions, and effects of the present invention more clear and distinct, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. The embodiments of the present invention are described below with reference to the accompanying drawings.

[0032] The intelligent friend adding method based on the large model provided by the embodiment of the present invention can be applied in the following situations: Figure 1 In an application environment, the system includes a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0033] The user may use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software (for example only).

[0034] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0035] The server 105 may be a server that provides various services, such as a backend server that provides support for the content browsed by the user using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (for example only). The backend server may analyze and process the received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to the user request) to the terminal device. The server 105 may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server 105 may also be a server for a distributed system, or a server combined with a blockchain.

[0036] It should be noted that the large-model-based intelligent friend adding method provided in the embodiments of the present application can generally be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103. Accordingly, the large-model-based intelligent friend adding apparatus provided in the embodiments of the present invention can also be provided in the first terminal device 101, the second terminal device 102, or the third terminal device 103. Alternatively, the large-model-based intelligent friend adding method provided in the embodiments of the present invention can generally be executed by the server 105. Accordingly, the large-model-based intelligent friend adding apparatus provided in the embodiments of the present invention can generally be provided in the server 105.

[0037] It should be understood that the numbers of the above terminal devices, networks and servers are merely illustrative and any number of terminal devices, networks and servers may be provided as required.

[0038] like Figure 2 As shown, the smart friend adding method based on the large model provided by the embodiment of the present invention specifically includes the following steps:

[0039] S201: Obtain basic information and behavior information of the target object to be added.

[0040] In this embodiment, for the target object that the user wants to add as a friend, the basic information and behavioral information of the target object are first obtained, and data is automatically obtained from various data sources through a preset script or tool, or data is obtained from an external platform (such as social media, health monitoring equipment) through an API interface based on the authorization of the target object. Among them, basic information specifically refers to the static attributes of the target object, which is used to preliminarily locate the characteristics of the target object. For example, basic information may include age, gender, occupation, geographic location, income level (in the financial field), medical history (in the medical field), etc.; behavioral information refers to the dynamic behavioral data of the target object in a specific scenario, which is used to deeply understand the preferences and needs of the target object. For example, behavioral information in the financial field may include credit card usage records, bank APP login frequency, interaction records with customer service, etc.; behavioral information in the medical field may include online consultation records, hospital official account interaction records, health monitoring equipment data, etc.

[0041] Specifically, you can obtain the target object's basic information and behavioral data from the customer management system (CRM) and business platforms (such as the bank's credit card management system and the hospital's electronic medical record system). You can also obtain supplementary data through cooperating third-party data platforms (such as financial credit institutions, medical health monitoring platforms, social platforms, etc.), such as video account viewing records, live broadcast participation records, public account subscription records, mini program usage records, etc.

[0042] For example, in the financial sector, a customer's basic information is collected: age (30 years old), gender (male), occupation (programmer), income level (annual salary of 300,000 yuan); behavioral information includes monthly credit card spending (5,000 yuan), credit card repayment record (on time), bank app login frequency (three times a week), and interaction records with customer service (inquiries about credit card promotions).

[0043] In the field of medical health, a patient's basic information is collected: the patient's age (45 years old), gender (female), medical history (hypertension, diabetes); behavioral information: online consultation records (once a month), hospital official account interaction records (pay attention to health science articles), and health monitoring equipment data (blood pressure, blood sugar monitoring records).

[0044] Through multi-source data collection, we ensure that the acquired data fully covers the characteristics and behaviors of the target object. In addition, we can remove irrelevant, invalid, erroneous and duplicate data through data cleaning, further improving the data quality of the target object and ensuring the reliability of subsequent processing.

[0045] S202 : Perform feature engineering on the basic information and behavior information of the target object according to the target business scenario, and extract key features in the target business scenario from the basic information and behavior information.

[0046] In this embodiment, different business scenarios are distinguished based on the friend adding application scenarios of different users. Customer behavior preference analysis can be carried out in a targeted manner in each business scenario. Therefore, feature engineering processing is first performed on the basic information and behavior information of the target object according to the current target business scenario, where the target business scenario refers to a specific application scenario, such as credit card promotion in the financial field or online consultation promotion in the medical field. Through feature engineering processing, key features in the target business scenario are extracted from the basic information and behavior information. The key features refer to features that have a significant impact on the success rate of adding friends, which are used for subsequent model predictions.

[0047] Specific feature engineering processes can include feature extraction and feature selection. For example, in the financial sector, features such as a customer's income level, credit card activity (e.g., monthly spending amount, repayment history), consumption preferences (e.g., consumption categories), and bank app login frequency can be extracted. Correlation analysis can then be used to select key features that are highly correlated with the success rate of adding friends in the target business scenario. By extracting key features based on specific business scenarios, the adaptability of the model is improved. Feature processing and selection can also reduce data redundancy and improve the predictive efficiency of subsequent models.

[0048] For example, in the financial field, features are extracted from the customer's basic information and behavioral information, such as income level (300,000 yuan), credit card activity (monthly consumption of 5,000 yuan, on-time repayment), consumption preferences (preferring electronic product consumption), and bank APP login frequency (3 times a week). The "income level" is standardized, and through correlation analysis, "income level", "credit card activity", and "consumption preference" are selected as key features.

[0049] In the field of medical health, features are extracted from patients' basic information and behavioral information, such as medical history (hypertension, diabetes), frequency of consultation (once a month), health monitoring data (blood pressure 140 / 90 mmHg, blood sugar 7.8 mmol / L), and hospital official account interaction records (paying attention to health popular science articles). Through correlation analysis, "medical history", "frequency of consultation", and "health monitoring data" are selected as key features.

[0050] S203: Input the key features into a large model that has been fine-tuned and trained in advance, and predict the optimal time to add, the optimal way to add, and the optimal words to add based on the key features.

[0051] In this embodiment, the big model refers to a pre-trained large language model (such as GPT-3, BERT, etc.), which has strong language generation and understanding capabilities and can handle complex natural language tasks. After fine-tuning the big model through training data, it can adapt to the prediction of adding friends for different target objects. Based on the big model that has completed fine-tuning training in advance, the extracted key features are input into the big model as input data to predict and generate the best adding time, best adding method and best adding words. For example, it may be predicted that a customer has a higher success rate of adding friends in the afternoon on weekdays, and it is recommended to push WeChat invitations through the bank APP and generate personalized words. Based on the basic information, behavior patterns and preferences of different target objects, the corresponding best adding time, method and words are predicted, so that it can adapt to the needs of different target objects, minimize invalid interference to the target objects, and improve the success rate and efficiency of adding friends.

[0052] For example, in the financial sector, the model predicts that a high-income customer (earning 300,000 yuan) has a higher success rate of adding WeChat on weekday afternoons. It recommends sending a WeChat invitation through the bank's app and generates personalized sales pitches: "Dear customer, thank you for your continued support! We recommend a high-end credit card with privileged services like airport VIP lounge access and high-end gift redemption. Add our customer service on WeChat now for more details!"

[0053] In the healthcare field, the model predicts that patients with chronic diseases (hypertension, diabetes) have a higher success rate of adding WeChat on Monday mornings. The model recommends sending an invitation through the hospital's official account and generates personalized text: "Dear patient, hello! We have noticed that your recent health monitoring data shows significant fluctuations in blood pressure and blood sugar. To better manage your health, we recommend adding our online customer service WeChat account. We will provide you with one-on-one health consultations and professional advice."

[0054] S204: Generate an optimal contact strategy for the target object based on the predicted optimal adding time, optimal adding method, and optimal adding words, and push the optimal contact strategy to the user.

[0055] In this embodiment, based on the predicted optimal contact time, optimal contact method, and optimal contact words, information is integrated to generate an optimal contact strategy for the target object. This optimal contact strategy is then pushed in real time to sales representatives or automated tools via an API interface. For example, in the financial sector, the optimal contact strategy can be pushed to bank customer service personnel through the bank's operating system, who can then send friend invitations to customers through the bank's app based on the personalized strategy. Alternatively, in the healthcare sector, the optimal contact strategy can be pushed to online consultation customer service personnel through the hospital's customer service system, who can then send friend invitations to patients through the hospital's official account. This personalized contact strategy can thus improve the success rate of friend addition and customer response rate.

[0056] In the above embodiment, the present invention discloses a smart friend adding method based on a large model, which obtains the basic information and behavior information of the target object to be added; performs feature engineering processing on the basic information and behavior information of the target object according to the target business scenario, and extracts the key features of the target business scenario from the basic information and behavior information; inputs the key features into the large model that has been fine-tuned and trained in advance, and predicts the optimal adding time, optimal adding method and optimal adding words according to the key features; generates the optimal contact strategy for the target object based on the predicted optimal adding time, optimal adding method and optimal adding words, and pushes the optimal contact strategy to the user. By collecting relevant information of the target object and combining it with the business scenario to make intelligent friend adding predictions, it is possible to obtain the optimal contact strategy based on the needs and behavioral preferences of the target object, thereby improving the success rate and efficiency of friend adding.

[0057] In one embodiment, after step S204, the method further includes:

[0058] receiving user feedback on the execution of the optimal contact strategy;

[0059] Collecting statistics on the execution feedback information according to a preset updating strategy to obtain feedback statistics;

[0060] The large model is updated according to the feedback statistical data.

[0061] In this embodiment, after a user executes the optimal contact strategy, the strategy's execution effectiveness is further optimized by collecting user feedback on the optimal contact strategy. For example, a user interface is provided to receive user evaluation and feedback on the contact strategy. Specifically, the identified feedback information can be qualitative feedback, such as selecting "add friend successfully" or "add friend failed," or quantitative feedback, such as a rating. Based on the collected execution feedback information, data statistics are generated according to a preset update strategy. For example, the preset update strategy may set user feedback statistics at daily or weekly intervals, or statistics are generated after a certain amount of feedback has been collected. Based on the preset update strategy, execution feedback information within a specified time period is collected when update conditions are met, generating feedback statistics that reflect the user's evaluation of the friend-adding strategy. For example, statistical analysis can be performed on execution feedback information within a specified time period, calculating indicators such as the success rate of friend additions as feedback statistics, thereby reflecting the effectiveness of the strategy prediction. Based on the feedback statistics, the large model is updated, for example, by adjusting model parameters such as the learning rate and batch size based on the feedback evaluation results. Periodically use the latest data to readjust the large model to maintain the timeliness and accuracy of the model. You can also set a set of monitoring indicators to track the operating status of the entire system to promptly discover and solve problems, thereby better adapting to the needs of adding friends in actual scenarios.

[0062] In one embodiment, Figure 3 As shown, step S202 includes:

[0063] S301, confirming the corresponding target business scenario according to the currently set scenario identifier;

[0064] S302: screening the basic information and behavior information of the target object according to the target business scenario to obtain data to be processed that matches the target business scenario;

[0065] S303: Perform feature extraction on the data to be processed, and extract corresponding key features from the data to be processed.

[0066] In this embodiment, when performing feature engineering processing on basic information and behaviors, the corresponding target business scenario is first confirmed based on the currently set scenario identifier. The scenario identifier is a label or code used to distinguish different business scenarios. The user can pre-set the default scenario identifier, or input a scene switching instruction to switch the scenario identifier according to different needs. For example, in the financial field, the scenario identifier can be "credit card promotion" or "financial product recommendation" and so on; in the medical field, the scenario identifier can be "chronic disease management" or "health check-up promotion" and so on. Based on the currently set scenario identifier, the system automatically identifies and confirms the corresponding target business scenario to ensure the accuracy of subsequent processing.

[0067] Data filtering rules are predefined based on different target business scenarios. These rules determine which data is useful for strategic predictions for the target business scenario. For example, in a credit card promotion scenario in the financial sector, it may be necessary to filter information such as customer A's income level, credit card usage frequency, and spending preferences. In a chronic disease management scenario in the healthcare sector, it may be necessary to filter patient B's medical history, health monitoring data (such as blood pressure and blood sugar), and medical records. These data filtering rules can be dynamically adjusted based on real-time feedback to adapt to different business needs. Feature extraction is then performed on the filtered data to be processed. For example, for text data, features such as keywords, topics, and sentiment can be extracted. For numerical data, statistical features and patterns can be extracted, such as active features based on online activity time. Key features representing the target subject's identity, sentiment, behavioral preferences, or habits are then extracted from the data to be processed. Data filtering rules are used to extract key features from the target subject's extensive information, identifying data that matches the target business scenario and ensuring that the filtered data and extracted key features are highly relevant to the target business scenario, improving the efficiency and accuracy of subsequent processing.

[0068] In one embodiment, Figure 4 As shown, step S203 includes:

[0069] S401: Input the key features into a pre-fine-tuned large model, perform encoding conversion and customer group analysis on the key features, and obtain corresponding key feature vectors and target groups;

[0070] S402: Calling the prediction network parameters corresponding to the target group to predict the key feature vector for adding a time task, adding a method task, and adding a speech task, and obtaining a predicted probability for each task;

[0071] S403: Determine the optimal time, method, and words for adding tasks based on the ranking of the predicted probabilities of each task.

[0072] In this embodiment, the extracted key features are input into a large model that has been fine-tuned and trained in advance, and the input key features are encoded and converted so that the model can process them. Specifically, categorical features (such as gender, occupation, medical history) can be One-Hot Encoding or Label Encoding, and numerical features (such as income level, blood pressure, blood sugar) can be standardized or normalized to eliminate the dimensional differences between different features. The key features after encoding conversion are converted into feature vectors for subsequent prediction tasks. Customer group analysis is also performed based on key features. For example, all customers are divided into several customer groups based on group characteristics in advance. Different customer groups have corresponding group characteristics. The distance between the key features and the characteristics of the center of each customer group is calculated, and the customer group closest to the center is confirmed as the target group to which the current target object belongs, so that more accurate processing can be achieved based on the characteristics of the target group in subsequent predictions.

[0073] The corresponding prediction network parameters are applied to the target group to which the current target object belongs. Specifically, during the fine-tuning training phase, the large model is trained for different customer groups and the corresponding model parameters are saved. This allows for flexible parameter application during the inference phase based on the target group to further improve prediction accuracy. Based on the prediction network parameters appropriate for the current target group, multi-task predictions are performed on key feature vectors, including the add time task, the add method task, and the add conversation task. The add time task predicts the success probability distribution for different add times. For example, the probability of a customer adding WeChat at different times is predicted. The add method task predicts the probability distribution for different add methods. For example, the success probability distribution for adding friends via the app, official account, SMS, email, and other methods is predicted. The add conversation task predicts the success probability distribution for different add conversation methods. The predicted probabilities for each task are ranked, and the option with the highest probability is selected as the optimal solution. For example, for the add time task, the time point with the highest probability is selected. Similarly, for the add method task, the method with the highest probability is selected. For the add conversation task, the conversation method with the highest probability is selected. This determines the optimal add time, method, and conversation method. By processing key features through large models, we can more accurately understand customer needs and preferences and provide personalized intelligent friend-adding services, thereby improving the customer's friend-adding success rate.

[0074] In one embodiment, Figure 5 As shown, before step S201, the method further includes:

[0075] S501, collecting historical addition samples of friend objects that have been added within a specified range and constructing a training data set;

[0076] S502: Load the pre-trained large model, input the training data set into the pre-trained large model to perform multi-task fine-tuning training until a preset convergence condition is met to obtain a large model that has completed fine-tuning training.

[0077] In this embodiment, during the fine-tuning training phase of the large model, data collection and preprocessing are first performed. Historical addition samples of friend objects that have been added within a specified range are collected through various data sources, such as internal databases, third-party data platforms, etc. The specific specified range can be a time range, business scope, or geographical range, etc. For example, in the financial field, customer data that has been successfully added as friends through bank apps, text messages, etc. in the past year can be collected; in the medical field, patient data that has been successfully added as friends through hospital official accounts, online consultations, etc. in the past year can be collected. The specific historical addition samples collected may include, for example, basic information (such as age, gender, occupation, etc.), behavioral information (such as credit card usage records, health monitoring data, etc.), friend addition information (such as the time of adding WeChat, the method of adding WeChat, the words used to add WeChat, and the content that triggers the customer to actively add WeChat), etc. Based on the collected historical addition samples, data cleaning is performed to remove invalid, erroneous, or duplicate samples to construct a training dataset for subsequent model fine-tuning training, ensuring the diversity of the training dataset and improving data quality.

[0078] Load the pre-trained large model, for example, choose the pre-trained GPT-3 as the basic model, or other large models suitable for natural language processing tasks, input the constructed training data set into the pre-trained large model for multi-task fine-tuning training. Similar to the inference process, before inputting the training data set into the large model, first perform mechanical energy feature engineering on the input data to extract the corresponding key features, such as customer age, gender, occupation, customer activity, historical success rate of adding WeChat, commonly used channels for adding WeChat, interaction in Moments, frequency of login of good car owners, reply rate of corporate WeChat, frequency of communication, etc., so that the large model can better learn the strategy of successfully adding friends from the training data set. The multi-task fine-tuning training includes adding a time task (predicting the optimal micro-adding time), adding a method task (predicting the optimal micro-adding method), and adding a word task (generating the optimal micro-adding word). Using a multi-task learning framework, the underlying feature representation is shared, while the loss functions of multiple tasks are optimized. Hyperparameters such as learning rate, batch size, and number of iterations are adjusted based on cross-validation methods. Model performance is evaluated simultaneously, and training is stopped until convergence conditions are met, such as reaching a preset number of iterations, or the loss function value is less than a preset value, or the performance index reaches a preset index, etc., thereby obtaining a large model that has completed fine-tuning training. Through multi-task fine-tuning training, the large model can adapt to the needs of different added factors, comprehensively improving the prediction accuracy of the large model under different added factors.

[0079] In one embodiment, step S501 includes:

[0080] Collect historical added samples of friend objects that have been added within a specified range, wherein the historical added samples include sample basic data, sample behavior data, and sample added data;

[0081] Performing sample group clustering processing based on the sample basic data and sample behavior data to obtain several sample groups and adding corresponding group labels;

[0082] Performing a time-consuming analysis on the sample addition data, and adding a short-duration label to samples that take less than a preset time;

[0083] Select some objects from each sample group, add corresponding strategy labels based on the sample data of the currently selected objects, and construct a training data set.

[0084] In this embodiment, when constructing the training data set, sample basic data, sample behavioral data, and sample added data of friend objects that have been added within a specified range are first collected. For example, basic data of customers who have added WeChat friends, such as age, gender, occupation, geographic location, etc.; behavioral data such as interaction records on the enterprise WeChat, such as chat records, circle of friends interactions (likes, comments, etc.); behavioral data on other platforms, such as video account viewing records, live broadcast participation records, public account subscription records, mini program usage records, etc. Added data includes data such as the way customers and salesmen add WeChat (active or passive), the source of the WeChat addition channel, the time of adding WeChat, the words used to add WeChat, and the content that triggers customers to actively add WeChat.

[0085] Afterwards, sample groups are clustered based on the collected sample basic data and sample behavior data. Specifically, the samples can be divided into groups through clustering algorithms (such as K-Means and DBSCAN), that is, the characteristics of different object groups are analyzed, and all added friend objects are divided into several sample groups with similar characteristics. For example, customers can be divided into high-income groups, middle-income groups, and low-income groups (in the financial field), or patients can be divided into chronic disease patient groups, acute disease patient groups, etc. (in the medical field), etc. Based on the clustering results, corresponding group labels are added to each sample to facilitate subsequent analysis and processing. In addition, a time-consuming analysis is performed based on sample addition data. This involves calculating the time it takes for each sample to be successfully added as a friend from the start of the micro-addition operation. For example, the time difference from sending the micro-addition invitation to the customer confirming the addition is recorded, and a preset time threshold is set, such as 1 hour or 24 hours. If the sample's time is less than the preset time, the sample is considered a "short-time sample" and a corresponding short-time label is added. Through time-consuming analysis, samples with quick responses are identified among all samples. These samples have a higher success rate for adding micro-additions and are more likely to add friends. The strategic factors such as the methods and words involved in the addition process are more valuable to learn from, thereby improving the fine-tuning effect. Afterwards, a portion of the objects in each sample group are randomly selected as labeled objects. That is, a portion of the sample addition data is labeled with the corresponding strategy labels, while a portion of the sample addition data is retained without labels, thereby constructing a training dataset. The large model is fine-tuned in multiple stages using the group labels of each sample in the training dataset, the short-time labels of some samples, and the strategy labels, thereby improving the effect of model fine-tuning.

[0086] In one embodiment, step S502 includes:

[0087] Loading a pre-trained large model, inputting the training data with the short-duration labels into the pre-trained large model to perform universal addition task learning for all groups;

[0088] Inputting a portion of the training data with the strategy labels into a large model that completes the general addition task learning, and performing multi-task supervised learning based on the strategy labels and group labels to learn to predict addition strategies for different object groups;

[0089] The remaining training data without policy labels is input into the large model that has completed supervised learning for multi-task unsupervised learning until the preset convergence conditions are met to obtain the large model that has completed fine-tuning training.

[0090] In this embodiment, the pre-trained large model is fine-tuned in multiple stages. After loading the pre-trained large model, the training data with short-duration labels are first input into the large model to perform universal adding task learning for all groups. These samples with short-duration labels have a shorter time consumption for adding friends, a higher success rate for adding friends on WeChat, and are easier to add friends. They are suitable as basic data for universal adding task learning. The common features and strategies of these samples are learned to learn how to successfully add friends in a shorter time. For example, the model may learn that the success rate of adding friends on WeChat through the bank APP is higher in the afternoons on weekdays, which provides a good foundation for subsequent multi-task learning.

[0091] The training data with policy labels is then fed into a large model that learns the general addition task. Since a subset of objects in each sample group are selected to have policy labels added, the training data with policy labels covers all sample groups. This labeled data provides clear guidance for the model, enabling it to better adapt to the characteristics of different groups. Therefore, the large model can perform multi-task supervised learning based on policy labels and group labels to learn to predict addition strategies for different groups of objects. Specifically, it simultaneously optimizes the tasks of predicting the time, method, and language for adding micro-objects for different groups, allowing the model to learn more precise patterns and strategies from labeled data.

[0092] Finally, to further enhance the model's ability to generalize to unknown data and ensure that the model performs well when processing unseen data, the remaining training data without policy labels is input into the large model that has completed supervised learning for multi-task unsupervised learning. The goal of unsupervised learning is to allow the model to automatically discover patterns and structures in the data without labeled data, further optimize the model's performance, and stop training when the preset convergence conditions are reached, resulting in a large model that has completed fine-tuning training.

[0093] This embodiment loads a pre-trained large model and sequentially performs general adding task learning, supervised learning, and unsupervised learning. Based on multi-stage fine-tuning training, the final large model can learn the adding strategies of different groups and has good generalization ability, providing reliable support for subsequent intelligent adding strategies, thereby providing personalized intelligent adding friend strategies based on the needs and behavioral preferences of different customer groups, and improving the success rate and efficiency of adding friends.

[0094] It should be noted that there is not necessarily a certain order between the above steps. A person skilled in the art can understand, based on the description of the embodiments of the present invention, that in different embodiments, the above steps may have different execution orders, that is, they may be executed in parallel, or may be executed interchangeably, etc.

[0095] Further references Figure 6 , as a response to the above Figure 2 The present invention provides an embodiment of an intelligent friend adding device based on a large model. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0096] like Figure 6 As shown, the large model-based intelligent friend adding device 60 described in this embodiment includes:

[0097] Acquisition module 601, used to acquire basic information and behavior information of the target object to be added;

[0098] A feature engineering module 602 is configured to perform feature engineering on the basic information and behavior information of the target object according to a target business scenario, and extract key features of the target business scenario from the basic information and behavior information;

[0099] Prediction module 603, configured to input the key features into a pre-fine-tuned large model, and predict the optimal time, method, and words for adding the content based on the key features;

[0100] The strategy push module 604 is used to generate the optimal contact strategy for the target object according to the predicted optimal adding time, optimal adding method and optimal adding words, and push the optimal contact strategy to the user.

[0101] The module referred to in the present invention refers to a series of computer program instruction segments that can complete specific functions. It is more suitable for describing the execution process of intelligent friend adding based on a large model than a program. For the specific implementation of each module, please refer to the corresponding method embodiment above, which will not be repeated here.

[0102] In one embodiment, the feature engineering module 602 includes:

[0103] A scenario confirmation unit, configured to confirm a corresponding target business scenario based on a currently set scenario identifier;

[0104] A data screening unit, configured to screen the basic information and behavior information of the target object according to the target business scenario to obtain data to be processed that matches the target business scenario;

[0105] The feature extraction unit is used to perform feature extraction on the data to be processed and extract corresponding key features from the data to be processed.

[0106] In one embodiment, the prediction module 603 includes:

[0107] An input unit is used to input the key features into a pre-fine-tuned large model, perform encoding conversion and customer group analysis on the key features, and obtain corresponding key feature vectors and target groups;

[0108] Calling a prediction unit, configured to call prediction network parameters corresponding to the target group to predict the key feature vector for adding a time task, adding a method task, and adding a speech task, to obtain a prediction probability for each task;

[0109] The ranking determination unit is used to determine the optimal adding time, optimal adding method and optimal adding words based on the ranking of the predicted probability of each task.

[0110] In one embodiment, the apparatus 60 further includes:

[0111] The sample collection module is used to collect historical samples of friend objects that have been added within a specified range and construct a training data set;

[0112] The fine-tuning module is used to load the pre-trained large model, input the training data set into the pre-trained large model to perform multi-task fine-tuning training until the preset convergence conditions are met to obtain a large model that has completed fine-tuning training.

[0113] In one embodiment, the sample collection module includes:

[0114] A sample collection unit, configured to collect historical added samples of friend objects that have been added within a specified range, wherein the historical added samples include sample basic data, sample behavior data, and sample added data;

[0115] A clustering unit, configured to perform sample group clustering processing based on the sample basic data and sample behavior data, obtain several sample groups and add corresponding group labels;

[0116] A time consumption analysis unit, configured to perform time consumption analysis on the sample addition data and add a short-duration label to samples whose time consumption is less than a preset time length;

[0117] The strategy labeling unit is used to select some objects in each sample group, add corresponding strategy labels according to the sample data of the currently selected objects, and construct a training data set.

[0118] In one embodiment, the fine-tuning module includes:

[0119] An initial training unit is used to load a pre-trained large model and input the training data with the short-duration labels into the pre-trained large model to perform general addition task learning for all groups;

[0120] A supervised learning unit is configured to input a portion of the training data having the strategy labels into a large model for completing the general addition task learning, and perform multi-task supervised learning based on the strategy labels and group labels to learn to predict addition strategies for different object groups;

[0121] The unsupervised learning unit is used to input the remaining training data without policy labels into the large model that has completed supervised learning to perform multi-task unsupervised learning until the preset convergence conditions are met to obtain the large model that has completed fine-tuning training.

[0122] In one embodiment, the apparatus 60 further includes:

[0123] A feedback receiving module, configured to receive user feedback on the execution of the optimal contact strategy;

[0124] A feedback statistics module is used to collect statistics on the execution feedback information according to a preset update strategy to obtain feedback statistics;

[0125] A model updating module is used to update the large model according to the feedback statistical data.

[0126] In the above embodiment, the present invention discloses an intelligent friend adding device based on a large model, which obtains the basic information and behavioral information of the target object to be added; performs feature engineering processing on the basic information and behavioral information of the target object according to the target business scenario, and extracts the key features of the target business scenario from the basic information and behavioral information; inputs the key features into the large model that has been fine-tuned and trained in advance, and predicts the optimal adding time, optimal adding method and optimal adding words according to the key features; generates the optimal contact strategy for the target object based on the predicted optimal adding time, optimal adding method and optimal adding words, and pushes the optimal contact strategy to the user. By collecting relevant information of the target object and combining it with the business scenario to make intelligent friend adding predictions, it is possible to obtain the optimal contact strategy based on the needs and behavioral preferences of the target object, thereby improving the success rate and efficiency of friend adding.

[0127] Another embodiment of the present invention provides a computer device, such as Figure 7 As shown, the computer device 70 includes:

[0128] One or more processors 701 and memory 702, Figure 7 In the description, a processor 701 is used as an example. The processor 701 and the memory 702 can be connected via a bus or other means. Figure 7 The bus connection is taken as an example.

[0129] The processor 701 is used to complete various control logics of the computer device 70. It can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine) or other programmable logic device, discrete gate or transistor logic, discrete hardware components or any combination of these components. In addition, the processor 701 can also be any traditional processor, microprocessor or state machine. The processor 701 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0130] Memory 702, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs, and modules, such as the program instructions corresponding to the large-model-based intelligent friend-adding method in the embodiments of the present invention. Processor 701 executes the non-volatile software programs, instructions, and modules stored in memory 702 to execute various functional applications and data processing of computer device 70, thereby implementing the large-model-based intelligent friend-adding method in the aforementioned method embodiments.

[0131] The memory 702 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device 70, etc. In addition, the memory 702 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 702 may optionally include a memory remotely located relative to the processor 701, and these remote memories may be connected to the computer device 70 via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. One or more units are stored in the memory 702, and when executed by one or more processors 701, the steps of the intelligent friend adding method based on the large model in any of the above-mentioned method embodiments are executed.

[0132] In the above embodiment, the present invention discloses a computer device, which obtains the basic information and behavioral information of the target object to be added; performs feature engineering processing on the basic information and behavioral information of the target object according to the target business scenario, and extracts the key features of the target business scenario from the basic information and behavioral information; inputs the key features into a large model that has been fine-tuned and trained in advance, and predicts the optimal time to add, the optimal way to add, and the optimal words to add according to the key features; generates the optimal contact strategy for the target object based on the predicted optimal time to add, the optimal way to add, and the optimal words to add, and pushes the optimal contact strategy to the user. By collecting relevant information of the target object and combining it with the business scenario to make intelligent predictions on friend addition, it is possible to obtain the optimal contact strategy based on the needs and behavioral preferences of the target object, thereby improving the success rate and efficiency of friend addition.

[0133] An embodiment of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the steps of the large-model-based intelligent friend adding method in any of the above-mentioned method embodiments are executed.

[0134] In the above embodiment, the present invention discloses a non-volatile computer-readable storage medium, which obtains basic information and behavioral information of a target object to be added; performs feature engineering on the basic information and behavioral information of the target object according to the target business scenario, and extracts key features of the target business scenario from the basic information and behavioral information; inputs the key features into a large model that has been fine-tuned and trained in advance, and predicts the optimal time to add, the optimal way to add, and the optimal words to add based on the key features; generates the optimal contact strategy for the target object based on the predicted optimal time to add, the optimal way to add, and the optimal words to add, and pushes the optimal contact strategy to the user. By collecting relevant information of the target object and combining it with the business scenario to make intelligent predictions on friend addition, it is possible to obtain the optimal contact strategy based on the needs and behavioral preferences of the target object, thereby improving the success rate and efficiency of friend addition.

[0135] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention, 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 enabling 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 invention.

[0136] The present invention can be used in a wide variety of general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present invention can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present invention can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0137] In summary, the method, apparatus, device and medium for intelligent friend adding based on a large model disclosed in the present invention include: obtaining basic information and behavioral information of the target object to be added; performing feature engineering processing on the basic information and behavioral information of the target object according to the target business scenario, and extracting key features under the target business scenario from the basic information and behavioral information; inputting the key features into a large model that has been fine-tuned and trained in advance, and predicting the optimal adding time, optimal adding method and optimal adding words according to the key features; generating the optimal contact strategy for the target object based on the predicted optimal adding time, optimal adding method and optimal adding words, and pushing the optimal contact strategy to the user. By collecting relevant information of the target object and performing intelligent friend adding predictions in combination with the business scenario, the optimal contact strategy can be obtained in combination with the needs and behavioral preferences of the target object, thereby improving the success rate and efficiency of adding friends.

[0138] Of course, those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware (such as a processor, controller, etc.) through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes in the above-described method embodiments. The storage medium can be a memory, a magnetic disk, a floppy disk, a flash memory, an optical storage device, etc.

[0139] It should be noted that if any software tools or components not developed by our company appear in the examples of this application, they are for illustration purposes only and do not represent actual use. It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications shall fall within the scope of protection of the appended claims.

Claims

1. A smart friend adding method based on a large model, characterized in that: include: Get the basic information and behavior information of the target object to be added; Perform feature engineering on the basic information and behavior information of the target object according to the target business scenario, and extract key features of the target business scenario from the basic information and behavior information; Input the key features into a pre-fine-tuned large model, and predict the optimal time, method, and words for adding the key features based on the key features; An optimal contact strategy for the target object is generated according to the predicted optimal adding time, optimal adding method, and optimal adding words, and the optimal contact strategy is pushed to the user.

2. The method for adding smart friends based on a large model according to claim 1, characterized in that: The step of performing feature engineering on the basic information and behavior information of the target object according to the target business scenario, and extracting key features of the target business scenario from the basic information and behavior information, includes: Confirm the corresponding target business scenario based on the currently set scenario identifier; Screening the basic information and behavior information of the target object according to the target business scenario to obtain data to be processed that matches the target business scenario; Feature extraction is performed on the data to be processed, and corresponding key features are extracted from the data to be processed.

3. The method for adding smart friends based on a large model according to claim 1, characterized in that: The key features are input into a pre-fine-tuned large model, and the optimal addition time, optimal addition method, and optimal addition words are predicted based on the key feature vector, including: Input the key features into a pre-fine-tuned large model, perform encoding conversion and customer group analysis on the key features, and obtain corresponding key feature vectors and target groups; Calling the prediction network parameters corresponding to the target group to predict the key feature vector for adding a time task, adding a method task, and adding a speech task to obtain a prediction probability for each task; The optimal time, method, and words for adding tasks are determined based on the ranking of the predicted probabilities of each task.

4. The method for adding smart friends based on a large model according to claim 1, characterized in that: Before obtaining the basic information and behavior information of the target object to be added, the method further includes: Collect historical samples of friend objects added within a specified range and construct a training dataset; The pre-trained large model is loaded, and the training data set is input into the pre-trained large model for multi-task fine-tuning training until the preset convergence conditions are met to obtain a large model that has completed fine-tuning training.

5. The method for adding smart friends based on a large model according to claim 4, characterized in that: The collecting of historical added samples of friend objects within a specified range and constructing a training dataset includes: Collect historical added samples of friend objects that have been added within a specified range, wherein the historical added samples include sample basic data, sample behavior data, and sample added data; Performing sample group clustering processing based on the sample basic data and sample behavior data to obtain several sample groups and adding corresponding group labels; Performing a time-consuming analysis on the sample addition data, and adding a short-duration label to samples that take less than a preset time; Select some objects from each sample group, add corresponding strategy labels based on the sample data of the currently selected objects, and construct a training data set.

6. The method for adding smart friends based on a large model according to claim 5, characterized in that: The loading of the pre-trained large model and inputting the training data set into the pre-trained large model for multi-task fine-tuning training until a preset convergence condition is met to obtain a large model that has completed fine-tuning training, including: Loading a pre-trained large model, inputting the training data with the short-duration labels into the pre-trained large model to perform universal addition task learning for all groups; Inputting a portion of the training data with the strategy labels into a large model that completes the general addition task learning, and performing multi-task supervised learning based on the strategy labels and group labels to learn to predict addition strategies for different object groups; The remaining training data without policy labels is input into the large model that has completed supervised learning for multi-task unsupervised learning until the preset convergence conditions are met to obtain the large model that has completed fine-tuning training.

7. The method for adding smart friends based on a large model according to any one of claims 1 to 6, characterized in that: After generating the optimal contact strategy for the target object based on the predicted optimal adding time, optimal adding method, and optimal adding words, and pushing the optimal contact strategy to the user, the method further includes: receiving user feedback on the execution of the optimal contact strategy; Collecting statistics on the execution feedback information according to a preset updating strategy to obtain feedback statistics; The large model is updated according to the feedback statistical data.

8. An intelligent friend adding device based on a large model, characterized in that: include: The acquisition module is used to obtain the basic information and behavior information of the target object to be added; A feature engineering module is used to perform feature engineering on the basic information and behavior information of the target object according to the target business scenario, and extract key features of the target business scenario from the basic information and behavior information; A prediction module is used to input the key features into a pre-fine-tuned large model and predict the optimal time, method, and words for adding the key features based on the key features; The strategy push module is used to generate the optimal contact strategy for the target object according to the predicted optimal adding time, optimal adding method and optimal adding words, and push the optimal contact strategy to the user.

9. A computer device, characterized in that: comprising at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the large model-based intelligent friend adding method described in any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by one or more processors, enable the one or more processors to execute the large model-based intelligent friend adding method described in any one of claims 1-7.

Images