Call strategy method and device based on predicted user behavior, equipment and medium
By generating a personalized call strategy based on a time series model that predicts user behavior, the problems of unclear strategies and low call effectiveness rate of agents are solved, and the call success rate and user satisfaction are improved.
Patent Information
- Application Number
- CN202410952348.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The strategy of the outbound call industry is unclear and the call effectiveness rate is low, resulting in a decrease in customer call success rate and effectiveness.
By obtaining the user's historical call behavior data, the trained time series model is used to extract feature vectors to generate personalized call strategies, including call timing selection, call content recommendation and call quality optimization, helping agents adjust call strategies based on predicted user behavior.
It improves the success rate and effectiveness of calls, meets users' personalized needs, and enhances user satisfaction and sales opportunities.
Smart Images

Figure CN118945270A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent seats, and in particular to a call strategy method, device, computer equipment and computer storage medium based on predicting user behavior. Background Art
[0002] The outbound call industry plays a vital role in many fields such as finance, e-commerce, insurance, education, and government affairs, and is an important bridge for communication between enterprises and customers. With the development of technology and the growth of market demand, the outbound call industry still faces many challenges and problems. In applications such as telemarketing and text marketing, traditional manual sales methods have been unable to help businesses find target customers quickly and accurately. A large number of potential users have become resistant to telemarketing due to long-term telemarketing, especially when they receive calls when they are busy at work. This not only causes trouble for users, but also greatly reduces the success rate and effectiveness of outbound calls. Therefore, how to improve the success rate and effectiveness of outbound calls to customers has become an urgent problem to be solved. Summary of the invention
[0003] The purpose of the present application is to provide a call strategy method, device, computer equipment and computer storage medium based on predicting user behavior, so as to at least solve the problems of unclear outbound call strategy and low call effectiveness rate of current agents.
[0004] To solve the above technical problems, the present application provides a call strategy method based on predicting user behavior, including: Obtain the user's historical call behavior data; Send the user's historical behavior data to the trained time series model; Extracting a feature vector of the historical behavior data based on the time series model, matching the feature vector according to the behavior feature rules learned by the time series model, and determining user behavior data matching the feature vector as the predicted behavior data of the user; Generating a personalized call strategy for the user according to the predicted behavior data, the call strategy including selection of call timing, personalized recommendation of call content, and optimization method of call quality; The call strategy is fed back to the agent so that the agent can initiate a call to the user according to the call strategy.
[0005] Optionally, the trained time series model is trained according to the following steps, including: Collecting call data of platform users, including at least one of the following behavior data: rejection, blacklisting, immediate hang-up, anti-marketing after answering, successful marketing after answering, and no business demand after answering; Marking the behavior data, call date, and call time in the call data to form a user behavior data set; Preprocessing the data of the user behavior data set, and constructing a user behavior data set to be tested according to the behavior data, call date and call time; The user behavior data set to be tested is input into the LSTM model for training to obtain a time series model that can predict user behavior, wherein the LSTM model is preset with the number of neuron cores, data input format, training method and verification method.
[0006] Optionally, the preprocessing of the data of the user behavior data set includes: The data of the user behavior data set is preprocessed by at least one of data cleaning, outlier removal, and data normalization.
[0007] Optionally, the step of inputting the user behavior data set to be tested into an LSTM model for training to obtain a time series model capable of predicting user behavior includes: The user behavior data set to be tested is divided into a training set and a validation set, the model is trained using the training set, and the weights and biases of the model are updated by a back propagation algorithm; wherein, during the training process, regularization or a Dropout layer is used to restrict the fitting of the model; Determining a performance indicator of the model according to the validation set, and adjusting a hyperparameter of the model according to the performance indicator to obtain an initially trained model; Using an independent test data set to evaluate the initially trained model, and calculating several evaluation indicators of the model; An optimization strategy for the model is configured according to the result of each evaluation indicator, and parameters are configured according to the optimization strategy to obtain a time series model that can predict user behavior.
[0008] Optionally, after sending the user's historical behavior data to the trained time series model, the method further includes: Importing the user's historical behavior data as input data into the time series model, transforming the input data, and passing it into the prediction function for single-step prediction to obtain a predicted value; Performing inverse scaling and inverse differentiation on the predicted value, restoring the predicted value to a preset value range, and obtaining a restored value; A corresponding feature vector is determined according to the restored value to match the predicted behavior of the user according to the feature vector.
[0009] Optionally, the time series model is further configured with an attention mechanism, and the attention mechanism is configured with learning attention weights of different data; The determining of the user behavior data matching the feature vector includes: The feature vector is configured according to the learned attention weights of different data, and user behavior data matching the feature vector is determined.
[0010] Optionally, generating a personalized call strategy for the user according to the predicted behavior data further includes: Acquire an external data source of the user, where the external data source includes at least one of the user's personal data, social network data, or other data related to the call behavior; A personalized call strategy for the user is generated based on the predicted behavior data and the external data source.
[0011] In order to solve the above technical problems, the present application also provides a call strategy device based on predicting user behavior, including: Data acquisition module: used to obtain the user's historical call behavior data; Data transmission module: used to send the user's historical behavior data to the trained time series model; Behavior prediction module: used to extract the feature vector of the historical behavior data based on the time series model, and match it according to the behavioral feature rules learned by the time series model, and determine the user behavior data matching the feature vector as the predicted behavior data of the user; Strategy matching module: used to generate a personalized call strategy for the user according to the predicted behavior data, wherein the call strategy includes the selection of call timing, personalized recommendation of call content, and optimization method of call quality; Strategy feedback module: used to feed back the call strategy to the agent so that the agent can initiate a call to the user according to the call strategy.
[0012] In order to solve the above technical problems, the present application also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the above-mentioned call strategy method based on predicting user behavior.
[0013] In order to solve the above technical problems, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the above-mentioned call strategy method based on predicting user behavior.
[0014] The beneficial effects of the embodiments created by the present application are: by acquiring the user's historical call behavior data; sending the user's historical behavior data to a trained time series model; extracting the feature vector of the historical behavior data based on the time series model, and matching it according to the behavioral feature rules learned by the time series model, determining the user behavior data matching the feature vector as the predicted behavior data of the user; generating a personalized call strategy for the user based on the predicted behavior data, the call strategy including the selection of call timing, personalized recommendation of call content, and optimization method of call quality; feeding back the call strategy to the agent so that the agent can initiate a call to the user according to the call strategy, outputting the important features of the user's historical behavior to the AI model, and the AI model predicting the user's behavioral response to marketing calls based on the user's historical behavior, and matching the call strategy for the user based on the behavioral response, which helps the agent to adopt different strategies to call and talk to the user, thereby improving the success rate and effectiveness of the call. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A basic flow chart of a call strategy method based on predicting user behavior according to a specific embodiment of the present application; Figure 2 A basic structural diagram of a call strategy device based on predicting user behavior according to a specific embodiment of the present application; Figure 3 This is a basic structural block diagram of a computer device according to a specific embodiment of the present application. DETAILED DESCRIPTION
[0016] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.
[0017] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.
[0018] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.
[0019] It will be understood by those skilled in the art that the term "terminal" as used herein includes both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capability, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of performing two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices, which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which may combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, a pager, Internet / intranet access, a web browser, a notepad, a calendar and / or a GPS (Global Positioning System) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with a radio frequency receiver. The "terminal" used here can be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable and / or configured to operate locally, and / or in a distributed form, at any other location on the earth and / or in space. The "terminal" used here can also be a communication terminal, an Internet terminal, a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device) and / or a mobile phone with music / video playback function, or a smart TV, a set-top box and other devices.
[0020] The hardware referred to by the names such as "server", "client", and "service node" in this application is essentially an electronic device with capabilities equivalent to those of a personal computer. It is a hardware device that has the necessary components revealed by the von Neumann principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device. The computer program is stored in its memory, and the central processing unit calls the program stored in the external memory into the internal memory for execution, executes the instructions in the program, and interacts with the input and output devices to complete specific functions.
[0021] It should be pointed out that the concept of "server" referred to in this application can also be extended to the case of server clusters. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or integrated into a physical computer or a set of computer clusters. Those skilled in the art should understand this flexibility, and should not use it to restrict the implementation of the network deployment method of this application.
[0022] Unless expressly specified, one or more technical features of the present application can be deployed on a server for implementation and accessed by a client through a remote call to obtain an online service interface provided by the server, or can be directly deployed and run on a client for access.
[0023] Unless explicitly specified, the AI models referenced or may be referenced in this application can be deployed on a remote server and remotely called on the client, or can be deployed and directly called by a client with sufficient device capabilities. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.
[0024] Unless explicitly specified, the various data involved in this application can be stored remotely on a server or on a local terminal device, as long as it is suitable for being called by the technical solution of this application.
[0025] Those skilled in the art should be aware that, although the various methods of the present application are described based on the same concept and thus present commonality to each other, unless otherwise specified, these methods can be independently executed. Similarly, for each embodiment disclosed in the present application, they are all proposed based on the same inventive concept, therefore, concepts with the same expression, and concepts that are appropriately changed for convenience despite different expressions, should be understood as equivalent.
[0026] Unless the mutually exclusive relationship between the embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in each embodiment can be cross-combined to flexibly construct a new embodiment, as long as such combination does not deviate from the creative spirit of this application and can meet the needs of the prior art or solve certain deficiencies in the prior art. Those skilled in the art should be aware of this flexibility.
[0027] See also Figure 1 , Figure 1 Schematic diagram of the basic flow of the call strategy method based on predicting user behavior in this embodiment.
[0028] like Figure 1 As shown, including: S1100, obtaining historical call behavior data of the user; This embodiment is applied to the outbound call scenarios of agents in multiple fields such as finance, e-commerce, insurance, and education. In this embodiment, an intelligent outbound call system is configured to connect users and agents. For example, the intelligent outbound call system can connect users and agents through telephone calls. The intelligent outbound call system can also be deployed as a function on a web page or client, connecting users and agents by providing network calls, or connecting users and agents through chat dialog boxes. The intelligent outbound call system is configured with certain call strategies. When the agent needs to contact the user or call the user, the intelligent outbound call system will feedback the call strategy matched for each user to the agent, and the agent will better serve each user according to the call strategy. Specifically, the user's historical call behavior data is first obtained. The historical call call data includes several behaviors of the agent after dialing the phone to the user, such as rejection, blacklisting, immediate hang-up, anti-marketing after answering, successful marketing after answering, and no business demand after connecting. In practice, a call record system can be established to record the call process between users and agents. The call record system can capture information such as the call start time, call duration, summary of the call content, and user satisfaction evaluation, and generate call behavior data based on this information. The call behavior data is stored in the database and associated with the user's identity information for subsequent use. The intelligent outbound call system can obtain the user's historical call behavior data in the database.
[0029] It should be noted that in order to protect the privacy of users, the call record system of this embodiment is also configured with the function of complying with relevant data protection regulations and privacy policies, and the call record system also uses multiple methods for encryption to ensure the secure storage and access of call records.
[0030] S1200, sending the user's historical behavior data to the trained time series model; After obtaining the user's historical call behavior data, the user's historical behavior data is sent to a trained time series model. The intelligent outbound call system has a built-in time series model, or the intelligent outbound call system can be linked to a time series model, which has been pre-trained and continuously updated and iterated according to real-time production data, and then the obtained user's historical behavior data is sent to the trained time series model. Specifically, the trained time series model is trained according to the following steps, including: S1: Collecting call data of platform users, the call data includes at least one of the following behavior data: rejection, blacklisting, immediate hang-up, anti-marketing after answering, successful marketing after answering, and no business demand after answering; In order to train the time series model, it is necessary to collect the call data of platform users. The call data should include behavioral data such as rejection, blacklisting, immediate hang-up, anti-marketing after answering, successful marketing after answering, and no business demand after answering. In one embodiment, these data can be obtained through the call records or user feedback of the call record system. Furthermore, the call data also includes the user's consumption capacity, historical consumption behavior, historical evaluation, historical browsing data, user shopping needs and other data.
[0031] S2. Marking the behavior data, call date, and call time in the call data to form a user behavior data set; For the collected call data, this embodiment uses the historical behavior data (action), call date (data), and call time (time) therein to train the model to predict the user's future behavior. Therefore, the behavior data, call date, and call time in the call data are marked, which can facilitate subsequent data processing and model training. In one embodiment, a corresponding label is added to each behavior data or binary encoding is used to represent different behavior types to form a user behavior data set.
[0032] S3, preprocessing the data of the user behavior data set, and constructing a user behavior data set to be tested according to the behavior data, call date and call time; Before model training, it is necessary to standardize the data set. In this embodiment, the user behavior data set is preprocessed. The preprocessing may include data cleaning, outlier processing, missing value filling and other steps. According to the preprocessed user behavior data set, the user behavior data set to be tested is constructed. The data of the user behavior data set to be tested will be used as the input data of the model. In this embodiment, a preset feature representation method is selected to convert the behavior data into a one-hot encoding or use an embedded representation, or the call date and call time are converted into an appropriate time series format. In one embodiment, the Data column and the Time column of the user behavior data set are connected, the Data-Time column and the user's historical behavior are retained, and the Data column is set as an index. For example, first convert the Datatime data into a two-dimensional array format. For example, after the first three rows of data are converted, they become: [[2.6838], [2.6882], [2.7048]], and then perform a differential conversion on it. That is, except for the first number, the other numbers are changed to the difference between itself and the previous number. For example, after the differential conversion of the first three rows of data, they become: [[2.6882-2.6838], [2.7048-2.6882]]. Converting time series data into supervised learning sets, for example, converting [
[10] ,
[11] ,
[12] ,
[13] ,
[14] ] into [[0,10],[10,11],[11,12],[12,13],[13,14]], that is, taking the former number as input and the latter number as the corresponding output, and scaling the user's historical behavior data to between [-1,1], can speed up convergence and improve the efficiency of model training.
[0033] It should be noted that features can also be constructed based on behavioral data, call date, and call time, such as calculating call duration, converting date and time formats, etc.
[0034] S4. Input the user behavior data set to be tested into the LSTM model for training to obtain a time series model that can predict user behavior, wherein the LSTM model is preset with the number of neuron cores, data input format, training method, and verification method.
[0035] The constructed user behavior data set to be tested is input into the LSTM (Long Short-Term Memory) model for training. The LSTM model is a recurrent neural network model suitable for processing sequence data. During the training process, the number of neuron cores, data input format, training method and verification method of the LSTM model can be set according to the actual situation, and a deep learning framework (such as TensorFlow, PyTorch, etc.) is used to implement the training process of the LSTM model. Through the training of the LSTM model, a time series model that can predict user behavior is obtained. The time series model can predict the user's future behavior based on the input user behavior data and previous time series information. Furthermore, this embodiment also uses model evaluation indicators (such as accuracy, recall rate, etc.) to evaluate the performance of the model, and then tune and improve it, so as to continuously optimize the time series model that can predict user behavior.
[0036] S1300, extracting a feature vector of the historical behavior data based on the time series model, and matching the feature vector according to the behavior feature rules learned by the time series model, and determining user behavior data matching the feature vector as the predicted behavior data of the user; After sending the user's historical behavior data to the trained time series model, the trained time series model is used to extract features from the historical behavior data to obtain a state representation of the hidden layer, and the state of the hidden layer is regarded as a feature vector of the historical behavior data. Next, matching is performed according to the behavioral feature law learned by the time series model, and the similarity between the feature vector of the extracted historical behavior data and the feature vector of the preset behavioral feature law is calculated. According to the threshold of the similarity matching degree, the user behavior data that matches the feature vector of the extracted historical behavior data is determined as the predicted behavior data of the user.
[0037] S1400, generating a personalized call strategy for the user according to the predicted behavior data, wherein the call strategy includes selection of a call timing, personalized recommendation of call content, and a method for optimizing call quality; After extracting the feature vector of the historical behavior data based on the time series model, matching it according to the behavioral feature rules learned by the time series model, and determining the user behavior data matching the feature vector as the predicted behavior data of the user, the call strategy includes the selection of call timing, personalized recommendation of call content, and optimization of call quality, wherein the appropriate call timing is determined based on the user's active time period and call history pattern to maximize the call connection rate and user participation. Secondly, by analyzing the user's interests, preferences and predicted behavior data, personalized call content is recommended for each user. For example, if the user often consults a certain type of product, relevant product information or promotional activities can be provided. In addition, technical means such as noise reduction algorithms and automatic gain control can be used to optimize call quality and ensure that the sound is clear and audible.
[0038] S1500: Feedback the call strategy to the agent so that the agent can initiate a call to the user according to the call strategy.
[0039] After generating the personalized call strategy of the user according to the predicted behavior data, the call strategy is fed back to the agent. In order to convey the call strategy to the agent, the following methods can be used. First, the generated personalized call strategy is integrated into the agent's work interface, such as displaying call strategy related information in the customer relationship management (CRM) system. Second, the call strategy can be sent to the agent in real-time notification, such as via email, mobile phone text message or instant messaging tool. Before talking with the user, the agent can view the call strategy and adjust his or her behavior according to the strategy guidance to better meet the user's needs and expectations.
[0040] The intelligent outbound calling system of this embodiment can achieve the following effects: Improve call efficiency: Generate personalized call strategies based on users' historical behavior data and predicted behavior data, so that agents can talk to users at the right time, reduce unnecessary waiting time and interruptions, and improve call efficiency.
[0041] Improve user satisfaction: By personalized recommendation of call content and optimization of call quality, we can meet users' personalized needs, provide better user experience and enhance user satisfaction with the service.
[0042] Provide agent guidance: present call strategies to agents, provide guidance and suggestions to agents, enable them to better understand user needs and behavioral characteristics, provide services more professionally during calls with users, and improve agent work efficiency and service quality.
[0043] Real-time adaptation to user behavior: By continuously updating and training time series models, it is possible to track and predict changes in user behavior in real time, thereby adjusting call strategies in a timely manner, maintaining close contact with users, and promptly adapting to users' changing needs.
[0044] Increase sales opportunities: By personalizing the content of recommendation calls, you can provide information about related products or services based on the user's interests and preferences, thereby increasing sales opportunities and cross-selling potential.
[0045] In the above implementation, historical call behavior data of the user is obtained; the historical behavior data of the user is sent to a trained time series model; a feature vector of the historical behavior data is extracted based on the time series model, and matched according to the behavioral feature rules learned by the time series model, and the user behavior data matching the feature vector is determined as the predicted behavior data of the user; a personalized call strategy for the user is generated according to the predicted behavior data, and the call strategy includes the selection of call timing, personalized recommendation of call content, and optimization method of call quality; the call strategy is fed back to the agent so that the agent can initiate a call to the user according to the call strategy, and the important features of the user's historical behavior are output to the AI model, and the AI model predicts the user's behavioral response to marketing calls based on the user's historical behavior, and matches the call strategy for the user based on the behavioral response, which helps the agent to adopt different strategies for calling and talking to the user, thereby improving the success rate and effectiveness of the call.
[0046] In some implementations, the preprocessing of the data of the user behavior dataset includes: The data of the user behavior data set is preprocessed by at least one of data cleaning, outlier removal, and data normalization.
[0047] In this embodiment, when preprocessing the data of the user behavior data set, the preprocessing includes data cleaning. Specifically, data cleaning includes data cleaning involving detecting and repairing errors, missing values and inconsistencies in the data set. First, basic statistical analysis is performed on the data, such as calculating the mean, variance, maximum value, minimum value, etc., to detect outliers and outliers in the data; secondly, the missing values are processed, which can be processed by filling the missing values, deleting the samples containing the missing values, or using the interpolation method; finally, the consistency of the data is checked, such as checking whether the data type matches, whether the data format is correct, etc. Missing value processing: Check whether there are missing values in the data set, which can be done by deleting the data samples containing the missing values or using the interpolation method to fill the missing values; outlier detection: Use statistics, methods or visualization tools to detect outliers. Outliers can be found by calculating statistics such as the mean and standard deviation of the data, or drawing visual graphics such as box plots and scatter plots. Outlier processing: For detected outliers, you can choose to delete the data samples where the outliers are located or use interpolation methods to correct them. In addition, outliers refer to data points that are significantly different from other observed values, which may interfere with the training and prediction of the model. The implementation method of removing outliers may include the following steps: first, detect outliers through statistical analysis or visualization tools, such as using box plots, Z-score and other methods, and then, you can choose to delete outliers or use reasonable alternative values for correction; further, when preprocessing the data of the user behavior data set, it also includes data normalization, converting data of different scales and ranges into a unified ratio to eliminate the deviation caused by different dimensions. The implementation method of data normalization may include the following steps: first, select a suitable normalization method, such as maximum-minimum normalization, standardization, etc. Secondly, calculate the statistics required for normalization, such as maximum value, minimum value, mean and variance. Then, the data is normalized and scaled according to the normalization method so that each data point falls within a unified range: In this embodiment, the data is converted to the range of [-1, 1] or [0, 1] by dividing by a fixed base (such as the maximum absolute value in the data).
[0048] This implementation can improve data quality, reduce noise and inconsistency by performing data cleaning, outlier removal and data normalization on the user behavior data set, thereby providing a more reliable data foundation for subsequent model training and analysis.
[0049] In some implementations, inputting the user behavior data set to be tested into an LSTM model for training to obtain a time series model capable of predicting user behavior includes: S41: dividing the user behavior data set to be tested into a training set and a validation set, using the training set to train the model, and updating the weights and biases of the model through a back propagation algorithm; wherein, during the training process, regularization or a Dropout layer is used to restrict the fitting of the model; S42: determining a performance indicator of the model according to the validation set, and adjusting a hyperparameter of the model according to the performance indicator to obtain an initially trained model; S43: Evaluate the initially trained model using an independent test data set, and calculate several evaluation indicators of the model; S44: configuring an optimization strategy of the model according to the result of each evaluation indicator, and configuring parameters according to the optimization strategy to obtain a time series model capable of predicting user behavior.
[0050] In this embodiment, in the process of inputting the user behavior data set to be tested into the LSTM model for training to obtain a time series model that can predict user behavior, specifically, the user behavior data set to be tested is divided into a training set and a validation set: for model operation, the user behavior data set to be tested needs to be divided into a training set and a validation set, wherein the training set is used for the training process of the model, and the validation set is used to evaluate the performance of the model; the method of dividing the training set and the validation set can adopt random sampling or chronological division to ensure the representativeness of the data set and a certain sample distribution balance. Use the training set to train the model, and update the weights and biases of the model through the back propagation algorithm: Use the divided training set to train the model, and the training process usually adopts an optimization method based on the back propagation algorithm, such as gradient descent, by calculating the gradient of the loss function, updating the weights and biases of the model, so that the model can better fit the training data; Use regularization or use the Dropout layer to restrict the model fitting: In order to avoid overfitting, regularization technology or Dropout layer can be used to restrict the model fitting during the model training process. Regularization techniques such as L1 regularization and L2 regularization can limit the size of model parameters by adding regularization terms to the loss function. The Dropout layer can randomly set the output of some neurons to zero, thereby reducing the dependency between neurons and improving the generalization ability of the model; Determine the performance indicators of the model based on the validation set, and adjust the hyperparameters of the model based on the performance indicators: Use the validation set to evaluate the performance of the model and calculate various performance indicators such as accuracy, recall, precision, etc. Based on the performance of the model on the validation set, the hyperparameters of the model, such as learning rate, batch size, hidden layer size, etc., can be adjusted to obtain better model performance. Methods such as grid search, random search, or Bayesian optimization can be used to search for the best hyperparameter combination. Configure the optimization strategy of the model based on the results of each evaluation indicator, and configure the parameters based on the optimization strategy to obtain a time series model that can predict user behavior: Based on the results of the evaluation indicators, the optimization strategy of the model can be determined. For example, if the accuracy of the model is low, you can try to increase the complexity of the model or adjust the model structure. If the model has an overfitting problem, you can further strengthen the regularization technology or adjust the probability of the Dropout layer. Parameters are configured and adjusted according to the optimization strategy, and finally a time series model that can predict user behavior is obtained.
[0051] By dividing the data set into a training set and a validation set, this implementation can evaluate the model and understand its performance on unseen data, which helps to determine the generalization ability and prediction accuracy of the model. By using an independent validation set to evaluate the model, it is possible to promptly detect whether there is an overfitting problem. Overfitting means that the model performs well on the training set but has poor prediction performance on new data. Through the evaluation of the validation set, the complexity of the model can be adjusted or regularization techniques can be used to avoid overfitting. Using an independent test data set to evaluate the initially trained model can verify the performance of the model and ensure its prediction effect in real scenarios, thereby increasing the reliability and credibility of the model.
[0052] In one embodiment, after sending the user's historical behavior data to the trained time series model, the method further includes: Importing the user's historical behavior data as input data into the time series model, transforming the input data, and passing it into the prediction function for single-step prediction to obtain a predicted value; Performing inverse scaling and inverse differentiation on the predicted value, restoring the predicted value to a preset value range, and obtaining a restored value; A corresponding feature vector is determined according to the restored value to match the predicted behavior of the user according to the feature vector.
[0053] In this embodiment, the user's historical behavior data is imported as input data into the time series model, including transforming the input data and passing it into the prediction function of the time series model for single-step prediction. Specifically, it is converted into a format suitable for the time series model, such as organizing the historical behavior data into a time series in chronological order, which includes processing methods such as sliding window segmentation and sequence filling of the data so that it can be imported as input data into the time series model. In the model, the input data is predicted in a single step through the prediction function to obtain the prediction value of the next time step. The predicted value is reverse scaled and reversely differentiated to restore the predicted value to a preset value range to obtain a restored value: the predicted value is usually normalized or standardized to adapt to the training process of the model. In order to restore the predicted value to the value range of the original data, reverse scaling and reverse differential operations are required. The reverse scaling operation can use the same scaling parameters as the training set to convert the predicted value back to the range of the original data. The reverse differential operation is to reverse the difference between the predicted value and the historical data to restore it to the original data. Determine the corresponding feature vector based on the restored value to match the user's predicted behavior based on the feature vector: The predicted user behavior data can be obtained through the restored value, but in order to match it with the actual user behavior, the restored value needs to be converted into a feature vector. The feature vector can include features related to the user behavior, such as user ID, timestamp, behavior type, and other information. Based on the matching of the feature vector, the predicted user behavior can be compared and matched with the actual user behavior to determine the accuracy and credibility of the prediction.
[0054] In this embodiment, the user's historical behavior data can be processed and predicted by the time series model to obtain a restored value, and the corresponding feature vector is determined based on the restored value to match the user's predicted behavior, thereby achieving the accuracy of the prediction of the user's future behavior.
[0055] In one embodiment, the time series model is further configured with an attention mechanism, wherein the attention mechanism is configured with learning attention weights for different data; The determining of the user behavior data matching the feature vector includes: The feature vector is configured according to the learned attention weights of different data, and user behavior data matching the feature vector is determined.
[0056] In this embodiment, the time series model is also configured with an attention mechanism, and the attention mechanism is configured with learning attention weights of different data; the attention mechanism can help the model pay more attention to important information when processing data. For each feature vector, the learning attention weights of different data can be configured through the attention mechanism. These weights represent the importance of each data to a specific feature vector. For example, in a time series model, a self-attention mechanism or a variant of an attention mechanism, such as multi-head attention in a Transformer model, can be used to calculate the learning attention weight. When determining the user behavior data that matches the feature vector, the user behavior data can be configured according to the feature vector by using the learning attention weight. The configuration method can be to perform weighted summation of the user behavior data according to the learning attention weight, and use the attention weight as a weight coefficient. In this way, in the user behavior data, the data corresponding to the feature vector will obtain a higher weight, thereby paying more attention to the user behavior data that matches the feature vector.
[0057] This implementation configures learning attention weights for different data in the time series model through the attention mechanism. According to the learning attention weights, the feature vector is configured to determine the user behavior data that matches the feature vector, so that the model can focus more accurately on important user behavior data during operation, thereby improving the prediction and matching performance of the model.
[0058] In one embodiment, generating a personalized call strategy for the user according to the predicted behavior data further includes: Acquire an external data source of the user, where the external data source includes at least one of the user's personal data, social network data, or other data related to the call behavior; A personalized call strategy for the user is generated based on the predicted behavior data and the external data source.
[0059] In this embodiment, when generating the personalized call strategy of the user according to the predicted behavior data, the external data source of the user is also obtained, wherein the external data source includes at least one of the user's personal profile, social network data or other data related to the call behavior: the external data source also includes the user's personal profile information, such as name, age, gender, and the user's social network data, such as friend relationships, hobbies, preferences, etc. It is also possible to consider obtaining other data related to the call behavior, such as the user's call records, phone book contacts and other information. By obtaining these external data sources, the user's personal characteristics and behavior background can be better understood. Then, the personalized call strategy of the user is generated according to the predicted behavior data and the external data source, and the predicted behavior data is combined with the external data source to generate the personalized call strategy of the user. First, according to the predicted behavior data, the user's call behavior trend, preference or need can be understood. For example, if it is predicted that the user prefers to talk or has a need for talking in a specific time period, the call strategy can be adjusted accordingly. Secondly, combined with the information of the external data source, a targeted call strategy can be customized according to the user's personal profile, social network data and other characteristics. For example, if the user shows a strong interest in a topic in a social network, the relevant topic can be mentioned in the call to increase the user's participation and satisfaction.
[0060] This embodiment can generate a personalized call strategy for users by utilizing predicted behavior data and external data sources. Such a strategy can better meet the needs and preferences of users, improve user experience, and increase the effectiveness and benefits of calls.
[0061] Please refer to Figure 2 , Figure 2 Schematic diagram of the basic structure of the call strategy device based on predicting user behavior in this embodiment.
[0062] like Figure 2 As shown, a call strategy device based on predicting user behavior includes: a data acquisition module 1100: used to obtain the user's historical call behavior data; a data transmission module 1200: used to send the user's historical behavior data to a trained time series model; a behavior prediction module 1300: used to extract the feature vector of the historical behavior data based on the time series model, and match it according to the behavioral feature rules learned by the time series model, and determine the user behavior data matching the feature vector as the predicted behavior data of the user; a strategy matching module 1400: used to generate a personalized call strategy for the user according to the predicted behavior data, and the call strategy includes the selection of call timing, personalized recommendation of call content and optimization method of call quality; a strategy feedback module 1500: used to feed back the call strategy to the agent, so that the agent can initiate a call to the user according to the call strategy.
[0063] The historical behavior data of the user of the call strategy device based on predicting user behavior can be processed and predicted by the time series model to obtain a restored value, and the corresponding feature vector is determined according to the restored value to match the predicted behavior of the user.
[0064] Optionally, the trained time series model is trained according to the following steps, including: Collecting call data of platform users, including at least one of the following behavior data: rejection, blacklisting, immediate hang-up, anti-marketing after answering, successful marketing after answering, and no business demand after answering; Marking the behavior data, call date, and call time in the call data to form a user behavior data set; Preprocessing the data of the user behavior data set, and constructing a user behavior data set to be tested according to the behavior data, call date and call time; The user behavior data set to be tested is input into the LSTM model for training to obtain a time series model that can predict user behavior, wherein the LSTM model is preset with the number of neuron cores, data input format, training method and verification method.
[0065] Optionally, the preprocessing of the data of the user behavior data set includes: The data of the user behavior data set is preprocessed by at least one of data cleaning, outlier removal, and data normalization.
[0066] Optionally, the step of inputting the user behavior data set to be tested into an LSTM model for training to obtain a time series model capable of predicting user behavior includes: The user behavior data set to be tested is divided into a training set and a validation set, the model is trained using the training set, and the weights and biases of the model are updated by a back propagation algorithm; wherein, during the training process, regularization or a Dropout layer is used to restrict the fitting of the model; Determining a performance indicator of the model according to the validation set, and adjusting a hyperparameter of the model according to the performance indicator to obtain an initially trained model; Using an independent test data set to evaluate the initially trained model, and calculating several evaluation indicators of the model; An optimization strategy for the model is configured according to the result of each evaluation indicator, and parameters are configured according to the optimization strategy to obtain a time series model that can predict user behavior.
[0067] Optionally, the behavior prediction module further includes executing: Importing the user's historical behavior data as input data into the time series model, transforming the input data, and passing it into the prediction function for single-step prediction to obtain a predicted value; Performing inverse scaling and inverse differentiation on the predicted value, restoring the predicted value to a preset value range, and obtaining a restored value; A corresponding feature vector is determined according to the restored value to match the predicted behavior of the user according to the feature vector.
[0068] Optionally, the time series model is further configured with an attention mechanism, wherein the attention mechanism is configured with learning attention weights of different data; The behavior prediction module determines the user behavior data matching the feature vector, including: The feature vector is configured according to the learned attention weights of different data, and user behavior data matching the feature vector is determined.
[0069] Optionally, the strategy matching module further includes executing: Acquire an external data source of the user, where the external data source includes at least one of the user's personal data, social network data, or other data related to the call behavior; A personalized call strategy for the user is generated based on the predicted behavior data and the external data source.
[0070] To solve the above technical problems, the present application also provides a computer device. Figure 3 , Figure 3 This is a basic structural block diagram of the computer device in this embodiment.
[0071] like Figure 3 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected via a system bus. Among them, the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions, and the database may store a control information sequence, and the computer-readable instructions are executed by the processor. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor can execute a call strategy method based on predicting user behavior. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 3The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0072] In this embodiment, the processor is used to execute Figure 2 The memory stores the program codes and various data required to execute the above modules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program codes and data required to execute all submodules in the call strategy device based on predicting user behavior, and the server can call the program codes and data of the server to execute the functions of all submodules.
[0073] The historical behavior data of computer device users can be processed and predicted by the time series model to obtain a restored value, and the corresponding feature vector can be determined based on the restored value to match the user's predicted behavior.
[0074] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the call strategy method based on predicting user behavior described in any of the above embodiments.
[0075] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).
[0076] Those skilled in the art will appreciate that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.
[0077] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A call strategy method based on predicting user behavior, characterized in that: include: Obtain the user's historical call behavior data; Send the user's historical behavior data to the trained time series model; Extracting a feature vector of the historical behavior data based on the time series model, matching the feature vector according to the behavior feature rules learned by the time series model, and determining user behavior data matching the feature vector as the predicted behavior data of the user; Generating a personalized call strategy for the user according to the predicted behavior data, the call strategy including selection of call timing, personalized recommendation of call content, and optimization method of call quality; The call strategy is fed back to the agent so that the agent can initiate a call to the user according to the call strategy.
2. The call strategy method based on predicting user behavior according to claim 1, characterized in that: The trained time series model is trained according to the following steps, including: Collecting call data of platform users, including at least one of the following behavior data: rejection, blacklisting, immediate hang-up, anti-marketing after answering, successful marketing after answering, and no business demand after answering; Marking the behavior data, call date, and call time in the call data to form a user behavior data set; Preprocessing the data of the user behavior data set, and constructing a user behavior data set to be tested according to the behavior data, call date and call time; The user behavior data set to be tested is input into the LSTM model for training to obtain a time series model that can predict user behavior, wherein the LSTM model is preset with the number of neuron cores, data input format, training method and verification method.
3. The call strategy method based on predicting user behavior according to claim 2, characterized in that: The preprocessing of the data of the user behavior data set includes: The data of the user behavior data set is preprocessed by at least one of data cleaning, outlier removal, and data normalization.
4. The call strategy method based on predicting user behavior according to claim 2, characterized in that: The step of inputting the user behavior data set to be tested into the LSTM model for training to obtain a time series model capable of predicting user behavior includes: The user behavior data set to be tested is divided into a training set and a validation set, the model is trained using the training set, and the weights and biases of the model are updated by a back propagation algorithm; wherein, during the training process, regularization or a Dropout layer is used to restrict the fitting of the model; Determining a performance indicator of the model according to the validation set, and adjusting a hyperparameter of the model according to the performance indicator to obtain an initially trained model; Using an independent test data set to evaluate the initially trained model, and calculating several evaluation indicators of the model; An optimization strategy for the model is configured according to the result of each evaluation indicator, and parameters are configured according to the optimization strategy to obtain a time series model that can predict user behavior.
5. The call strategy method based on predicting user behavior according to claim 1, characterized in that: After sending the user's historical behavior data to the trained time series model, the method further includes: Importing the user's historical behavior data as input data into the time series model, transforming the input data, and passing it into the prediction function for single-step prediction to obtain a predicted value; Performing inverse scaling and inverse differentiation on the predicted value, restoring the predicted value to a preset value range, and obtaining a restored value; A corresponding feature vector is determined according to the restored value to match the predicted behavior of the user according to the feature vector.
6. The call strategy method based on predicting user behavior according to claim 1, characterized in that: The time series model is further configured with an attention mechanism, wherein the attention mechanism is configured with learning attention weights of different data; The determining of the user behavior data matching the feature vector includes: The feature vector is configured according to the learned attention weights of different data, and user behavior data matching the feature vector is determined.
7. The call strategy method based on predicting user behavior according to claim 1, characterized in that: The generating of the personalized call strategy of the user according to the predicted behavior data further includes: Acquire an external data source of the user, where the external data source includes at least one of the user's personal data, social network data, or other data related to the call behavior; A personalized call strategy for the user is generated based on the predicted behavior data and the external data source.
8. A call strategy device based on predicting user behavior, characterized in that: include: Data acquisition module: used to obtain the user's historical call behavior data; Data transmission module: used to send the user's historical behavior data to the trained time series model; Behavior prediction module: used to extract the feature vector of the historical behavior data based on the time series model, and match it according to the behavioral feature rules learned by the time series model, and determine the user behavior data matching the feature vector as the predicted behavior data of the user; Strategy matching module: used to generate a personalized call strategy for the user according to the predicted behavior data, wherein the call strategy includes the selection of call timing, personalized recommendation of call content, and optimization method of call quality; Strategy feedback module: used to feed back the call strategy to the agent so that the agent can initiate a call to the user according to the call strategy.
9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the call strategy method based on predicting user behavior as described in any one of claims 1 to 7.
10. A storage medium storing computer-readable instructions, which, when executed by one or more processors, causes the one or more processors to execute the steps of the call strategy method based on predicting user behavior as described in any one of claims 1 to 7.
Citation Information
Cited By
Scene call method and system based on intelligent scene recommendation, and medium
CN119484698A
