Promotion strategy optimization method and device based on operator big data analysis

Abstract

The invention relates to a promotion strategy optimization method and device based on operator big data analysis. The method comprises the steps that user data information is acquired through at least one data source; performing feature extraction on the obtained user data information; constructing an improved random forest model to perform data analysis and prediction on the extracted feature information, and dividing the users into different groups; the groups at least comprise high-value users, high-potential users, potential lost users and negative users; and according to the analysis and prediction result of the improved random forest model, making a personalized promotion strategy for each user group. According to the promotion strategy optimization method based on operator big data analysis provided by the invention, customers can be subdivided more finely, and personalized marketing strategies better meeting customer requirements are formulated, so that the marketing effect and efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a promotion strategy optimization method and device based on operator big data analysis. Background Art

[0002] As market competition becomes increasingly fierce, more and more companies are beginning to pay attention to customer relationship management (CRM), hoping to improve customer satisfaction and loyalty through refined customer classification and personalized marketing strategies, thereby increasing sales and market share. Among them, telecom operators are a typical CRM application scenario. They need to continuously launch new products and services to meet customer needs, while also facing fierce competition and high marketing costs. In this context, how to effectively use the big data resources of operators to conduct accurate customer segmentation and formulate effective marketing strategies has become an important challenge facing telecom operators.

[0003] At present, most telecom operators mainly rely on traditional customer segmentation methods, such as ARPU (average revenue per user), credit rating and historical behavior, to divide customer groups and implement different marketing strategies. However, this method has certain limitations because it cannot take into account the individual differences and real-time changing needs of customers, resulting in poor marketing results and a large waste of marketing resources.

[0004] Therefore, it is necessary to propose a new promotion strategy optimization method based on operator big data analysis, which can segment customers more finely and formulate personalized marketing strategies that better meet customer needs, thereby improving the effectiveness and efficiency of marketing. Summary of the invention

[0005] This application provides a promotion strategy optimization method based on operator big data analysis, which can segment customers more finely and formulate personalized marketing strategies that better meet customer needs, thereby improving the effectiveness and efficiency of marketing.

[0006] An embodiment of the present application provides a promotion strategy optimization method based on operator big data analysis, the optimization method comprising: Acquire user data information through at least one data source; Perform feature extraction on the acquired user data information; Constructing an improved random forest model to perform data analysis and prediction on the extracted feature information, and classifying users into different groups; the groups at least include high-value users, high-potential users, potential churn users, and passive users; According to the analysis and prediction results of the improved random forest model, a personalized promotion strategy is formulated for each user group.

[0007] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the user data information includes a first data set, a second data set and a third data set; The first data set is data information legally and compliantly obtained from the operator's database, and the first data set at least includes the user's basic information, usage, device information and location records; The second data set is public information of users on social media obtained through an API interface, and the second data set at least includes social interactions, activity participation and public evaluation; The third data set is data information legally and compliantly obtained from a third-party data platform, and the third data set at least includes user shopping data and market research data.

[0008] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the feature extraction of the acquired user data information includes: Extract user data information through ETL, clean, transform and standardize the extracted user information, and input the processed data information into the autoencoder; The autoencoder is used to extract features from the acquired user information; the extracted features include at least the user's average monthly usage time, average monthly usage volume, average monthly consumption amount, and user active time period; The feature information extracted by the autoencoder is stored.

[0009] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the expression encoded by the autoencoder is: ; The expression of the self-encoder decoding is: ; in, For status, For each user's data information, is the encoding weight matrix, is the encoding bias vector, is the encoding activation function, is the decoding activation function, is the decoding weight matrix, To decode the bias vector, Feature vector extracted for each user.

[0010] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred scheme, the high-value users are those who respond to promotion activities more times than the average, and the amount of promotional consumption by users is greater than the average; high-potential users are those who respond to promotion activities less than the average, and the amount of promotional consumption by users is greater than the average; potential churn users are those who respond to promotion activities less than the average, and the amount of promotional consumption by users is less than the average; and passive users are users who have never responded to promotional activities.

[0011] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the construction of an improved random forest model to perform data analysis and prediction on the extracted feature information includes: The data in the database are divided into a training set and a test set, wherein 70% is used as the training set and 30% is used as the test set; Initialize parameters; Multiple sample subsets are randomly extracted from the original data set with replacement. The size of each subset is the same as the original data set. A part of the features are randomly selected from all the features as the feature set of the subset. The number of selected features is less than the number of original features, that is: ; For each sample subset, a decision tree is constructed using the selected feature set, expressed as follows: ; The training forest automatically adjusts parameters according to the distribution of the reference data nodes: ; Iteratively optimize the model based on the model parameter adjustment results; in: The parameters include the number of trees , the maximum depth of the tree , the maximum number of features used when splitting each node , the minimum number of sample splits of the tree and the maximum number of sample splits ; is a sample subset, is the original data set; To construct a decision tree function, is a set of parameters; is the error rate of each decision tree, is the error rate of the entire forest; when , readjust the parameters, and the parameter adjustment process is: ; ; in, is the re-adjusted parameter set, For a real set, is the prediction set, is the loss function, is the expected forecast error, Classify and group customer groups. To predict the effect of different promotion strategies.

[0012] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the formulation of personalized promotion strategies for each user group includes: For high-value users, push the first-level promotion strategy; For high-potential users, push the second-level promotion strategy; For potential churn users, push the third-level promotion strategy; For passive users, push the fourth level promotion strategy; The first-level promotion strategy, the second-level promotion strategy, the third-level promotion strategy and the fourth-level promotion strategy have higher activity amounts and / or push frequencies respectively.

[0013] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the optimization method also includes: Obtain feedback data from each user group on the current promotion strategy being pushed; Evaluate current promotional strategies based on feedback data; Dynamically adjust the current promotion strategy being pushed based on the evaluation results.

[0014] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the step of obtaining feedback data of each user group on the pushed promotion strategy includes: During the promotional activities, user behavior data is collected in real time through multiple channels; the multiple channels include real-time records of users using operator services, real-time feedback from users on social media, and shopping behavior data of users on third-party platforms; Processing the collected real-time data; the data processing includes data cleaning, screening and conversion operations; Integrate the processed real-time data with the original user data to update the user's characteristic information; Using the updated user feature information, the improved random forest model is reused for data analysis and prediction to evaluate the changes in user behavior during promotional activities.

[0015] The present application also provides a promotion strategy optimization device based on operator big data analysis, the promotion strategy optimization device comprising: An information collection module, the information collection module is used to obtain user data information through at least one data source; A feature extraction module, wherein the feature extraction module is used to extract features from the acquired user data information; A group division module, which is used to construct an improved random forest model to perform data analysis and prediction on the extracted feature information, and divide users into different groups; the groups include at least high-value users, high-potential users, potential churn users, and passive users; A promotion strategy formulation module is used to formulate personalized promotion strategies for each user group based on the analysis and prediction results of the improved random forest model.

[0016] According to the promotion strategy optimization method and device based on operator big data analysis of the above-mentioned embodiment, user data information is obtained and integrated through multiple data sources, and these scattered and differently formatted data are cleaned, converted and feature extracted by using ETL process and autoencoder, thereby overcoming the problems of data dispersion and inconsistency in the prior art, and bringing about a significant improvement in data consistency and integrity; by constructing an improved random forest model to analyze and predict user behavior, and segmenting users into high-value users, high-potential users, potential churn users and passive users, the user's diverse behavior patterns are deeply understood, the technical problem of insufficient understanding of user behavior is solved, and the operator's ability to accurately identify different user groups is improved; by formulating personalized promotion strategies for each segmented user group, not only can marketing activities be customized according to the user's historical behavior and potential needs, but also the transformation of promotion activity effect evaluation from rough statistics to refined analysis is realized, effectively solving the problem of limitations in promotion activity effect evaluation and improving the efficiency of marketing resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Flow chart of the promotion strategy optimization method provided in the embodiment of the present application (I); Figure 2 Flow chart of the promotion strategy optimization method provided in the embodiment of the present application (II); Figure 3 Flow chart of the promotion strategy optimization method provided in the embodiment of the present application (III); Figure 4 This is a functional block diagram of the promotion strategy optimization device provided in an embodiment of the present application.

[0018] In the figure: 401, information collection module; 402, feature extraction module; 403, group classification module; 404, promotion strategy formulation module. DETAILED DESCRIPTION

[0019] The present application is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are intended to enable the present application to be better understood. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, and methods. In some cases, some operations related to the present application are not shown or described in the specification, in order to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0020] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0021] The serial numbers assigned to the components herein, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).

[0022] Please refer to Figure 1 In the embodiment of the present application, a promotion strategy optimization method based on operator big data analysis is provided, which can segment customers more finely and formulate personalized marketing strategies that better meet customer needs, thereby improving the effectiveness and efficiency of marketing. The optimization method includes the following steps: Step S101: Acquire user data information through at least one data source.

[0023] The selection of data sources is diverse and targeted. It not only covers the massive databases accumulated by the operators themselves, but also extends to external social media platforms and professional third-party data platforms. By establishing a secure and reliable data connection interface with these data sources and using efficient data collection algorithms and tools, it ensures that a rich and diverse user data information can be obtained comprehensively, accurately and timely, providing a solid data foundation for subsequent in-depth analysis. This data information covers all aspects of the user's use of communication services, including but not limited to the user's basic identity characteristics, daily usage behavior patterns, details of the terminal equipment used, and activity trajectories in different geographical locations.

[0024] Step S102: extract features from the acquired user data information.

[0025] Through a series of complex and rigorous data preprocessing and feature extraction technologies, key information is extracted from complex data, and noise and redundant data are removed, so that subsequent model analysis can focus more on users' core behavior patterns and consumption tendencies, improve the accuracy and efficiency of the model, and lay the foundation for accurate user classification and personalized strategy formulation.

[0026] Step S103: construct an improved random forest model to perform data analysis and prediction on the extracted feature information, and divide the users into different groups; the groups at least include high-value users, high-potential users, potential churn users and passive users.

[0027] Based on the traditional random forest algorithm, the improved random forest model is optimized and improved according to the characteristics of operator data, such as innovations in feature selection, tree construction strategy, and model evaluation indicators. By inputting the extracted feature information into the model and using the model's powerful classification and prediction capabilities, the consumption patterns and behavior patterns hidden behind user data are deeply explored, thereby accurately dividing users into different groups with significant feature differences. Each group represents different consumption potential and market value, providing a clear target object for the subsequent formulation of personalized marketing strategies.

[0028] Step S104: formulate a personalized promotion strategy for each user group based on the analysis and prediction results of the improved random forest model.

[0029] Based on the in-depth analysis of different user groups by the model, combined with the unique needs and behavioral characteristics of each group, highly matched personalized promotion strategies are formulated. These strategies cover multiple aspects, from product package design, service discount provision to marketing channel selection and communication method customization, aiming to meet the needs of different user groups to the greatest extent, stimulate their consumption desire, improve user satisfaction and loyalty, and ultimately achieve an increase in operator market share and growth in business revenue.

[0030] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the user data information includes a first data set, a second data set and a third data set.

[0031] The first data set is data information legally and compliantly obtained from the operator's database, and the first data set at least includes the user's basic information, usage, device information and location records.

[0032] In some embodiments, the first data set obtained from the operator database directly records various detailed information of the user in the process of using the operator's communication service. The basic information of the user includes the identity identification information provided by the user when joining the network, such as name, ID number, contact information, etc., as well as basic demographic characteristics such as occupation, age, and gender that are further verified through the operator's real-name authentication system. In terms of usage, the user's call records are recorded in detail, including the start and end time of the call, call duration, call object, call type (local call, long-distance call, international call, etc.), the number and time of text messages sent and received, and the corresponding contacts, and the usage of traffic, including the total amount of traffic used, the time period distribution of traffic usage, and the category of applications using traffic. The device information covers the brand, model, operating system version, IMEI code and other unique identification information of the terminal device used by the user, as well as the hardware configuration parameters of the device, such as processor model, memory size, storage capacity, etc. This information helps to analyze the user's device's support capabilities for communication services and potential service needs. Location records use the operator's base station positioning technology and GPS positioning data to accurately record the user's geographical location coordinates at different time points, as well as the length of stay and activity trajectory at each location, providing an important basis for analyzing the user's living and working areas, travel habits, and location-based service needs.

[0033] The second data set is public information of users on social media obtained through an API interface, and the second data set at least includes social interactions and public evaluations of activity participation.

[0034] Through the API interface, data is securely connected to the social media platform, and the second data set obtained adds rich social dimension information to the user portrait. Social interaction information includes records of users' participation in various topic discussions, likes, comments, forwarding, and other behaviors on social media platforms, which records in detail the user's areas of concern, interests and hobbies, and the level of activity and influence in social networks. Activity participation and public evaluation information covers the user's participation in various online and offline activities held on social media platforms, including event registration information, behavioral data during participation, and user evaluation content and ratings of the event after the event. This information can reflect the user's consumption preferences, social behavior patterns, and attitudes and opinions on different products and services, providing strong support for operators to gain in-depth understanding of users' non-communication service-related needs and potential market opportunities.

[0035] The third data set is data information legally and compliantly obtained from a third-party data platform, and the third data set at least includes user shopping data and market research data.

[0036] Third-party data platforms gather user data from multiple fields and channels. By establishing legal data cooperation relationships with these platforms, the third-party data sets obtained further enrich the user's consumption portrait. User shopping data includes the user's purchase records on various e-commerce platforms, offline retailers and other consumption scenarios, including the name, category, brand, price, purchase time, purchase frequency and other detailed information of the purchased goods or services. These data can intuitively reflect the changing trends of users' consumption capacity, consumption preferences and consumption habits. Market research data is feedback information collected by third-party institutions through professional market research methods on users' demand, satisfaction, expected improvement direction for different products and services, as well as macro market information such as market competition situation and industry development trends. This information helps operators to stand from a broader market perspective, formulate more forward-looking and competitive promotion strategies, lay out the market in advance, and meet users' future potential needs.

[0037] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the feature extraction of the acquired user data information in step S102 specifically includes the following steps: Step S201: extract user data information through ETL, perform data cleaning, conversion and standardization on the extracted user information, and input the processed data information into the autoencoder.

[0038] The ETL (Extract, Transform, Load) process is the basic link of data processing. First, in the extraction stage, efficient data extraction tools and algorithms are used to accurately obtain the required user data information from different data sources and integrate them into a temporary data storage area. Then, in the data cleaning stage, by writing complex data cleaning rules and algorithms, duplicate records, erroneous data, missing values, and data outliers that do not conform to business logic are identified and removed to ensure the accuracy and consistency of the data. For example, for numerical data such as age and income, outliers are screened by setting a reasonable value range; for text data such as address information, address parsing and standardization tools are used to unify the format and correct errors. Then, in the data conversion stage, according to the requirements of subsequent analysis and model training, data of different formats and types are uniformly converted, such as unifying the date format to "YYYY-MM-DD", converting string-type numbers to numeric types, and encoding conversion of categorical variables. Finally, in the standardization stage, a standardization algorithm, such as Z-score standardization or Min-Max standardization, is used to map the feature values ​​of the data to a specific range, eliminating the impact of different features due to differences in dimensions and numerical ranges, so that the data can be compared and analyzed on the same scale. The data information after the above processing is input into the autoencoder to prepare for subsequent deep feature extraction.

[0039] Step S202: Use the autoencoder to extract features from the acquired user information; the extracted features include at least the user's average monthly usage time, average monthly usage volume, average monthly consumption amount, and user active time period.

[0040] The autoencoder is an unsupervised learning model based on deep learning. It automatically learns the potential feature representation in the data through the process of encoding and decoding the input data. In this process, the data after ETL processing is input into the input layer of the autoencoder. The high-dimensional original data is compressed into a low-dimensional feature vector through the nonlinear transformation of the hidden layer, and then decoded and reconstructed through the output layer to restore the original data as much as possible. By continuously adjusting the parameters of the autoencoder, the reconstruction error is minimized, so that the features learned by the hidden layer can retain the key information of the original data to the greatest extent. In addition to the basic features mentioned above, the difference characteristics of users' usage behavior in different time periods (such as weekdays and weekends, daytime and night), the fluctuation characteristics of users' consumption amount (such as the standard deviation of monthly consumption amount, coefficient of variation, etc.), and the continuity characteristics of users' use of services (such as continuous use days, number of interruptions, etc.) can also be extracted. These rich features can more comprehensively characterize users' behavior patterns and consumption characteristics, and provide more powerful support for subsequent user classification and model prediction.

[0041] Step S203: store the feature information extracted by the autoencoder.

[0042] The feature information extracted by the autoencoder is stored in a dedicated feature database, which should have efficient data storage and retrieval capabilities and be able to meet the storage requirements of large-scale feature data. During the storage process, the feature data is reasonably organized and indexed so that the required feature information can be quickly and accurately obtained during the subsequent model training and analysis process. At the same time, in order to ensure the security and integrity of the data, technical means such as data backup and encrypted storage are used to prevent data loss and leakage. In addition, a data version management mechanism can be established to record the update history and changes of feature data, so as to conduct backtracking and comparative analysis during model optimization and strategy adjustment to ensure data consistency and traceability.

[0043] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the expression encoded by the autoencoder is: ; The expression of the self-encoder decoding is: ; in, For status, For each user's data information, is the encoding weight matrix, is the encoding bias vector, is the encoding activation function, is the decoding activation function, is the decoding weight matrix, To decode the bias vector, Feature vector extracted for each user.

[0044] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred scheme, the high-value users are those who respond to promotion activities more times than the average, and the amount of promotional consumption by users is greater than the average; high-potential users are those who respond to promotion activities less than the average, and the amount of promotional consumption by users is greater than the average; potential churn users are those who respond to promotion activities less than the average, and the amount of promotional consumption by users is less than the average; and passive users are users who have never responded to promotional activities.

[0045] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the construction of an improved random forest model to perform data analysis and prediction on the extracted feature information specifically includes the following steps: Step S301, divide the data in the database into a training set and a test set, wherein 70% is used as the training set and 30% is used as the test set.

[0046] Step S302: Initialize parameters.

[0047] Step S303: randomly extract multiple sample subsets from the original data set with replacement, the size of each subset is the same as the original data set, and randomly select a part of the features from all the features as the feature set of the subset, and the number of selected features is less than the number of original features, that is: .

[0048] Step S304: For each sample subset, a decision tree is constructed using the selected feature set, which is expressed by the following formula: .

[0049] Step S305: The training forest automatically adjusts parameters according to the distribution of the reference data nodes: .

[0050] Step S306: iteratively optimize the model according to the model parameter adjustment result.

[0051] in: The parameters include the number of trees , the maximum depth of the tree , the maximum number of features used when splitting each node , the minimum number of sample splits of the tree and the maximum number of sample splits ; is a sample subset, is the original data set; To construct a decision tree function, is a set of parameters; is the error rate of each decision tree, is the error rate of the entire forest; when , readjust the parameters, and the parameter adjustment process is: ; ; in, is the re-adjusted parameter set, For a real set, is the prediction set, is the loss function, is the expected prediction error, Classify and group customer groups. To predict the effect of different promotion strategies.

[0052] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred scheme, personalized promotion strategies are formulated for each user group, including: for high-value users, pushing the first-level promotion strategy; for high-potential users, pushing the second-level promotion strategy; for potential churn users, pushing the third-level promotion strategy; for passive users, pushing the fourth-level promotion strategy; the first-level promotion strategy, the second-level promotion strategy, the third-level promotion strategy and the fourth-level promotion strategy have higher activity amounts and / or push frequencies respectively.

[0053] In some embodiments, for high-value users, a promotional activity is recommended once a month, with the activity amount greater than the average, and the high-cost call fee refund and value-added services are increased, and exclusive customer service channels and priority services are provided. The first-level promotion strategy focuses on service quality and exclusive experience, such as providing top-level network speed guarantees, including priority access to 5G networks and high-speed and stable network experience; designing ultra-large traffic packages to meet their high traffic needs; and providing high-quality value-added services, such as preferential packages for international roaming services and exclusive online entertainment platform membership rights. The push method is more precise and regular, such as one-on-one telephone communication with exclusive account managers or customized email push, to maintain their sense of dignity and strengthen close ties with operators.

[0054] In some embodiments, for high-potential users, a promotional activity is recommended once a month, and the activity amount increases in direct proportion to the user's own consumption amount; the activity launches a tiered reward plan, and when the consumption amount reaches the quota, it is accompanied by an upgrade reward. The second-level promotion strategy focuses on stimulating consumption potential, providing service packages at preferential prices, such as new business experience packages that include additional traffic, call duration, and number of SMS messages, and also giving a free trial period for newly launched value-added services to guide them to convert into high-value users. The push frequency is appropriately increased, using weekly SMS push or twice-monthly in-app message reminders to promptly inform of preferential activities and service updates.

[0055] In some embodiments, for potential lost users, a promotional activity is recommended once a week, and a limited-time discount is provided (such as extending the validity period of the existing package, giving extra traffic or call minutes) and setting up return rewards; the reasons for loss are analyzed, and a special retention plan is formulated (more discounts are given to price-sensitive users, and after-sales service guarantees are strengthened for users who are dissatisfied with the service quality). The third-level promotion strategy aims to retain users through more attractive preferential measures. For example, a substantial package upgrade discount is provided, and the upgraded package has a significant increase in traffic, call duration and number of text messages, and additional practical peripheral products or services are given. The push frequency is further increased, with SMS push every three days and telephone return visits once a week to understand the user experience and changes in demand, emphasize the limited time and uniqueness of the preferential activities, and reduce the user churn rate.

[0056] In some embodiments, for passive users, promotional activities are recommended once a quarter, and lightweight packages and low-threshold incentives (small call fee subsidies, rewards after completing simple tasks) are launched. The fourth-level promotion strategy uses more radical methods to attract their attention and activate usage behavior, such as launching limited-time ultra-low discount packages, providing basic communication services and targeted traffic for popular applications; holding highly participatory interactive promotional activities, such as online lucky draws, with generous prizes and a high probability of winning. The push frequency is once a day, and multiple channels are used to ensure that users are aware of promotional information in a timely manner and arouse their interest in participation.

[0057] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the optimization method also includes: obtaining feedback data of each user group on the current promotion strategy being pushed; evaluating the current promotion strategy based on the feedback data; and dynamically adjusting the current promotion strategy being pushed according to the evaluation results.

[0058] In the above-mentioned promotion strategy optimization method based on operator big data analysis, as a preferred solution, the feedback data of each user group on the pushed promotion strategy is obtained, including: collecting user behavior data in real time through multiple channels during the promotion activity; the multiple channels include real-time records of users using operator services, real-time feedback of users on social media, and shopping behavior data of users on third-party platforms; processing the collected real-time data; the data processing includes data cleaning, screening and conversion operations; integrating the processed real-time data with the original user data, and updating the user's feature information; using the updated user feature information, reusing the improved random forest model for data analysis and prediction, and evaluating the changes in user behavior during the promotion activity.

[0059] Please refer to Figure 4 The present application also provides a promotion strategy optimization device based on operator big data analysis, the promotion strategy optimization device comprising: An information collection module 401, the information collection module 40 is used to obtain user data information through at least one data source; A feature extraction module 402, wherein the feature extraction module 402 is used to extract features from the acquired user data information; A group division module 403, which is used to construct an improved random forest model to perform data analysis and prediction on the extracted feature information, and divide users into different groups; the groups at least include high-value users, high-potential users, potential churn users, and passive users; The promotion strategy formulation module 404 is used to formulate a personalized promotion strategy for each user group according to the analysis and prediction results of the improved random forest model.

[0060] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.

[0061] The above specific examples are used to illustrate the present application, which is only used to help understand the present application and is not intended to limit the present application. For technicians in the technical field to which the present application belongs, they can also make some simple deductions, deformations or substitutions based on the ideas of the present application.

Claims

1. A promotion strategy optimization method based on operator big data analysis, characterized in that: The optimization method comprises: Acquire user data information through at least one data source; Perform feature extraction on the acquired user data information; Constructing an improved random forest model to perform data analysis and prediction on the extracted feature information, and classifying users into different groups; the groups at least include high-value users, high-potential users, potential churn users, and passive users; According to the analysis and prediction results of the improved random forest model, a personalized promotion strategy is formulated for each user group.

2. The promotion strategy optimization method based on operator big data analysis according to claim 1, characterized in that: The user data information includes a first data set, a second data set and a third data set; The first data set is data information legally and compliantly obtained from the operator's database, and the first data set at least includes the user's basic information, usage, device information and location records; The second data set is public information of users on social media obtained through an API interface, and the second data set at least includes social interactions, activity participation and public evaluation; The third data set is data information legally and compliantly obtained from a third-party data platform, and the third data set at least includes user shopping data and market research data.

3. The promotion strategy optimization method based on operator big data analysis according to claim 1, characterized in that: The feature extraction of the acquired user data information includes: Extract user data information through ETL, clean, transform and standardize the extracted user information, and input the processed data information into the autoencoder; The autoencoder is used to extract features from the acquired user information; the extracted features include at least the user's average monthly usage time, average monthly usage volume, average monthly consumption amount, and user active time period; The feature information extracted by the autoencoder is stored.

4. The promotion strategy optimization method based on operator big data analysis according to claim 3 is characterized in that: The expression of the autoencoder encoding is: ； The expression of the self-encoder decoding is: ； in, For status, For each user's data information, is the encoding weight matrix, is the encoding bias vector, is the encoding activation function, is the decoding activation function, is the decoding weight matrix, To decode the bias vector, Feature vector extracted for each user.

5. The promotion strategy optimization method based on operator big data analysis according to claim 1, characterized in that: The high-value users are those who respond to promotional activities more times than the average and whose promotional spending amount is greater than the average; the high-potential users are those who respond to promotional activities less times than the average and whose promotional spending amount is greater than the average; the potential churn users are those who respond to promotional activities less times than the average and whose promotional spending amount is less than the average; and the passive users are those who have never responded to a promotion.

6. The promotion strategy optimization method based on operator big data analysis according to claim 1, characterized in that: The improved random forest model is constructed to perform data analysis and prediction on the extracted feature information, including: The data in the database are divided into a training set and a test set, wherein 70% is used as the training set and 30% is used as the test set; Initialize parameters; Multiple sample subsets are randomly extracted from the original data set with replacement. The size of each subset is the same as the original data set. A part of the features are randomly selected from all the features as the feature set of the subset. The number of selected features is less than the number of original features, that is: ； For each sample subset, a decision tree is constructed using the selected feature set, expressed as follows: ； The training forest automatically adjusts parameters according to the distribution of the reference data nodes: ； Iteratively optimize the model based on the model parameter adjustment results; in: The parameters include the number of trees , the maximum depth of the tree , the maximum number of features used when splitting each node , the minimum number of sample splits of the tree and the maximum number of sample splits ; is a sample subset, is the original data set; To construct a decision tree function, is a set of parameters; is the error rate of each decision tree, is the error rate of the entire forest; when , readjust the parameters, and the parameter adjustment process is: ； ； in, is the re-adjusted parameter set, For a real set, is the prediction set, is the loss function, is the expected prediction error, Classify and group customer groups. To predict the effect of different promotion strategies.

7. The promotion strategy optimization method based on operator big data analysis according to claim 1, characterized in that: The personalized promotion strategy is formulated for each user group, including: For high-value users, push the first-level promotion strategy; For high-potential users, push the second-level promotion strategy; For potential churn users, push the third-level promotion strategy; For passive users, push the fourth level promotion strategy; The first-level promotion strategy, the second-level promotion strategy, the third-level promotion strategy and the fourth-level promotion strategy have higher activity amounts and / or push frequencies respectively.

8. The promotion strategy optimization method based on operator big data analysis according to claim 1, characterized in that: The optimization method further comprises: Obtain feedback data from each user group on the current promotion strategy being pushed; Evaluate current promotional strategies based on feedback data; Dynamically adjust the current promotion strategy being pushed based on the evaluation results.

9. The promotion strategy optimization method based on operator big data analysis according to claim 8, characterized in that: The obtaining of feedback data from each user group on the pushed promotion strategy includes: During the promotional activities, user behavior data is collected in real time through multiple channels; the multiple channels include real-time records of users using operator services, real-time feedback from users on social media, and shopping behavior data of users on third-party platforms; Processing the collected real-time data; the data processing includes data cleaning, screening and conversion operations; Integrate the processed real-time data with the original user data to update the user's characteristic information; Using the updated user feature information, the improved random forest model is reused for data analysis and prediction to evaluate the changes in user behavior during promotional activities.

10. A promotion strategy optimization device based on operator big data analysis, characterized in that: The sales promotion strategy optimization device comprises: An information collection module, the information collection module is used to obtain user data information through at least one data source; A feature extraction module, wherein the feature extraction module is used to extract features from the acquired user data information; A group division module, which is used to construct an improved random forest model to perform data analysis and prediction on the extracted feature information, and divide users into different groups; the groups include at least high-value users, high-potential users, potential churn users, and passive users; A promotion strategy formulation module is used to formulate personalized promotion strategies for each user group based on the analysis and prediction results of the improved random forest model.

Images