Enterprise digital customer relationship management method and system

Abstract

The application relates to the technical field of customer relationship management, in particular to an enterprise digital customer relationship management method and system. The method comprises the following steps: obtaining attribute data and behavior data of a target user; analyzing the attribute data and the behavior data to generate a plurality of independent portrait modules; combining the independent portrait modules through a genetic algorithm to generate an initial user portrait containing a multi-branch candidate label set; obtaining real-time interaction data of the target user, pruning and optimizing the multi-branch candidate label set in the initial user portrait according to the real-time interaction data, and obtaining an accurate user portrait for customer relationship management. The method is helpful for more comprehensively analyzing user demand and formulating personalized service and marketing strategies. Through the acquisition and analysis of real-time interaction data, the initial user portrait can be updated in real time, irrelevant labels can be removed, and the most relevant labels to the latest behavior of the user can be retained, so that the accuracy and timeliness of the accurate user portrait can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of customer relationship management, and in particular to an enterprise digital customer relationship management method and system. BACKGROUND

[0002] In today's business environment, enterprise digital customer relationship management has become one of the core strategies for building long-term customer relationships. With data-driven decision-making gradually becoming the mainstream, accurate and comprehensive user portraits have become increasingly important in customer management.

[0003] Traditional user portrait construction methods rely on the collection and analysis of a large amount of data, and then form user groups through cluster analysis, and infer customer needs based on this. However, how to construct a high-precision user portrait has become an important challenge for enterprises. SUMMARY

[0004] The present application provides an enterprise digital customer relationship management method and system to solve the above problems.

[0005] In a first aspect, the present application provides an enterprise digital customer relationship management method, the method comprising:

[0006] obtaining attribute data and behavior data of a target user;

[0007] analyzing the attribute data and the behavior data to generate a plurality of independent portrait modules;

[0008] combining the independent portrait modules through a genetic algorithm to generate an initial user portrait containing a multi-branch candidate label set;

[0009] obtaining real-time interaction data of the target user, and pruning and optimizing the multi-branch candidate label set in the initial user portrait according to the real-time interaction data to obtain an accurate user portrait for customer relationship management.

[0010] By this scheme, the attribute data and behavior data of the target user are obtained, which helps to provide the basic information for constructing the initial user portrait, including the personal information and behavior record of the user, and helps to identify the user demand and preference, and formulate personalized service and marketing strategy. By analyzing the attribute data, the basic characteristics of the user are determined, such as age, gender, geographical location, etc., which helps to identify the basic attributes and background information of the user. At the same time, by analyzing the behavior data, the unique characteristics of the user are determined, such as click behavior, purchase behavior, stay time, etc., which helps to identify the behavior pattern and preference of the user. The process of combining the independent portrait modules optimized by the genetic algorithm generates the initial user portrait of the multi-branch candidate label set, reflecting the diversity and complexity of the user, which helps to more comprehensively analyze the user demand and formulate personalized service and marketing strategy. Through the acquisition and analysis of real-time interaction data, the initial user portrait is updated in time, the irrelevant labels are removed, and the labels most relevant to the latest behavior of the user are retained, which helps to improve the accuracy and timeliness of the accurate user portrait, so as to better meet the user demand, improve the user satisfaction and loyalty.

[0011] Optionally, the analysis of the attribute data and the behavior data generates a plurality of independent portrait modules, including:

[0012] analyzing the attribute data to determine the basic characteristics;

[0013] analyzing the behavior data to determine the unique characteristics;

[0014] predicting the implicit characteristics according to the basic characteristics and the unique characteristics;

[0015] determining a basic portrait module according to the basic characteristics;

[0016] determining a unique portrait module according to the unique characteristics;

[0017] determining an implicit portrait module according to the implicit characteristics;

[0018] combining the basic portrait module, the unique portrait module and the implicit portrait module as a plurality of independent portrait modules.

[0019] By identifying the basic characteristics, the basic attributes and background information of the user are identified, so as to better position and segment the user group. By determining the unique characteristics, the behavior patterns and preferences of the user are identified, so as to better predict the needs and behaviors of the user. By predicting the implicit characteristics, the inner needs and potential behaviors of the user are identified in depth, so as to develop more accurate marketing and service strategies. The basic portrait module is the basis of user portrait construction, which helps to establish the basic profile and characteristics of the user. The unique portrait module is the key to user portrait construction, which helps to distinguish the differences and personalized needs between users. The implicit portrait module helps to discover the inner needs and potential behaviors of the user, so as to develop more accurate marketing and service strategies. The basic portrait module, the unique portrait module and the implicit portrait module are used as multiple independent portrait modules, which helps to analyze and identify the user in multiple dimensions, making the construction of user portrait more flexible.

[0020] Optionally, the attribute data includes age, gender, and geographic location; and the prediction of the implicit characteristics based on the basic characteristics and the unique characteristics includes:

[0021] Analyzing the behavior data to determine click behavior, purchase behavior, and dwell time;

[0022] Analyzing the age and gender to determine mass demand;

[0023] Analyzing the geographic location, the click behavior, the purchase behavior, and the dwell time to determine potential demand;

[0024] Predicting the implicit characteristics based on the mass demand and the potential demand.

[0025] By this scheme, through the analysis of behavior data, the user's interest points, purchase preferences and participation are identified. Click behavior reflects the user's focus and interest points, purchase behavior reveals the user's consumption habits and preferences, and dwell time evaluates the user's interest and participation in content. By analyzing age and gender, the mass demand of different groups is identified. For example, young users prefer innovative products and online shopping, while older users focus on product quality and after-sales service. By analyzing data such as geographic location, click behavior, purchase behavior and dwell time, the potential demand of users in different geographic locations and behavior data is identified. For example, users in different regions have higher interest in different product categories, or users' browsing behavior in different time periods indicates upcoming purchase behavior. By predicting the implicit characteristics, the inner needs and potential behaviors of the user are identified in depth, which helps to develop more accurate marketing and service strategies, such as recommending interested products or services to users based on the prediction results.

[0026] Optionally, the combining of the independent portrait modules through the genetic algorithm generates an initial user portrait containing a multi-branch candidate label set, including:

[0027] According to the basic portrait module, an initial genetic template is determined;

[0028] Through the genetic algorithm, the unique portrait modules are analyzed to determine the co-occurrence probability of any two unique portrait modules;

[0029] According to the co-occurrence probability, the module connection between any two unique portrait modules is determined;

[0030] According to the module connection, the unique portrait modules are randomly crossed and combined to generate a mutant candidate label;

[0031] According to the implicit portrait module, a number of implicit labels are determined;

[0032] The mutant candidate label and the number of implicit labels are analyzed to determine the label conflict rate;

[0033] According to the label conflict rate, the initial genetic template is adjusted to generate an initial user portrait containing a multi-branch candidate label set.

[0034] Through this scheme, the initial genetic template provides a basic framework for the operation of the genetic algorithm, helps to maintain the core features of the user portrait, provides a starting point for the construction of the user portrait, and ensures the consistency and stability of the user portrait. The co-occurrence probability reflects the frequency of the simultaneous occurrence of different unique portrait modules. By analyzing the co-occurrence probability, the relationship between different unique portrait modules is identified. The module connection reflects the relevance and interaction between different unique portrait modules. By determining the module connection, the complexity and diversity of the user portrait are constructed, making the user portrait more comprehensive and accurate. By randomly crossing and combining the unique portrait modules, new user portrait mutant candidate labels are generated, thereby enriching the diversity of the user portrait. The implicit label reflects the user's potential needs and preferences, such as potential purchase intention, future behavior trend, etc. By determining the implicit label, the user's internal needs and potential behavior are revealed, providing deeper insights. The label conflict rate reflects the inconsistency and contradiction between the mutant candidate label and the implicit label. By analyzing the label conflict rate, conflicts and contradictions in the user portrait are identified and resolved, improving the accuracy and consistency of the user portrait. Adjusting the initial genetic template helps to optimize the construction process of the user portrait and reduce the label conflict rate. Generating an initial user portrait containing a multi-branch candidate label set helps to more comprehensively analyze customer needs and explore different user portrait combinations, thereby better meeting customer needs.

[0035] Optionally, according to the real-time interaction data, the multi-branch candidate label set in the initial user portrait is pruned and optimized to obtain an accurate user portrait, including:

[0036] extracting interaction types, interaction frequencies, and interaction depths in the real-time interaction data;

[0037] matching the interaction types with the multi-branch candidate labels to determine a set of central association labels;

[0038] calculating real-time confidence scores of the set of central association labels based on the interaction frequencies and interaction depths;

[0039] comparing the real-time confidence scores with a preset dynamic threshold, and filtering the set of central association labels according to a comparison result to obtain a qualified label;

[0040] generating the accurate user portrait according to the qualified label.

[0041] According to the present solution, by obtaining interaction types, interaction frequencies, and interaction depths, analyzing real-time behavior data of users helps to timely adjust service and marketing strategies to adapt to rapidly changing market demands. By matching, a set of central association labels associated with real-time interactions of users is determined, which helps to more accurately locate the needs of users and provide more personalized services. Real-time confidence scores help to evaluate the accuracy and relevance of the set of central association labels, and also help to identify which set of central association labels can better reflect the real needs of users. By filtering, the set of central association labels that is no longer associated with users is removed, and the set of central association labels that accurately reflects the current state of users is retained, which helps to maintain the accuracy and timeliness of the user portrait. The accurate user portrait reflects the latest behavior and preferences of users, providing a basis for more accurate market positioning and personalized services, which helps to improve customer satisfaction and loyalty and enhance market competitiveness.

[0042] Optionally, the prediction of the implicit characteristics according to the mass demand and the potential demand comprises:

[0043] analyzing the behavior data to determine a set of high-frequency behavior labels;

[0044] constructing a set of mass labels and a set of potential labels according to the mass demand and the potential demand;

[0045] calculating first set similarity and second set similarity between the set of high-frequency behavior labels and the set of mass labels and the set of potential labels, respectively;

[0046] comparing the first set similarity and the second set similarity with a preset similarity threshold, respectively, and if any one of the first set similarity and the second set similarity is less than the preset similarity threshold, extracting a time-series behavior sequence in the behavior data;

[0047] According to the time sequence behavior sequence, the coincident behaviors in the public label set and the potential label set are removed.

[0048] The removed public label set and the removed potential label set are input into a preset demand prediction model to predict the implicit characteristics.

[0049] According to the present solution, the high-frequency behavior label set helps to identify the daily behavior habits and preferences of the user, so as to better position and segment the user group. According to the public demand and the potential demand, the public label set and the potential label set are constructed, which helps to perform market segmentation and personalized marketing. By calculating the first set similarity and the second set similarity between the high-frequency behavior label set and the public label set and the potential label set, it is determined whether the behavior data of the user needs to be further analyzed. By comparing the first set similarity and the second set similarity with a preset similarity threshold, it is judged whether the behavior data of the user is consistent with the preset demand characteristics. If the first set similarity and the second set similarity are lower than the preset similarity threshold, it indicates that the behavior data of the user is inconsistent with the preset demand characteristics, and the time sequence behavior sequence of the user needs to be further analyzed. The time sequence behavior sequence in the behavior data is extracted, which helps to find the trend and change of the user behavior, so as to better identify the demand and behavior data of the user. The coincident behaviors in the public label set and the potential label set are removed, which helps to simplify the preset demand prediction model and improve the efficiency and accuracy of the preset demand prediction model. The removed public label set and the removed potential label set are input into the preset demand prediction model to predict the implicit characteristics of the user, which helps to more comprehensively identify the user and develop personalized service and marketing strategies, and improve the user satisfaction.

[0050] Optionally, the behavior data is analyzed to determine the unique characteristics, including:

[0051] The event type and the behavior timestamp in the behavior data are extracted;

[0052] According to the event type, a high-frequency behavior combination is mined through an FP-growth algorithm to determine a behavior pattern primitive;

[0053] The behavior timestamp is analyzed to calculate a time interval distribution of adjacent behaviors;

[0054] According to the time interval distribution, a behavior time characteristic is determined;

[0055] According to the behavior pattern primitive and the behavior time characteristic, the unique characteristics are determined.

[0056] By extracting the event type and behavior timestamp, the interaction behavior and time pattern of the user are identified, which helps to better analyze the behavior habit of the user. The behavior pattern base element reveals the common interaction mode of the user, which helps to identify the behavior preference of the user. The behavior timestamp is analyzed, the time interval distribution between adjacent behaviors is calculated, and the time law of the user behavior is analyzed. The time interval distribution reveals the time pattern of the user behavior, which helps to identify the behavior cycle and active time period of the user. According to the time interval distribution, the behavior time characteristics of the user, such as the active time period and the behavior cycle, are determined, which helps to analyze the behavior pattern and time preference of the user and provides a basis for personalized service. Combined with the behavior pattern base element and the behavior time characteristics, the unique characteristics of the user are determined, which reflects the personalized behavior pattern of the user and helps to more deeply identify the personalized needs of the user and provide support for customized services and marketing strategies.

[0057] Optionally, the adjusting the initial gene template according to the label conflict rate comprises:

[0058] According to the variation candidate label and the plurality of recessive labels, a conflict type is determined.

[0059] The conflict type is analyzed to determine a conflict mode.

[0060] According to the conflict mode and the label conflict rate, a template adjustment scheme is determined.

[0061] According to the template adjustment scheme, the initial gene template is adjusted.

[0062] Through the scheme, by analyzing the conflict between the variation candidate label and the plurality of recessive labels, the inconsistency between the variation candidate label and the plurality of recessive labels is identified, which provides a basis for conflict mode analysis. By analyzing the conflict type, the conflict mode is identified, which reflects the mutual relationship between the variation candidate label and the plurality of recessive labels, and provides guidance for the template adjustment scheme. According to the conflict mode and the label conflict rate, the template adjustment scheme is formulated, which improves the accuracy and consistency of the user portrait. According to the template adjustment scheme, the initial gene template is adjusted, thereby optimizing the construction process of the user portrait.

[0063] Optionally, the method further comprises:

[0064] Continuously monitoring the continuous interaction data of the target user;

[0065] Analyzing the continuous interaction data to determine update data;

[0066] According to the update data, the multi-branch candidate label set of the accurate user portrait is dynamically expanded, and the expanded multi-branch candidate label set is pruned and optimized.

[0067] By continuously monitoring the continuous interaction data of the user, the behavior pattern of the user is analyzed in real time, which helps to timely adjust the service strategy and adapt to the rapidly changing market demand. By analyzing the continuous interaction data of the user, the user portrait data that needs to be updated is identified, which helps to timely adjust the user portrait and reflect the latest behavior and preference of the user, improving the accuracy and timeliness of the user portrait. According to the updated data, the multi-branch candidate label set of the accurate user portrait is dynamically expanded, which helps to identify the diversified needs and behaviors of the user and provide support for personalized services and marketing strategies. The expanded multi-branch candidate label set is optimized using the pruning process in the genetic algorithm, and redundant or irrelevant candidate labels are removed, which helps to simplify the user portrait, reduce redundant information, and improve the efficiency and practicality of the user portrait.

[0068] In a second aspect, the present application provides an enterprise digital customer relationship management system, comprising:

[0069] A data acquisition module is configured to acquire attribute data and behavior data of a target user.

[0070] A data analysis module is configured to analyze the attribute data and the behavior data to generate a plurality of independent portrait modules.

[0071] An initial portrait generation module is configured to combine the independent portrait modules by a genetic algorithm to generate an initial user portrait comprising a multi-branch candidate label set.

[0072] A portrait optimization module is configured to acquire real-time interaction data of the target user, prune and optimize the multi-branch candidate label set in the initial user portrait according to the real-time interaction data, and obtain an accurate user portrait for customer relationship management.

[0073] Optionally, when the data analysis module analyzes the attribute data and the behavior data to generate a plurality of independent portrait modules, it is configured to:

[0074] analyze the attribute data to determine a basic feature;

[0075] analyze the behavior data to determine a unique feature;

[0076] predict a latent feature according to the basic feature and the unique feature;

[0077] determine a basic portrait module according to the basic feature;

[0078] determine a unique portrait module according to the unique feature;

[0079] determine a latent portrait module according to the latent feature;

[0080] The base image module, the unique image module and the implicit image module are taken as a plurality of independent image modules.

[0081] Optionally, the attribute data includes age, gender, geographical location; when the data analysis module predicts implicit characteristics according to the base characteristics and the unique characteristics, is used for:

[0082] Analyzing the behavior data to determine click behavior, purchase behavior and stay time;

[0083] Analyzing the age and gender to determine public demand;

[0084] Analyzing the geographical location, the click behavior, the purchase behavior and the stay time to determine potential demand;

[0085] Predicting implicit characteristics according to the public demand and the potential demand.

[0086] Optionally, when the initial image generation module combines the independent image modules by a genetic algorithm to generate an initial user image containing a multi-branch candidate label set, is used for:

[0087] Determining an initial gene template according to the base image module;

[0088] Analyzing the unique image module by a genetic algorithm to determine the co-occurrence probability of any two unique image modules;

[0089] Determining the module connection between any two unique image modules according to the co-occurrence probability;

[0090] Randomly crossing and combining the unique image modules according to the module connection to generate a variation candidate label;

[0091] Determining a plurality of implicit labels according to the implicit image module;

[0092] Analyzing the variation candidate label and the plurality of implicit labels to determine a label conflict rate;

[0093] Adjusting the initial gene template according to the label conflict rate to generate an initial user image containing a multi-branch candidate label set.

[0094] Optionally, when the image optimization module prunes and optimizes the multi-branch candidate label set in the initial user image according to the real-time interaction data to obtain an accurate user image, is used for:

[0095] Extracting the interaction type, interaction frequency and interaction depth in the real-time interaction data;

[0096] Matching the interaction type with the multi-branch candidate label to determine a centralized associated label;

[0097] calculate a real-time confidence score of the centralized associated label based on the interaction frequency and the interaction depth;

[0098] compare the real-time confidence score with a preset dynamic threshold, and filter the centralized associated label according to a comparison result to obtain a qualified label;

[0099] generate the accurate user portrait according to the qualified label.

[0100] Optionally, when predicting implicit characteristics according to the public demand and the potential demand, the data analysis module is configured to:

[0101] analyze the behavior data to determine a high-frequency behavior label set;

[0102] construct a public label set and a potential label set according to the public demand and the potential demand;

[0103] calculate a first set similarity and a second set similarity between the high-frequency behavior label set and the public label set and the potential label set respectively;

[0104] compare the first set similarity and the second set similarity with a preset similarity threshold respectively, and if any one of the first set similarity and the second set similarity is less than the preset similarity threshold, extract a time sequence behavior sequence in the behavior data;

[0105] remove coincident behaviors in the public label set and the potential label set according to the time sequence behavior sequence;

[0106] input the removed public label set and the removed potential label set into a preset demand prediction model to predict implicit characteristics.

[0107] Optionally, when analyzing the behavior data to determine unique characteristics, the data analysis module is configured to:

[0108] extract an event type and a behavior timestamp in the behavior data;

[0109] determine a behavior pattern primitive by mining a high-frequency behavior combination according to an FP-growth algorithm according to the event type;

[0110] analyze the behavior timestamp to calculate a time interval distribution of adjacent behaviors;

[0111] determine a behavior time feature according to the time interval distribution;

[0112] determine unique characteristics according to the behavior pattern primitive and the behavior time feature.

[0113] Optionally, the initial image generation module adjusts the initial gene template according to the label conflict rate, for:

[0114] According to the variation candidate label and the plurality of recessive labels, a conflict type is determined;

[0115] The conflict type is analyzed to determine a conflict mode;

[0116] According to the conflict mode and the label conflict rate, a template adjustment scheme is determined;

[0117] According to the template adjustment scheme, the initial gene template is adjusted.

[0118] Optionally, the enterprise digital customer relationship management system further comprises an image expansion module, configured to:

[0119] Continuously monitor the continuous interaction data of the target user;

[0120] The continuous interaction data is analyzed to determine update data;

[0121] According to the update data, the multi-branch candidate label set of the accurate user image is dynamically expanded, and the expanded multi-branch candidate label set is pruned and optimized. BRIEF DESCRIPTION OF DRAWINGS

[0122] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0123] Figure 1 An application scenario diagram is provided for an embodiment of the present application;

[0124] Figure 2 A flowchart of an enterprise digital customer relationship management method is provided for an embodiment of the present application;

[0125] Figure 3 An enterprise digital customer relationship management system structure diagram is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0126] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0127] In addition, the term "and / or" in this paper is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents an "or" relationship between the associated objects before and after it, unless otherwise specified.

[0128] The embodiments of the present application will be described in further detail below with reference to the drawings of the specification.

[0129] The traditional user portrait construction method relies on the collection and analysis of a large amount of data, and then forms a user group through clustering analysis, and infers customer demand based on this. However, how to construct a high-precision user portrait has become an important challenge for enterprises.

[0130] Based on this, the application provides an enterprise digital customer relationship management method and system, attribute data and behavior data of a target user are obtained, the attribute data and the behavior data are analyzed to generate a plurality of independent portrait modules, the independent portrait modules are combined through a genetic algorithm to generate an initial user portrait containing a multi-branch candidate label set, real-time interaction data of the target user is obtained, and the multi-branch candidate label set in the initial user portrait is pruned and optimized according to the real-time interaction data to obtain an accurate user portrait for customer relationship management. Obtaining attribute data and behavior data of a target user helps to provide basic information for constructing an initial user portrait, including personal information and behavior records of the user, and helps to identify user needs and preferences and develop personalized services and marketing strategies. By analyzing attribute data, the basic characteristics of the user, such as age, gender, and geographic location, are determined, which helps to identify the basic attributes and background information of the user. At the same time, by analyzing behavior data, the unique characteristics of the user, such as click behavior, purchase behavior, and dwell time, are determined, which helps to identify the behavior patterns and preferences of the user. The process of combining the independent portrait modules optimized by the genetic algorithm generates an initial user portrait containing a multi-branch candidate label set, which reflects the diversity and complexity of the user, and helps to more comprehensively analyze user needs and develop personalized services and marketing strategies. Through the acquisition and analysis of real-time interaction data, the initial user portrait is updated in real time, irrelevant labels are removed, and the most relevant labels to the latest behavior of the user are retained, which helps to improve the accuracy and timeliness of the accurate user portrait, thereby better meeting user needs and improving user satisfaction and loyalty.

[0131] Figure 1 An application scenario diagram is provided for the application, and the method provided by the application is applied in customer relationship management.

[0132] Specifically, the method provided by the application is applied in any server, the server interacts with an internal database to obtain attribute data and behavior data of a target user in the internal database, the attribute data is analyzed to determine the basic characteristics of the user, and the behavior data is analyzed to determine the unique characteristics of the user. The process of combining the independent portrait modules optimized by the genetic algorithm generates an initial user portrait containing a multi-branch candidate label set, the multi-branch candidate label set in the initial user portrait is pruned and optimized through the acquisition and analysis of real-time interaction data to obtain an accurate user portrait, and the accurate user portrait is stored in the internal database for customer relationship management. The specific implementation mode can be referred to the following embodiments.

[0133] Figure 2 A flowchart of an enterprise digital customer relationship management method is provided for an embodiment of the application, and the method of the embodiment can be applied in the server in the above scenario. As shown in the figure, Figure 2 the method includes:

[0134] S201, acquire attribute data and behavior data of a target user;

[0135] The target user can be an individual or a group who wants to identify the target user's needs, behaviors and preferences, and establish and maintain relationships with the user.

[0136] The attribute data can be data related to the target user, such as age, gender, geographic location, occupation, education background, etc.

[0137] The behavior data can be dynamic behavior data of the target user, such as click behavior, purchase behavior, browsing behavior, search behavior, interaction behavior, etc.

[0138] Specifically, the attribute data of the user is collected through registration forms, questionnaires, social media data scraping, etc. The behavior data of the user on the product, website or application is collected through website analysis tools, in-application tracking, third-party data providers, etc.

[0139] S202, analyze the attribute data and behavior data to generate a plurality of independent portrait modules;

[0140] The independent portrait module can be a module that divides the user portrait into multiple independent modules, each module representing a certain aspect of the user's characteristics, such as a basic portrait module, a unique portrait module, and a hidden portrait module.

[0141] Specifically, the collected attribute data is preprocessed to remove missing values, handle outliers, and unify data formats. The basic characteristics of the user are extracted from the processed attribute data. The collected behavior data is preprocessed to clean and standardize. The unique characteristics of the user are extracted from the processed behavior data using data mining techniques. According to the basic characteristics and unique characteristics, a plurality of independent portrait modules are generated.

[0142] S203, combine the independent portrait modules through a genetic algorithm to generate an initial user portrait containing a plurality of branch candidate label sets;

[0143] The genetic algorithm can be an optimization algorithm that simulates the natural selection and genetic process, and finds the optimal solution or approximate optimal solution through iterative search and combination operations.

[0144] The plurality of branch candidate label sets can be a plurality of branch candidate label sets generated by optimizing the module combination through the genetic algorithm, each label set containing multiple branches representing different attribute data and behavior data of the user.

[0145] The initial user portrait can be a preliminary user portrait generated by combining the independent portrait modules through the genetic algorithm.

[0146] Specifically, the construction of the user portrait is the core link of customer management. The existing user portrait construction process is: collection, processing, analysis (clustering), and labeling. This method requires a large amount of data to achieve a certain accuracy of the portrait. Another method is to map the user to a template based on basic information, and then make the portrait more comprehensive through growth. This method also requires a large amount of data to achieve a certain accuracy of the portrait.

[0147] In this embodiment, the independent portrait module is encoded as a "chromosome" in the genetic algorithm, that is, an initial portrait. A certain number of initial portraits are generated. A fitness function is defined for evaluating the quality of each initial portrait. According to the evaluation result of the fitness function, the initial user portrait with higher fitness is selected for breeding using methods such as tournament selection. The selected initial user portrait is randomly crossed to generate a new accurate user portrait. The accurate user portrait is subjected to mutation operation to simulate gene mutation. The new accurate user portrait is added to the initial user portrait to replace the user portrait with lower fitness, forming a new initial user portrait. The genetic algorithm simulates natural selection and genetic process, and iteratively optimizes the module combination to generate an initial user portrait containing a multi-branch candidate label set.

[0148] S204, obtaining real-time interaction data of the target user, pruning and optimizing the multi-branch candidate label set in the initial user portrait according to the real-time interaction data, and obtaining an accurate user portrait for customer relationship management.

[0149] The real-time interaction data can be real-time interaction records such as click behavior, purchase behavior, browsing behavior, search behavior, and interaction behavior of the target user with products, services, platforms, etc.

[0150] The accurate user portrait can be a final user portrait obtained by pruning and optimizing the multi-branch candidate label set in the initial user portrait.

[0151] Customer relationship management can be a series of strategies, activities, and measures taken to establish, maintain, and develop relationships with users.

[0152] Specifically, the Web analysis tool, in-event tracking, social media monitoring, and other methods are used to obtain real-time interaction data such as interaction type, interaction frequency, and interaction depth of the target user. According to the real-time interaction data, the multi-branch candidate label set is pruned and optimized. Based on the real-time interaction data of the user, the confidence score of each multi-branch candidate label is calculated. A dynamic threshold is set for filtering the confidence of the multi-branch candidate label. According to the filtered label, the user portrait is updated to generate an accurate user portrait, thereby performing customer relationship management.

[0153] By this scheme, the attribute data and behavior data of the target user are obtained, which helps to provide the basic information for constructing the initial user portrait, including the personal information and behavior record of the user, and helps to identify the user's needs and preferences, and develop personalized services and marketing strategies. By analyzing the attribute data, the basic characteristics of the user are determined, such as age, gender, geographic location, etc., which helps to identify the basic attributes and background information of the user. At the same time, by analyzing the behavior data, the unique characteristics of the user are determined, such as click behavior, purchase behavior, dwell time, etc., which helps to identify the behavior patterns and preferences of the user. The process of genetic algorithm optimization of independent portrait module for combination generates the initial user portrait of the multi-branch candidate label set, reflecting the diversity and complexity of the user, which helps to more comprehensively analyze the user's needs and develop personalized services and marketing strategies. Through the acquisition and analysis of real-time interaction data, the initial user portrait is updated in time, irrelevant labels are removed, and the most relevant labels to the user's latest behavior are retained, which helps to improve the accuracy and timeliness of the accurate user portrait, so as to better meet the user's needs and improve the user's satisfaction and loyalty.

[0154] In some embodiments, the attribute data is analyzed to determine the basic characteristics, the behavior data is analyzed to determine the unique characteristics, the implicit characteristics are predicted according to the basic characteristics and the unique characteristics, the basic portrait module is determined according to the basic characteristics, the unique portrait module is determined according to the unique characteristics, and the implicit portrait module is determined according to the implicit characteristics. The basic portrait module, the unique portrait module, and the implicit portrait module are used as the multiple independent portrait modules.

[0155] The basic characteristics can be static attributes of the user, such as age, gender, geographic location, education level, occupation, income level, etc.

[0156] The unique characteristics can be dynamic behavior characteristics of the user, such as click behavior, purchase history, browsing record, search record, interaction behavior, etc.

[0157] The implicit characteristics can be potential and not easily directly observed characteristics of the user.

[0158] The basic portrait module can be the basic part of the user portrait.

[0159] The unique portrait module can be the unique part of the user portrait.

[0160] The implicit portrait module can be the potential part of the user portrait.

[0161] Specifically, the collected attribute data is statistically analyzed to identify basic characteristics representing the user. The user's behavior data is analyzed to identify unique characteristics representing the user. The basic characteristics and unique characteristics are combined to use a preset demand prediction model to predict the user's implicit characteristics. A basic portrait module is created according to the basic characteristics. A unique portrait module is created according to the unique characteristics. An implicit portrait module is created according to the implicit characteristics. The basic portrait module, the unique portrait module, and the implicit portrait module are integrated together to form multiple independent portrait modules.

[0162] Through the scheme, identifying basic characteristics helps to identify the basic attributes and background information of the user, so as to better position and segment the user group. Determining unique characteristics helps to identify the behavior patterns and preferences of the user, so as to better predict the needs and behaviors of the user. Predicting the implicit characteristics of the user helps to better identify the inner needs and potential behaviors of the user, so as to develop more accurate marketing and service strategies. The basic portrait module is the basis of user portrait construction, which helps to establish the basic outline and characteristics of the user. The unique portrait module is the key to user portrait construction, which helps to distinguish the differences and personalized needs between users. The implicit portrait module helps to discover the inner needs and potential behaviors of the user, so as to develop more accurate marketing and service strategies. Integrating the basic portrait module, the unique portrait module, and the implicit portrait module as multiple independent portrait modules helps to analyze and identify the user in multiple dimensions, making the construction of user portrait more flexible.

[0163] In some embodiments, the behavior data is analyzed to determine click behavior, purchase behavior, and dwell time; age and gender are analyzed to determine mass demand; geographic location, click behavior, purchase behavior, and dwell time are analyzed to determine potential demand; and implicit characteristics are predicted based on mass demand and potential demand.

[0164] Click behavior can be a click operation such as clicking a link, button, picture, etc. on a website, application, or advertisement.

[0165] Purchase behavior can be the behavior of a user purchasing goods or services.

[0166] Dwell time can be the length of time spent on a webpage, application page, or content.

[0167] Mass demand can be a demand or preference that is generally present in a group.

[0168] Geographic location can be the physical location of the user.

[0169] Potential demand can be a demand that has not been explicitly expressed or satisfied by the user.

[0170] Specifically, according to the behavior data, the click behavior, the purchase behavior, and the stay duration on different pages or contents of the user are analyzed using data mining techniques. The age and gender of the user are analyzed using statistical analysis methods, so as to determine the public demand. In combination with the geographic location, the click behavior, the purchase behavior, and the stay duration of the user, the sequence pattern mining technique is used for analysis, so as to determine the potential demand of the user. The implicit characteristics of the user are predicted using a preset demand prediction model.

[0171] Through the scheme, the interest points, the purchase preferences, and the participation of the user are identified through analysis of the behavior data. The click behavior reflects the attention points and the interest points of the user, the purchase behavior reveals the consumption habits and the preferences of the user, and the stay duration evaluates the interest and the participation of the user in the content. Through analysis of the age and the gender, the public demand of different groups is identified. For example, young users prefer innovative products and online shopping, while older users pay more attention to product quality and after-sales service. Through analysis of the geographic location, the click behavior, the purchase behavior, and the stay duration, the potential demand of the user under different geographic locations and behavior data is identified. For example, users in different regions have higher interest in different product categories, or the browsing behavior of the user in different time periods indicates upcoming purchase behavior. Through prediction of the implicit characteristics, the innermost needs and the potential behavior of the user are more deeply identified, which helps to develop more accurate marketing and service strategies, for example, the user is recommended with interested products or services according to the prediction results.

[0172] In some embodiments, according to the basic portrait module, an initial gene template is determined; through a genetic algorithm, the unique portrait module is analyzed, and the co-occurrence probability of any two unique portrait modules is determined; according to the co-occurrence probability, the module connection between any two unique portrait modules is determined; according to the module connection, the unique portrait modules are randomly crossed and combined to generate a variation candidate label; according to the implicit portrait module, a plurality of implicit labels are determined; the variation candidate label and the plurality of implicit labels are analyzed to determine a label conflict rate; according to the label conflict rate, the initial gene template is adjusted to generate an initial user portrait containing a plurality of branch candidate label sets.

[0173] The initial gene template can be a preset framework for representing the basic characteristics of the user portrait.

[0174] The co-occurrence probability can be the frequency of simultaneous occurrence of any two unique portrait modules.

[0175] The module connection can be the correlation and interaction between different unique portrait modules.

[0176] Random crossing can be, in a genetic algorithm, exchanging part of the genes of two unique portrait modules to generate a new unique portrait module.

[0177] The variation candidate label can be a new label generated by exchanging the genes of the unique portrait module in the random cross process.

[0178] The implicit label can be a demand not explicitly expressed or not fully met in the user portrait.

[0179] The label conflict rate can be the degree of inconsistency and contradiction between the variation candidate label and the implicit label in the user portrait.

[0180] Specifically, according to the basic portrait module, an initial gene template is constructed. The unique portrait module is encoded as a chromosome in the genetic algorithm, that is, a user portrait. Through the iterative process of the genetic algorithm, the user portrait is continuously optimized, thereby determining the co-occurrence probability of any two unique portrait modules. The association rule learning algorithm is used to analyze the co-occurrence probability to determine the association rules between different unique portrait modules. According to the co-occurrence probability and the association rule learning result, the module connection between any two unique portrait modules is determined. According to the module connection, the unique portrait module with higher fitness is selected for cross combination. The selected unique portrait module is subjected to cross operation to exchange part of the features. The user portrait generated by the cross is subjected to mutation operation to randomly change part of the features. Through the cross and mutation operations, a new user portrait variation candidate label is generated. According to the implicit portrait module, a plurality of implicit labels are defined. The variation candidate label and the implicit label are aligned, the label conflict is identified using a statistical method, and the label conflict rate is calculated. According to the result of the label conflict rate, the initial gene template is adjusted. Using the adjusted initial gene template, the cross and mutation operations of the genetic algorithm are used to generate an initial user portrait containing a multi-branch candidate label set.

[0181] By this scheme, the initial gene template provides a basic framework for the genetic algorithm operation, helps to maintain the core features of the user portrait, provides a starting point for the construction of the user portrait, and ensures the consistency and stability of the user portrait. The co-occurrence probability reflects the frequency of the simultaneous occurrence of different unique portrait modules. By analyzing the co-occurrence probability, the relationship between different unique portrait modules is identified. The module connection reflects the relevance and interaction between different unique portrait modules. By determining the module connection, the complexity and diversity of the user portrait are constructed, making the user portrait more comprehensive and accurate. Through the random cross combination of unique portrait modules, new user portrait variation candidate labels are generated, thereby enriching the diversity of the user portrait. The implicit label reflects the user's potential needs and preferences, such as potential purchase intention, future behavior trend, etc. By determining the implicit label, the user's internal needs and potential behavior are revealed, providing deeper insights. The label conflict rate reflects the inconsistency and contradiction between the variation candidate label and the implicit label. By analyzing the label conflict rate, conflicts and contradictions in the user portrait are identified and resolved, improving the accuracy and consistency of the user portrait. Adjusting the initial gene template helps to optimize the construction process of the user portrait and reduce the label conflict rate. Generating the initial user portrait of the multi-branch candidate label set helps to more comprehensively analyze customer needs and explore different user portrait combinations, thereby better meeting customer needs.

[0182] In some embodiments, the interaction type, interaction frequency, and interaction depth in the real-time interaction data are extracted; the interaction type is matched with the multi-branch candidate label to determine the central association label; the real-time confidence score of the central association label is calculated based on the interaction frequency and the interaction depth; the real-time confidence score is compared with the preset dynamic threshold, and the central association label is filtered according to the comparison result to obtain the qualified label; and the accurate user portrait is generated according to the qualified label.

[0183] The interaction type can be the way the user interacts with the product, service, or platform.

[0184] The interaction frequency can be the number of times the user performs the interaction type within a certain time.

[0185] The interaction depth can be the user's input on the interaction type.

[0186] The central association label can be the label most relevant to the user's real-time interaction.

[0187] The real-time confidence score can be the confidence score of the central association label calculated based on the interaction frequency and the interaction depth.

[0188] The preset dynamic threshold can be a dynamic threshold for filtering the central association label according to business needs and behavior data. It is pre-stored in the server and called when used.

[0189] The target label can be a label screened according to a comparison result of the real-time confidence score and a preset dynamic threshold.

[0190] Specifically, real-time data stream processing technology is used to capture and analyze the interaction type, interaction frequency and interaction depth in real-time interaction data. A matching algorithm is used to match the interaction type in the real-time interaction data with a predefined multi-branch candidate label set. Through the matching result, the set of associated labels in real-time interaction with the user is determined. According to the interaction frequency and the interaction depth, a statistical analysis method is used to calculate the real-time confidence score of the set of associated labels. A dynamic threshold is set, which is adjusted according to the change of real-time interaction data. The real-time confidence score is compared with the preset dynamic threshold, and the label with a score higher than the preset dynamic threshold is screened out, i.e. the target label. According to the screened target label, the user portrait is updated, and an accurate user portrait is generated.

[0191] Through the scheme, by obtaining the interaction type, interaction frequency and interaction depth, analyzing the real-time behavior data of the user, it is helpful to timely adjust the service and marketing strategy to adapt to the rapidly changing market demand. Through matching, the set of associated labels in real-time interaction with the user is determined, which is helpful to more accurately locate the needs of the user and provide more personalized services. The real-time confidence score is helpful to evaluate the accuracy and relevance of the set of associated labels, and at the same time to identify which set of associated labels can better reflect the real needs of the user. Through screening, the set of associated labels that is no longer associated with the user is removed, and the set of associated labels that accurately reflects the current state of the user is retained, which is helpful to maintain the accuracy and timeliness of the user portrait. The accurate user portrait reflects the latest behavior and preference of the user, providing a basis for more accurate market positioning and personalized services, which is helpful to improve customer satisfaction and loyalty and enhance market competitiveness.

[0192] In some embodiments, the behavior data is analyzed to determine a high-frequency behavior label set; a popular label set and a potential label set are constructed according to popular needs and potential needs; a first set similarity and a second set similarity of the high-frequency behavior label set with the popular label set and the potential label set are calculated; the first set similarity and the second set similarity are compared with a preset similarity threshold, and if any one of the first set similarity and the second set similarity is less than the preset similarity threshold, a time sequence behavior sequence in the behavior data is extracted; according to the time sequence behavior sequence, overlapping behaviors in the popular label set and the potential label set are removed; the removed popular label set and the removed potential label set are input into a preset demand prediction model to predict implicit characteristics.

[0193] The high-frequency behavior label set can be a set of behavior data labels most commonly seen by the user identified by analyzing the behavior data.

[0194] The public tag set can be a set of tags representing the needs of the public, derived from market research and behavioral data analysis.

[0195] The potential tag set can be a set of tags representing the potential needs of the user, inferred by analyzing behavioral data and market trends.

[0196] The first set similarity can be the similarity between the high-frequency behavior tag set and the public tag set.

[0197] The second set similarity can be the similarity between the high-frequency behavior tag set and the potential tag set.

[0198] The preset similarity threshold can be a preset similarity threshold for determining whether the similarity between the high-frequency behavior tag set and the public tag set or the potential tag set is high enough. It is pre-stored in the server and called when used.

[0199] The time-series behavior sequence can be a sequence of behaviors of the user within a time period.

[0200] The coincident behavior can be a behavior that exists in both the public tag set and the potential tag set.

[0201] The preset demand prediction model can be a model pre-constructed for predicting the potential needs of the user. It is pre-stored in the server and called when used.

[0202] Specifically, the behavior data of the user is analyzed using data mining techniques to identify high-frequency behaviors and create a high-frequency behavior tag set. According to the public needs and potential needs, a public tag set and a potential tag set are constructed. The first set similarity between the high-frequency behavior tag set and the public tag set is calculated using a similarity calculation method. The second set similarity between the high-frequency behavior tag set and the potential tag set is calculated using a similarity calculation method. The calculated first set similarity and second set similarity are compared with the preset similarity threshold. If either the first set similarity or the second set similarity is less than the preset similarity threshold, the time-series behavior sequence in the behavior data of the user is extracted. The time-series behavior sequence is analyzed using time series analysis techniques. According to the analysis results of the time-series behavior sequence, the coincident behavior with the public tag set and the potential tag set is identified. According to the identified coincident behavior, the data cleaning technique is used to remove the coincident behavior in the public tag set and the potential tag set. The removed public tag set and potential tag set are input into the preset demand prediction model, and the machine learning algorithm is used for prediction. According to the prediction results, the implicit characteristics are identified.

[0203] By the scheme, the high-frequency behavior tag set helps to identify the daily behavior habits and preferences of the user, so as to better position and segment the user group. According to the mass demand and potential demand, the mass tag set and the potential tag set are constructed, which helps to carry out market segmentation and personalized marketing. By calculating the first set similarity and the second set similarity of the high-frequency behavior tag set and the mass tag set and the potential tag set, it is determined whether the behavior data of the user needs to be further analyzed. By comparing the first set similarity and the second set similarity with the preset similarity threshold, it is judged whether the behavior data of the user is consistent with the preset demand characteristics. If the first set similarity and the second set similarity are lower than the preset similarity threshold, it indicates that the behavior data of the user is inconsistent with the preset demand characteristics, and the time series behavior sequence of the user needs to be further analyzed. The time series behavior sequence in the behavior data is extracted, which helps to find the trend and change of the user behavior, so as to better identify the demand and behavior data of the user. The coincident behaviors in the mass tag set and the potential tag set are removed, which helps to simplify the preset demand prediction model and improve the efficiency and accuracy of the preset demand prediction model. The mass tag set and the potential tag set after removal are input into the preset demand prediction model, and the implicit characteristics of the user are predicted, which helps to more comprehensively identify the user and develop personalized service and marketing strategies to improve user satisfaction.

[0204] In some embodiments, the event type and the behavior timestamp in the behavior data are extracted; according to the event type, the high-frequency behavior combination is mined through the FP-growth algorithm to determine the behavior pattern primitive; the behavior timestamp is analyzed to calculate the time interval distribution of adjacent behaviors; according to the time interval distribution, the behavior time feature is determined; according to the behavior pattern primitive and the behavior time feature, the unique characteristics are determined.

[0205] The event type can be a specific event of the user on the platform.

[0206] The behavior timestamp can be a specific time point at which the user performs part of the behavior.

[0207] The FP-growth algorithm can be a frequent item set mining algorithm for efficiently discovering frequent item sets and association rules from a large amount of behavior data.

[0208] The high-frequency behavior combination can be a frequently occurring user behavior combination identified by the FP-growth algorithm.

[0209] The behavior pattern primitive can be a basic unit of the user behavior pattern.

[0210] The adjacent behaviors can be two or more user behaviors that occur continuously in a time series.

[0211] The time interval distribution can be the distribution of the time interval between adjacent behaviors.

[0212] The behavior time feature can be an active period, a behavior cycle, a response time, and the like of the user behavior.

[0213] Specifically, event types and behavior timestamps are extracted from the behavior data using data extraction techniques. The FP-growth algorithm is used to mine frequent item sets for the event types, and high-frequency behavior combinations are identified. According to the high-frequency behavior combinations, behavior pattern primitives are determined. According to the behavior timestamps, the time difference between adjacent behaviors in the time-series behavior sequence is calculated. The time interval data of the user group is summarized to construct a time interval distribution. The time interval distribution is analyzed to identify the behavior time features of the user. The unique characteristics of the user are determined in combination with the behavior pattern primitives and the behavior time features.

[0214] Through the scheme, by extracting event types and behavior timestamps, the interactive behavior and time pattern of the user are identified, which helps to better analyze the behavior habits of the user. The behavior pattern primitives reveal the common interactive patterns of the user, which helps to identify the behavior preferences of the user. The behavior timestamps are analyzed, the time interval distribution between adjacent behaviors is calculated, and the time regularity of the user behavior is analyzed. The time interval distribution reveals the time pattern of the user behavior, which helps to identify the behavior cycle and active period of the user. According to the time interval distribution, the behavior time features of the user, such as the active period and the behavior cycle, are determined, which helps to analyze the behavior pattern and time preference of the user, and provides a basis for personalized services. In combination with the behavior pattern primitives and the behavior time features, the unique characteristics of the user are determined, which reflects the personalized behavior pattern of the user, and helps to more deeply identify the personalized needs of the user, and provides support for customized services and marketing strategies.

[0215] In some embodiments, a conflict type is determined according to the variation candidate label and the plurality of implicit labels; a conflict pattern is determined by analyzing the conflict type; a template adjustment scheme is determined according to the conflict pattern and a label conflict rate; and the initial gene template is adjusted according to the template adjustment scheme.

[0216] The conflict type can be an inconsistent situation between the variation candidate label and the plurality of implicit labels in the process of constructing the user portrait.

[0217] The conflict pattern can be a co-occurrence pattern, a dependency relationship, a competition relationship, and the like presented by the conflict type in the process of constructing the user portrait.

[0218] The template adjustment scheme can be an adjustment scheme for identifying conflicts in the user portrait, and adjusting the template for constructing the user portrait.

[0219] Specifically, the mutation candidate label generated by the genetic algorithm and the implicit label of the user are analyzed. According to the analysis result between the mutation candidate label and the several implicit labels, the conflict situation is determined, and the conflict type is identified. The conflict type is analyzed using statistical analysis techniques, and according to the analysis result, the conflict mode is determined. The identified conflict mode is analyzed, and the label conflict rate, i.e. the frequency and severity of conflicts between the mutation candidate label and the several implicit labels, is analyzed. According to the analysis result of the conflict mode and the label conflict rate, the template adjustment strategy is formulated. According to the template adjustment scheme, the initial genetic template is adjusted.

[0220] Through the scheme, by analyzing the conflicts between the mutation candidate label and the several implicit labels, the inconsistency between the mutation candidate label and the several implicit labels is identified, which provides a basis for conflict mode analysis. By analyzing the conflict type, the conflict mode is identified, reflecting the mutual relationship between the mutation candidate label and the several implicit labels, and providing guidance for the template adjustment scheme. According to the conflict mode and the label conflict rate, the template adjustment scheme is formulated, which improves the accuracy and consistency of the user portrait. According to the template adjustment scheme, the initial genetic template is adjusted, thereby optimizing the construction process of the user portrait.

[0221] In some embodiments, the continuous interaction data of the target user is continuously monitored; the continuous interaction data is analyzed to determine update data; and the multi-branch candidate label set of the accurate user portrait is dynamically expanded according to the update data, and the expanded multi-branch candidate label set is pruned and optimized.

[0222] The continuous interaction data can be data generated by the continuous interaction behavior of the user on the platform.

[0223] The update data can be data analyzed from the continuous interaction data and needed to be updated to the user portrait.

[0224] Specifically, the continuous interaction data of the user is continuously monitored using real-time data stream processing technology. The continuous interaction data is deeply analyzed using data mining and analysis techniques to determine the update data. According to the update data, the multi-branch candidate label set of the accurate user portrait is dynamically expanded, and the expanded multi-branch candidate label set is optimized using the pruning process in the genetic algorithm.

[0225] This solution utilizes continuous monitoring of user interaction data and real-time analysis of user behavior patterns to facilitate timely adjustments to service strategies and adapt to rapidly changing market demands. By analyzing continuous user interaction data, it identifies user profiles requiring updates, enabling timely adjustments to these profiles to reflect the latest user behaviors and preferences, thus improving their accuracy and timeliness. Based on updated data, the solution dynamically expands the multi-branch candidate tag set for precise user profiles, helping to identify diverse user needs and behaviors and supporting personalized services and marketing strategies. Optimizing the expanded multi-branch candidate tag set using a pruning process within a genetic algorithm removes redundant or irrelevant candidate tags, simplifying user profiles, reducing redundant information, and improving the efficiency and usability of user profiles.

[0226] Figure 3 A schematic diagram of the structure of an enterprise digital customer relationship management system provided in one embodiment of this application is shown below. Figure 3 As shown, the enterprise digital customer relationship management system 300 of this embodiment includes: a data acquisition module 301, a data analysis module 302, an initial profile generation module 303, and a profile optimization module 304.

[0227] Data acquisition module 301 is used to acquire attribute data and behavioral data of the target user;

[0228] Data analysis module 302 is used to analyze the attribute data and the behavioral data to generate multiple independent profile modules;

[0229] The initial profile generation module 303 is used to combine the independent profile modules through a genetic algorithm to generate an initial user profile containing a multi-branch candidate tag set.

[0230] The profile optimization module 304 is used to acquire the real-time interaction data of the target user, and to perform pruning and optimization on the multi-branch candidate tag set in the initial user profile based on the real-time interaction data to obtain an accurate user profile for customer relationship management.

[0231] Optionally, when the data analysis module 302 analyzes the attribute data and the behavioral data to generate multiple independent profile modules, it is used for:

[0232] Analyze the attribute data to determine the basic characteristics;

[0233] Analyze the behavioral data to determine unique characteristics;

[0234] Based on the aforementioned basic characteristics and unique characteristics, predict latent characteristics;

[0235] Based on the aforementioned fundamental characteristics, the basic profile module is determined;

[0236] According to the unique characteristics, determine a unique image module;

[0237] According to the implicit characteristics, determine an implicit image module;

[0238] The basic image module, the unique image module, and the implicit image module are used as a plurality of independent image modules.

[0239] Optionally, the attribute data includes age, gender, and geographical location; when the data analysis module 302 predicts implicit characteristics according to the basic characteristics and the unique characteristics, it is used for:

[0240] Analyzing the behavior data to determine click behavior, purchase behavior, and stay duration;

[0241] Analyzing the age and gender to determine public demand;

[0242] Analyzing the geographical location, the click behavior, the purchase behavior, and the stay duration to determine potential demand;

[0243] According to the public demand and the potential demand, predict the implicit characteristics.

[0244] Optionally, when the initial image generation module 303 combines the independent image modules by a genetic algorithm to generate an initial user image containing a multi-branch candidate label set, it is used for:

[0245] According to the basic image module, determine an initial gene template;

[0246] By a genetic algorithm, analyze the unique image module to determine the co-occurrence probability of any two unique image modules;

[0247] According to the co-occurrence probability, determine the module connection between any two unique image modules;

[0248] According to the module connection, randomly cross-combine the unique image modules to generate a variation candidate label;

[0249] According to the implicit image module, determine a plurality of implicit labels;

[0250] Analyze the variation candidate label and the plurality of implicit labels to determine a label conflict rate;

[0251] According to the label conflict rate, adjust the initial gene template to generate an initial user image containing a multi-branch candidate label set.

[0252] Optionally, when the image optimization module 304 prunes and optimizes the multi-branch candidate label set in the initial user image according to the real-time interaction data to obtain an accurate user image, it is used for:

[0253] extracting an interaction type, an interaction frequency and an interaction depth in the real-time interaction data;

[0254] matching the interaction type with the multi-branch candidate label to determine a centralized associated label;

[0255] calculating a real-time confidence score of the centralized associated label based on the interaction frequency and the interaction depth;

[0256] comparing the real-time confidence score with a preset dynamic threshold, and filtering the centralized associated label according to a comparison result to obtain a qualified label;

[0257] generating the accurate user portrait according to the qualified label.

[0258] Optionally, when predicting implicit characteristics according to the public demand and the potential demand, the data analysis module 302 is configured to:

[0259] analyzing the behavior data to determine a high-frequency behavior label set;

[0260] constructing a public label set and a potential label set according to the public demand and the potential demand;

[0261] calculating a first set similarity and a second set similarity between the high-frequency behavior label set and the public label set and the potential label set respectively;

[0262] comparing the first set similarity and the second set similarity with a preset similarity threshold respectively, and if any one of the first set similarity and the second set similarity is less than the preset similarity threshold, extracting a time sequence behavior sequence in the behavior data;

[0263] according to the time sequence behavior sequence, eliminating coincident behaviors in the public label set and the potential label set;

[0264] inputting the eliminated public label set and the eliminated potential label set into a preset demand prediction model to predict implicit characteristics.

[0265] Optionally, when analyzing the behavior data to determine unique characteristics, the data analysis module 302 is configured to:

[0266] extracting an event type and a behavior timestamp in the behavior data;

[0267] determining a behavior pattern primitive by mining a high-frequency behavior combination according to the event type through an FP-growth algorithm;

[0268] analyzing the behavior timestamp to calculate a time interval distribution of adjacent behaviors;

[0269] determine a behavior time feature according to the time interval distribution;

[0270] determine a unique feature according to the behavior pattern primitive and the behavior time feature.

[0271] Optionally, the initial image generation module 303 adjusts the initial gene template according to the label conflict rate, for:

[0272] determine a conflict type according to the mutation candidate label and the plurality of recessive labels;

[0273] analyze the conflict type to determine a conflict mode;

[0274] determine a template adjustment scheme according to the conflict mode and the label conflict rate;

[0275] adjust the initial gene template according to the template adjustment scheme.

[0276] Optionally, the enterprise digital customer relationship management system further comprises an image expansion module 305, configured to:

[0277] continuously monitor the continuous interaction data of the target user;

[0278] analyze the continuous interaction data to determine update data;

[0279] dynamically expand the multi-branch candidate label set of the accurate user image according to the update data, and prune and optimize the expanded multi-branch candidate label set.

[0280] The system of the embodiment can be used to execute the method of any of the above embodiments, and has similar implementation principles and technical effects, which will not be described here.

Claims

1. A method for enterprise digital customer relationship management, characterized in that, include: Obtain the target user's attribute and behavioral data; Analyzing the attribute data and the behavioral data, multiple independent profile modules are generated, including: Analyze the attribute data to determine the basic characteristics; Analyze the behavioral data to determine unique characteristics, including: Extract the event type and behavior timestamp from the behavioral data; Based on the event type, the FP-growth algorithm is used to mine high-frequency behavior combinations and determine the behavior pattern primitives. Analyze the timestamps of the actions and calculate the time interval distribution between adjacent actions; Based on the time interval distribution, determine the behavioral temporal characteristics; Based on the behavioral pattern primitives and the behavioral temporal characteristics, unique features are determined; Based on the aforementioned basic characteristics and unique characteristics, latent characteristics are predicted, including: Analyze the behavioral data to determine a set of high-frequency behavioral tags; Based on public demand and potential demand, construct a set of public tags and a set of potential tags; Calculate the similarity between the high-frequency behavior tag set and the first set and the second set of the mass tag set and the potential tag set, respectively; The similarity of the first set and the similarity of the second set are compared with a preset similarity threshold. If either the similarity of the first set or the similarity of the second set is less than the preset similarity threshold, the temporal behavior sequence in the behavior data is extracted. Based on the time-series behavior sequence, overlapping behaviors in the public tag set and the potential tag set are eliminated; The set of public labels after removal and the set of potential labels after removal are input into the preset demand prediction model to predict latent characteristics; Based on the aforementioned fundamental characteristics, the basic profile module is determined; Based on the aforementioned unique characteristics, a unique profile module is determined; Based on the aforementioned implicit characteristics, the implicit profiling module is determined; The basic portrait module, the unique portrait module, and the implicit portrait module are treated as multiple independent portrait modules; The independent profiling modules are combined using a genetic algorithm to generate an initial user profile containing a multi-branch candidate tag set, including: Based on the aforementioned basic profiling module, an initial gene template is determined; The unique profile modules are analyzed using a genetic algorithm to determine the co-occurrence probability of any two unique profile modules. Based on the co-occurrence probability, determine the module relationship between any two unique profile modules; Based on the module relationships, the unique profile modules are randomly cross-combined to generate variant candidate tags; Based on the implicit profiling module, several implicit tags are determined; Analyze the candidate variant labels and the plurality of latent labels to determine the label conflict rate; Based on the tag conflict rate, the initial gene template is adjusted to generate an initial user profile containing a multi-branch candidate tag set; The system acquires real-time interaction data of the target user, and prunes and optimizes the multi-branch candidate tag set in the initial user profile based on the real-time interaction data to obtain an accurate user profile for customer relationship management.

2. The method according to claim 1, characterized in that, The attribute data includes age, gender, and geographic location; the prediction of latent traits based on the basic characteristics and the unique characteristics includes: Analyze the behavioral data to determine click behavior, purchase behavior, and dwell time; Analyze the stated age and gender to determine the needs of the general public; Analyze the geographic location, click behavior, purchase behavior, and dwell time to determine potential demand; Based on the stated public demand and the stated potential demand, predict latent characteristics.

3. The method according to claim 1, characterized in that, The step of pruning and optimizing the multi-branch candidate tag set in the initial user profile based on the real-time interaction data to obtain an accurate user profile includes: Extract the interaction type, interaction frequency, and interaction depth from the real-time interaction data; The interaction type is matched with the multi-branch candidate tags to determine the centrally associated tags; Based on the interaction frequency and interaction depth, calculate the real-time confidence score of the centralized associated tags; The real-time confidence score is compared with a preset dynamic threshold, and the clustered associated tags are filtered based on the comparison results to obtain qualified tags; The precise user profile is generated based on the qualified tags.

4. The method according to claim 3, characterized in that, The step of adjusting the initial gene template according to the tag conflict rate includes: Based on the candidate variant labels and the plurality of latent labels, determine the conflict type; Analyze the conflict types to determine the conflict patterns; Based on the conflict mode and the tag conflict rate, a template adjustment scheme is determined; The initial gene template is adjusted according to the template adjustment scheme.

5. The method according to claim 1, characterized in that, The method further includes: Continuously monitor the target user's ongoing interaction data; Analyze the continuous interaction data to determine the data to be updated; Based on the updated data, the multi-branch candidate tag set of the accurate user profile is dynamically expanded, and the expanded multi-branch candidate tag set is pruned and optimized.

6. A digital customer relationship management system for enterprises, characterized in that, The method applied to any one of claims 1-5 includes: The data acquisition module is responsible for acquiring the attribute and behavioral data of the target user. The data analysis module is used to analyze the attribute data and the behavioral data to generate multiple independent profile modules; The initial profile generation module is used to combine the independent profile modules using a genetic algorithm to generate an initial user profile containing a multi-branch candidate tag set. The profile optimization module is used to acquire real-time interaction data of the target user, and prune and optimize the multi-branch candidate tag set in the initial user profile based on the real-time interaction data to obtain an accurate user profile for customer relationship management.

Images