Customer relationship management data analysis method and system based on artificial intelligence

Abstract

The present invention relates to the field of data analysis technology, and specifically to a customer relationship management data analysis method and system based on artificial intelligence, comprising the following steps: step one, collecting customer data; step two, storing the data; step three, analyzing and mining the data; step four, building a customer portrait; step five, providing intelligent decision support for customer relationship management based on the customer portrait and data analysis results; the customer relationship management data analysis method and system based on artificial intelligence of the present invention can significantly improve the customer relationship management level of an enterprise. Through artificial intelligence deep learning, the system can automatically analyze a customer's massive conversation records, social media interactions, and transaction data, quickly identify customer needs and preferences, and then generate personalized customer portraits and behavior analysis reports, to help enterprises accurately locate potential customer groups, optimize service strategies and marketing plans, and thus improve customer satisfaction and loyalty.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to an artificial intelligence-based customer relationship management data analysis method and system. Background Art

[0002] In today's digital age, the business operating environment has undergone profound changes, and customer relationship management faces unprecedented challenges and opportunities. With the rapid development of information technology, companies can obtain massive amounts of customer data from a wide range of sources, including internal sales transaction records, customer service interaction logs, marketing campaign feedback, as well as external social media platform data, online browsing behavior tracking data, etc. Traditional customer relationship management methods are unable to cope with such large and complex data.

[0003] Traditional customer relationship management mainly relies on manual analysis and simple statistical tools, and it is difficult to extract deep customer insights from massive data. For example, when analyzing customer purchasing behavior, manual analysis can only process a limited amount of data and is unable to capture subtle changes in customer purchasing patterns and potential correlations between purchases of different products in a timely manner. This leads to a lack of precision in the formulation of marketing strategies by companies, making it difficult to provide personalized product recommendations and service solutions for different customer groups, thus missing out on many market opportunities. To this end, a customer relationship management data analysis method and system based on artificial intelligence is proposed. Summary of the invention

[0004] In view of this, the present invention provides a customer relationship management data analysis method and system based on artificial intelligence to solve or alleviate the technical problems existing in the prior art and at least provide a beneficial option.

[0005] The technical solution of the present invention is implemented as follows: a customer relationship management data analysis method based on artificial intelligence comprises the following steps:

[0006] Step 1: Collect customer data

[0007] Collect customer data through various channels, including but not limited to basic customer information (such as name, age, gender, contact information, etc.), transaction data (such as the type, quantity, amount, time, etc. of purchased products or services), conversation records (such as customer service chat records, online consultation records, etc.) and social media interaction data (such as likes, comments, shares, etc.). Let the collected customer data matrix be D, where D ij Represents the value of the jth data type of the ith customer, i = 1, 2, ..., m, m is the number of customers, j = 1, 2, ..., n, n is the number of data types.

[0008] Step 2: Store the data

[0009] A distributed storage system is used to store the collected customer data to ensure data security and scalability. The data is stored in structured and unstructured forms to facilitate subsequent data processing and analysis. The data structure after storage is assumed to be S, where S can be represented as a combination of a series of database tables and file systems to meet the needs of fast data retrieval and update.

[0010] Step 3: Analyze and mine the data

[0011] 1. Data preprocessing

[0012] Clean the collected data, remove noise data and outliers, and use the data standardization formula where x ij is the original data, μ j is the mean value of the jth data type, σ j is the standard deviation of the jth data type, x' ij The data is standardized, so that data of different dimensions can be unified into a standard range, which is convenient for subsequent analysis and comparison.

[0013] 2. Feature extraction

[0014] The principal component analysis (PCA) algorithm is used to extract features from the preprocessed data. The goal of PCA is to find a set of new orthogonal eigenvectors that maximize the variance of the projection of the data on these new eigenvectors. Let the original data matrix X (after standardization) and the covariance matrix be Solve for the eigenvalue λ of the covariance matrix C k and the eigenvector v k , k = 1, 2, ..., n, select the first p eigenvectors in descending order of eigenvalue to form the projection matrix P, then the projection of the data in the new feature space is Y = XP, where Y ik It represents the projection value of the i-th customer on the k-th principal component. Through feature extraction, the dimension of the data can be reduced while retaining the main information of the data.

[0015] 3. Customer behavior pattern recognition

[0016] Clustering algorithms (such as K-Means clustering) are used to identify customer behavior patterns on the extracted feature data. The goal of the K-Means algorithm is to divide the data points into K clusters so that the sum of the distances between the data points in the cluster is minimized. Let the cluster center be μ k , k=1,2,…,K, data point y i To cluster center μ k The distance is calculated using the Euclidean distance formula By continuously iteratively updating the cluster center and the cluster to which the data point belongs until the convergence conditions are met, the clustering results can help companies identify different types of customer behavior patterns, such as high-value customers, potential churn customers, etc.

[0017] Step 4: Build a customer profile

[0018] Build a customer profile based on the customer's basic information, transaction data, conversation records, social media interaction data, and behavior pattern recognition results. The customer profile can be represented as a vector P i =(a i1 ,a i2 ,…,a iq ), where a ij represents the attribute value of the i-th customer on the j-th portrait dimension, and q is the number of portrait dimensions. For example, a i1 Can represent the age range of customers, a i2 It can represent the customer's consumption preference level, etc. By building customer portraits, companies can understand the customer's characteristics and needs more intuitively.

[0019] Step 5: Provide intelligent decision support for customer relationship management based on customer portraits and data analysis results

[0020] 1. Customer value assessment

[0021] Use the customer value evaluation model to evaluate the value of customers, assuming that the customer value function is V i =f(P i ,Y i ), where Y i is the behavioral characteristic data of the ith customer (such as purchase frequency, purchase amount, etc.), f is the value evaluation function, and a linear regression model can be used where β j and γ k is the regression coefficient, which is estimated through training data. According to the customer value assessment results, the company can manage customers in different levels and formulate different marketing strategies and service plans for customers of different values.

[0022] 2. Personalized recommendations

[0023] Use collaborative filtering algorithm to provide personalized recommendations for customers. Let the user-item rating matrix be R, where R ij represents the rating of the i-th customer on the j-th item (if no rating is given, it is a missing value). For the target customer u, find a set of k neighbor customers N similar to it. u , the similarity between customers uses the cosine similarity formula Recommend items or services that may be of interest to target customers based on their neighbor customers’ purchasing behavior and ratings.

[0024] The customer relationship management data analysis system based on artificial intelligence includes data collection module, data storage module, data analysis module, customer portrait module, intelligent decision-making module and system management module;

[0025] The data collection module collects data through the enterprise's internal sales system, customer service system, marketing system and social media platform;

[0026] The data storage module selects a database storage type according to the structure and type of the data;

[0027] The data analysis module uses data mining algorithms and machine learning techniques to perform in-depth mining on the pre-processed data;

[0028] The customer profiling module uses a clustering algorithm to divide customers into different groups, each group having similar characteristics and behavior patterns;

[0029] The intelligent decision-making module formulates corresponding intelligent decision-making strategies based on the trained model and data analysis results, combined with the business goals and strategies of the enterprise;

[0030] The system management module configures and adjusts various parameters of the system.

[0031] Further preferably, the data collection module collects relevant data from the modules of the customer relationship management system, collects massive customer data from the sales system, marketing platform, and customer service records, and cleans, converts and integrates these multi-source heterogeneous data to remove noise and inconsistent data.

[0032] Further preferably, the data storage module, the database is responsible for storing the original data and the data after preliminary processing, and provides data writing, reading, updating and deleting operations.

[0033] Further preferably, the data analysis module utilizes the computing resources of the cloud computing platform to perform model training, generate a trained model and apply it to the analysis of actual data.

[0034] Further preferably, the customer portrait module performs visual display and management of customer portraits, and can present customer portrait data to the company's marketing, sales and customer service teams in the form of intuitive charts and reports, and provides a customer portrait update mechanism to update customer portraits in a timely manner when there is new customer data or changes in customer behavior.

[0035] Further preferably, the intelligent decision-making module simulates different decision-making schemes before implementing the decision to predict their effects and impacts.

[0036] Further preferably, the system management module is responsible for user management, authority management, system configuration and monitoring of the entire system.

[0037] The embodiment of the present invention adopts the above technical solution, and has the following innovative features:

[0038] 1. Multi-source data fusion and deep mining innovation

[0039] 1. Comprehensive data collection channel integration

[0040] The present invention innovatively integrates multiple data sources such as the company's internal sales system, customer service system, marketing system, and social media platforms. Unlike traditional customer relationship management data analysis methods that rely only on a single or limited data source, this multi-source data collection can obtain comprehensive customer information, such as transaction details from the sales system, customer feedback from the customer service system, and customer reputation and social influence from social media platforms, thereby building a rich and three-dimensional customer data system, providing a more solid data foundation for accurate analysis of customer behavior and needs.

[0041] 2. Application of deep data mining and analysis technology

[0042] The use of advanced data mining algorithms and machine learning techniques to deeply process the massive multi-source heterogeneous data collected, and the use of algorithms such as principal component analysis (PCA) for feature extraction can reduce the data dimension while retaining key information, effectively solving the "dimensionality disaster" problem caused by high-dimensional data, and improving the efficiency and accuracy of data analysis. For example, when processing a data set containing many customer attributes and behavioral data, PCA can screen out the most valuable feature combinations for customer classification or prediction, so that subsequent clustering analysis or value assessment models can run more accurately. At the same time, using K-Means clustering algorithms and other algorithms to identify customer behavior patterns can unearth customer group characteristics and behavioral patterns hidden in the data, providing a strong basis for companies to formulate personalized marketing strategies, which is difficult to achieve with traditional experience-based or simple statistical analysis methods.

[0043] 2. Innovation in building accurate customer portraits

[0044] 1. Multi-dimensional image feature extraction

[0045] When constructing customer portraits, multi-dimensional information such as demographic characteristics, consumer behavior characteristics, social behavior characteristics, and value characteristics are taken into consideration. Compared with previous portrait construction methods that only focused on some customer characteristics, this multi-dimensional portrait can portray the customer image more comprehensively and meticulously. For example, by analyzing the customer's interactive behavior on social media, we can understand their social circles and interests, and then combine their consumer behavior characteristics to predict their potential demand for new products or services. Taking a young customer who often follows technology products on social media and purchases them frequently as an example, his portrait will not only show his basic information such as age, gender, and purchase history, but also highlight his social attributes as a technology enthusiast and his value characteristics of high consumption potential, enabling companies to more accurately customize marketing and services to meet their needs.

[0046] 2. Dynamic update mechanism

[0047] A dynamic update mechanism for customer portraits has been established, which can respond to changes in customer data and behavior in a timely manner. Traditional customer portraits are often static and difficult to reflect the real-time status of customers. When customers have new purchasing behaviors, social interactions or other data updates, the present invention quickly incorporates new data into the analysis process and updates portrait features. For example, when a customer purchases a new high-end product, the system will immediately update its consumption behavior characteristics and value characteristics, and further adjust its social behavior characteristics and potential demand predictions based on its sharing and evaluation of the product on social media, ensuring that the company always grasps the latest situation of customers and provides guarantees for the continuous provision of personalized services.

[0048] 3. Intelligent decision-making supports innovation

[0049] 1. Model-based decision simulation and optimization

[0050] In the intelligent decision-making module, by performing model-based simulations on different decision-making plans, the decision effects and impacts can be predicted in advance. This is fundamentally different from the traditional way of making decisions based on experience or intuition. For example, when formulating a marketing plan, historical data and trained customer behavior models are used to simulate different advertising strategies, promotional activities, etc., analyze the possible sales growth, customer satisfaction improvement and other effects, and select the optimal plan based on the simulation results. At the same time, cross-validation, grid search and other technologies are used for optimization during the model training process, which improves the generalization ability and stability of the model, so that the decision-making model can more accurately adapt to different market environments and changes in customer groups, and provide more reliable decision-making support for enterprises.

[0051] 2. Personalized decision-making strategy generation

[0052] It is possible to generate personalized decision-making strategies based on customer portraits and data analysis results. Traditional customer relationship management decisions are often a one-size-fits-all model, which is difficult to meet the personalized needs of different customers. The present invention accurately classifies and analyzes customers and formulates exclusive decision-making strategies for different types of customer groups. For example, for high-value customers, personalized value-added services and exclusive preferential strategies are formulated; for potential lost customers, targeted retention measures are formulated. This personalized decision-making strategy can effectively improve customer satisfaction and loyalty, and enhance the company's market competitiveness.

[0053] IV. Innovation in System Management and Performance Assurance

[0054] 1. Flexible system configuration and adaptive adjustment

[0055] The system management module can flexibly configure various system parameters according to the personalized needs of the enterprise. Whether it is the selection and adjustment of database storage type or the allocation of computing resources of the cloud computing platform, it can be dynamically optimized according to the development and changes of the enterprise business. For example, when the enterprise business expands to new regions or launches new product lines, the system management module can timely adjust the data collection scope and frequency, optimize the database storage structure to adapt to the new data type and scale, and ensure that the system can continue to operate efficiently and meet the ever-changing needs of the enterprise. This is difficult to achieve with traditional fixed-configuration customer relationship management systems.

[0056] 2. Comprehensive system monitoring and intelligent maintenance

[0057] A comprehensive system monitoring mechanism has been established to monitor key indicators such as data flow, server load, and model operation efficiency in real time. Unlike ordinary system monitoring that only focuses on the operating status of some hardware or software, the monitoring of the present invention covers the entire process from data acquisition to decision output. When the system encounters abnormal conditions, such as data transmission interruption, server overload, or excessive model prediction error, it can automatically issue an alarm and take corresponding intelligent maintenance measures. For example, when the server load is too high, the load balancing mechanism is automatically started or resource allocation is dynamically adjusted; when the model prediction error increases, the model retraining or parameter adjustment is automatically triggered, which ensures the stability and reliability of the system and reduces the business risks caused by system failures or performance degradation.

[0058] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0061] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present invention. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0062] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0063] like Figure 1 As shown, the embodiment of the present invention provides a customer relationship management data analysis method based on artificial intelligence, comprising the following steps:

[0064] Step 1: Collect customer data;

[0065] Step 2: Store the data;

[0066] Step 3: Analyze and mine the data;

[0067] Step 4: Build a customer profile;

[0068] Step 5: Provide intelligent decision support for customer relationship management based on customer portraits and data analysis results;

[0069] Step 6: Evaluate the performance of the AI ​​model used.

[0070] The customer relationship management data analysis system based on artificial intelligence includes data collection module, data storage module, data analysis module, customer portrait module, intelligent decision-making module and system management module;

[0071] The data collection module collects data through the company's internal sales system, customer service system, marketing system and social media platform;

[0072] Data storage module, selects database storage type according to the structure and type of data;

[0073] The data analysis module uses data mining algorithms and machine learning techniques to conduct in-depth mining of pre-processed data;

[0074] The customer profiling module uses clustering algorithms to divide customers into different groups, each with similar characteristics and behavior patterns;

[0075] Intelligent decision-making module, based on trained models and data analysis results, combined with the business goals and strategies of the enterprise, formulates corresponding intelligent decision-making strategies;

[0076] System management module, configures and adjusts various system parameters.

[0077] In one embodiment, the data collection module collects relevant data from the modules of the customer relationship management system, collects massive amounts of customer data from the sales system, marketing platform, and customer service records, and cleans, converts, and integrates these multi-source heterogeneous data to remove noise and inconsistent data; performs preliminary integration and cleaning on the collected data to remove duplicate data, erroneous data, and incomplete data, unifies the data format, and standardizes the text data.

[0078] In this embodiment, the acquisition of multi-source customer data can be achieved in the following ways:

[0079] (1) The data collection module establishes a connection with the company's internal sales system to obtain detailed information about the products or services purchased by customers, including structured data such as product name, purchase quantity, purchase time, and payment amount, as well as unstructured data such as notes during the sales process. For example, the list of items purchased by a customer each time and the corresponding transaction amount are extracted from the sales order database, and the customer's special needs or preferences recorded by the salesperson in the order notes are collected.

[0080] (2) Connecting with the customer service system to collect conversation records such as customer consultations, complaints, and suggestions. These data may be stored in text form and cover customer feedback on product usage issues, evaluation of service quality, etc. For example, the customer service chat record database can be used to obtain the text of the communication between customers and customer service staff regarding product function questions, as well as customer complaints about the response speed of after-sales service.

[0081] (3) Collect customer participation data in various marketing activities from the marketing system, such as the open rate and click-through rate of email marketing, event registration information, and the receipt and use of promotional coupons. For example, analyze the database of email marketing activities, count the number of times specific customer groups open emails with different themes and click on links, as well as the time and usage scenarios of customers receiving coupons during promotional activities.

[0082] (4) Using the API (application programming interface) of social media platforms to capture customer interaction data related to the company on social platforms, including customers’ likes, comments, and sharing behaviors on the company’s official social media accounts, as well as content posted by customers that mention the company’s brand or products. For example, through the Xiaohongshu API, collect the number of likes, comments, and sharing links of customers on the relevant pages of the Xiaohongshu account, as well as information about customers mentioning the company’s products on Weibo.

[0083] In this embodiment, data cleaning and integration can be achieved in the following ways:

[0084] Remove duplicate data: Use an algorithm based on data hash values ​​or unique identifiers to check for duplicates in data collected from various data sources. For example, for sales order data, if there are identical order numbers and exactly the same order details, they are considered duplicate data and one of them is deleted. For customer service records, if there are records with the same timestamp, the same customer ID, and exactly the same conversation content, they are also deduplicated.

[0085] Correction of erroneous data: For obvious errors in the data, such as data type mismatch (such as recording the date format incorrectly as a string) and values ​​outside the reasonable range (such as recording the customer's age as 200 years old), corrections are made based on the logical relationship and business rules of the data. For data type errors, data conversion functions are used to convert them into the correct data type. For values ​​outside the reasonable range, if they can be inferred and corrected based on other relevant data (such as inferring the possible age range of the customer based on the type of product purchased and the purchase time), they are corrected. Otherwise, they are marked as abnormal data for further verification.

[0086] Incomplete data processing: For data records with missing key information, we use a variety of data filling strategies. For missing numerical data, such as missing customer purchase amounts, if there is historical purchase data for the customer and the purchase behavior has a certain regularity, we fill it in according to the average or median of the historical data. If there is not enough historical data, we use the mean filling method based on similar customer groups. For textual missing data, such as missing customer occupation information, we use natural language processing technology to analyze other relevant text data of the customer (such as customer service conversation content, social media comments), extract keywords that may be related to the occupation, and thus infer the customer's occupation. If it cannot be effectively inferred, it will be marked as data to be supplemented.

[0087] Text data standardization: Use the natural language processing toolkit to convert all text data into lowercase letters, remove stop words (such as common conjunctions, prepositions, etc.), and standardize professional terms. For example, "iPhone" and "Apple phone" are uniformly mapped to "i Phone" to facilitate subsequent text analysis and mining.

[0088] In one embodiment, the data storage module, the database is responsible for storing the original data and the data after preliminary processing, providing data writing, reading, updating and deleting operations; and storing and organizing the data in layers to facilitate subsequent data analysis and query.

[0089] In this embodiment, the database selection and construction can be implemented in the following ways:

[0090] For structured data, such as basic customer information (name, gender, age, contact information, etc.), sales transaction details, marketing activity participation data, etc., choose a relational database (such as MySQL) for storage and design a reasonable database table structure. For example, create a "customer information table" containing fields such as "customer ID", "name", "gender", "age", "contact number", etc.; create a "sales record table" containing fields such as "sales order ID", "customer ID", "product ID", "purchase quantity", "purchase time", and "payment amount". By establishing primary keys, foreign key constraints and indexes, the efficiency of data writing, reading, updating and deleting operations can be improved. For example, in the "sales record table", set the "sales order ID" as the primary key, and set the "customer ID" and "product ID" as foreign keys respectively, associate the "customer information table" and the "product information table", and create an index on the "customer ID" field to quickly query the sales records of specific customers.

[0091] For unstructured data, such as customer service conversation records and social media comments, non-relational databases (such as MongoDB) are used for storage. The data is stored in the form of documents, and each document represents a data record. For example, a customer service conversation record document can contain fields such as "conversation ID", "customer ID", "conversation time", and "conversation content". The flexible data structure of non-relational databases can better adapt to the diversity and complexity of unstructured data, and facilitate data storage and subsequent queries.

[0092] In this embodiment, the hierarchical data storage organization can be implemented in the following ways:

[0093] Raw data layer: The unprocessed raw data collected from various data sources is stored in its entirety in the raw data layer. For example, the raw sales order data files directly exported from the sales system and the raw JSON interactive data files captured from the social media platform are stored in a dedicated raw data storage area to ensure data traceability and integrity, and facilitate backtracking and reprocessing when necessary.

[0094] Preprocessing data layer: stores preliminary processed data after cleaning, conversion and integration. For example, customer information data after deduplication, error correction and unified format is stored in the "Customer Information Preprocessing Table" of the preprocessing data layer, and standardized text data is stored in the corresponding preprocessed text data set. The preprocessing data layer provides high-quality data input for the data analysis module, improving the accuracy and efficiency of data analysis.

[0095] Analysis result data layer: stores various analysis results, model parameters, customer portraits and other data generated by the data analysis module. For example, the customer behavior pattern classification results generated by the data mining algorithm are stored in the "Customer Behavior Analysis Result Table", the model parameter files obtained by machine learning model training are stored in a special model parameter storage area, and the constructed customer portrait data is stored in the customer portrait data set in JSON or other suitable formats. The analysis result data layer facilitates data calls and decision making by the intelligent decision-making module and other related modules. At the same time, a complete data directory and metadata management mechanism is established to record and manage metadata information such as data storage location, data structure, and data source, so as to quickly locate and retrieve required data.

[0096] In one embodiment, the data analysis module uses the computing resources of the cloud computing platform to perform model training, generates a trained model and applies it to the analysis of actual data; and uses the K-Means clustering algorithm to divide customers into different groups so that the company can formulate personalized marketing strategies for different groups.

[0097] In this embodiment, the cloud computing platform utilization and model training can be completed in the following ways:

[0098] Use the powerful computing resources of cloud computing platforms (such as Alibaba Cloud, Tencent Cloud, Huawei Cloud, etc.) to train the model. First, divide the preprocessed data into training set, validation set and test set according to a certain ratio (such as 70% training set, 20% validation set, 10% test set). During the training process, select appropriate data mining algorithms and machine learning techniques according to the data characteristics and analysis objectives. For example, for customer purchase behavior prediction, if the data has an obvious linear relationship, a linear regression algorithm can be used; if the data relationship is more complex and has nonlinear characteristics, a neural network algorithm or a decision tree algorithm can be selected. Taking the neural network algorithm as an example, determine the network structure (such as the number of layers including the input layer, hidden layer and output layer and the number of nodes in each layer), set the initial parameters (such as learning rate, weight initialization method, etc.), input the training set data into the neural network for training, and evaluate the prediction error of the model by calculating the loss function (such as mean square error, cross entropy, etc.), and use the back propagation algorithm to continuously adjust the weights and bias of the neural network. In order to reduce the loss function value, during the training process, a cross-validation technique (such as 5-fold cross-validation) is used to divide the training set into 5 subsets, and one of the subsets is used as the validation set and the remaining subsets are used as the training set in turn. Training and validation are performed 5 times, and the performance indicators of the model on the validation set (such as accuracy, recall rate, F1 value, etc.) are recorded each time. Finally, the average performance indicator is taken as the evaluation result of the model. Through the grid search technology, a comprehensive search is performed within the pre-set parameter and hyperparameter range (such as the learning rate range is [0.001, 0.1], the number of hidden layer nodes range is [10, 100], etc.), and the model is trained and evaluated for each parameter combination to find the parameter combination that optimizes the model performance. For example, for a simple neural network model, training and evaluation are performed under different combinations of learning rates and hidden layer nodes. It is found that when the learning rate is 0.01 and the number of hidden layer nodes is 50, the model has the highest accuracy on the validation set, and this parameter combination is determined to be the optimal parameter combination.

[0099] In this embodiment, the application of the K-Means clustering algorithm can be implemented in the following ways:

[0100] Use the K-Means clustering algorithm to divide customers into different groups. First, determine the number of clusters K. You can use the elbow method, silhouette coefficient method, and other methods to determine it. Take the elbow method as an example to calculate the intra-cluster sum of square errors (SSE) under different K values ​​(such as K from 2 to 10). In the specific calculation, for each K value, randomly initialize K cluster centers, assign each customer data point to the cluster to which the nearest cluster center belongs, calculate the sum of squares of the distances from each data point in the cluster to its cluster center, add the sum of squares of the distances of all clusters to get the SSE, and draw a K-SSE curve. When the curve has an obvious inflection point, the corresponding K value is the more appropriate number of clusters. For example, when K gradually increases from 2 to 6, the SSE decreases significantly, and when K is greater than 6, the SSE decreases significantly. After that, the downward trend of SSE slows down and an obvious turning point appears, then K=6 is determined as the number of clusters, and then 6 cluster centers are randomly initialized, and each customer data point is assigned to the cluster to which the nearest cluster center belongs. The mean of the data points in each cluster is calculated as the new cluster center, and this process is repeated until the cluster center no longer changes significantly (such as the coordinate change of the cluster center in two iterations is less than the set threshold) or the preset number of iterations (such as 100 times) is reached. Through cluster analysis, customers are divided into groups with similar characteristics and behavior patterns, such as customers with high consumption frequency and high consumption amount are clustered into one category, and customers with low frequency and low consumption but sensitive to promotions are clustered into another category, so that companies can formulate personalized marketing strategies for different groups.

[0101] In one embodiment, the customer portrait module visualizes and manages customer portraits, and can present customer portrait data to the company's marketing, sales, and customer service teams in the form of intuitive charts and reports, and provide a customer portrait update mechanism to promptly update customer portraits when there is new customer data or changes in customer behavior; based on the above analysis and model prediction results, a customer portrait is generated to present the customer's characteristics, behavior patterns, preferences, and value information in the form of intuitive charts, graphs, or text descriptions.

[0102] In this embodiment, the portrait construction and visualization display can be achieved in the following ways:

[0103] Based on the above data analysis and model prediction results, a customer profile is constructed, which covers many aspects:

[0104] Demographic characteristics: Extract the customer's age, gender, occupation, region, family income and other information from the customer information data. For example, determine the customer's geographical distribution by analyzing the address information filled in by the customer in the registration information or purchase record, and count the number of customers in different regions and the proportion of consumption; analyze the consumption preferences and behavior patterns of customers with different occupations from the customer's occupation information (if any) or occupation information inferred based on other relevant data.

[0105] Consumption behavior characteristics: including purchase frequency, purchase amount, purchase preference, consumption cycle, etc. For example, by analyzing sales record data, the number of purchases made by customers within a certain period of time (such as the past year) is calculated as the purchase frequency, the total purchase amount is counted, and the association analysis algorithm is used to mine the product combinations that customers frequently purchase, determine their purchase preferences, analyze the time intervals between customers' purchase behaviors, and determine the consumption cycle.

[0106] Social behavior characteristics: Extract information such as the customer's activity, influence, and social circle type on social platforms from social media interaction data. For example, evaluate the customer's activity by counting the number of likes, comments, and shares on social media; evaluate the customer's influence based on factors such as the number of fans and the influence of the interaction objects; use social network analysis algorithms to analyze the customer's social relationship network and determine the type of social circle (such as a circle of technology enthusiasts, a circle of fashion experts, etc.).

[0107] Value characteristics: Calculate customer lifetime value, potential value, etc. Customer lifetime value can be calculated by considering factors such as the customer's historical consumption amount, purchase frequency, customer retention rate, etc., using a specific calculation model (such as an extended model of the RFM model). Potential value can be evaluated based on factors such as customer behavior trends, market trends, and customer potential demand forecasts for new products or services. These characteristics are presented in the form of intuitive charts, graphs, or text descriptions to present customer characteristics, behavior patterns, preferences, and value information. For example, a bar chart can be used to show the distribution of consumption amounts of customers of different age groups, a pie chart can be used to show the proportion of customers' preferences for different product categories, and a text description can be used to summarize the typical behavior patterns of high-value customers, such as "high-value customers usually purchase high-end products 3-5 times a month, respond positively to new product recommendations, have a high influence on social media, and often share product usage experiences." These visual displays can intuitively present customer portrait data to the company's marketing, sales, and customer service teams, enabling them to quickly understand the full picture of customers and provide strong support for precision marketing and personalized services.

[0108] In this embodiment, the portrait update mechanism can be achieved by:

[0109] Establish a customer portrait update mechanism. When there is new customer data or customer behavior changes, update the customer portrait in a timely manner. For example, when a customer completes a new purchase transaction, the data collection module collects the new purchase data and transmits it to the data analysis module. The data analysis module re-analyzes whether the customer's consumption behavior characteristics (such as purchase frequency, purchase amount, purchase preferences, etc.) have changed. If there are changes, the updated information is fed back to the customer portrait module to update the consumption behavior characteristics part of the customer portrait. When the customer has new interactive behaviors on social media (such as posting new comments, liking new content, etc.), the data collection module collects new social data, and the data analysis module analyzes the changes in its social behavior characteristics, such as whether the activity level has increased, whether the influence has changed, etc., and passes the updated information to the customer portrait module for corresponding updates. At the same time, regularly (such as monthly or quarterly) conduct a comprehensive review and update of the customer portrait to ensure that the portrait can accurately reflect the customer's current status and trends. For example, regularly recalculate the customer's lifetime value and potential value, and re-evaluate the customer's value characteristics according to changes in the market environment and adjustments to the company's business strategy; update the customer's consumption preferences and behavior patterns and other information based on new data analysis results and market research data.

[0110] In one embodiment, the intelligent decision-making module simulates different decision-making plans before implementing the decision to predict their effects and impacts; selects appropriate machine learning, deep learning or other artificial intelligence models according to specific decision-making problems and data characteristics, and formulates corresponding intelligent decision-making strategies based on the trained models and data analysis results, combined with the company's business goals and strategies, and uses historical data to train the selected models. By adjusting the parameters and hyperparameters of the models, the models can learn the laws and patterns in the data, so that they can accurately predict and analyze customer behaviors and market trends. During the training process, cross-validation, grid search and other technologies are used to optimize the model to improve the generalization ability and stability of the model and avoid the occurrence of overfitting.

[0111] In this embodiment, the decision-making scheme simulation and model selection can be implemented in the following ways:

[0112] Before implementing a decision, simulate different decision plans to predict their effects and impacts. For example, a company plans to launch a new product and considers two marketing decision plans: Plan 1 is to conduct large-scale advertising for all customers; Plan 2 is to first conduct precision marketing for high-value customers and customers with potential demand for new product categories, and then gradually expand the marketing scope based on market response. Use historical data and the constructed customer portraits and data analysis models to simulate these two plans. For Plan 1, predict the cost of advertising, possible sales conversion rate, profit and other indicators based on the size of all customers, consumption preferences, response rate to advertising and other data; for Plan 2, predict the cost of precision marketing, initial sales volume, word-of-mouth communication effect and other indicators based on the characteristics and behavior patterns of high-value customers and potential customers and their possible response rate to precision marketing, and further simulate the market response after the marketing scope is expanded. According to the specific decision-making problem and data characteristics, select appropriate machine learning, deep learning or other artificial intelligence models. For example, for customer churn warning and retention decisions, you can use survival model. The analysis model predicts the probability and time of customer churn, and formulates corresponding retention strategies based on the prediction results; for product recommendation decisions, collaborative filtering algorithms or content-based recommendation algorithms can be used to recommend personalized products or services to customers. Taking the collaborative filtering algorithm as an example, the similarity between customers is calculated based on the customer's historical purchase behavior data (such as the user-based collaborative filtering algorithm calculates the interest similarity between users), and then based on the purchase behavior of similar customers, recommend products that may be of interest to the target customers. Based on the trained model and data analysis results, combined with the company's business goals and strategies, formulate corresponding intelligent decision-making strategies. For example, if the company's business goal is to improve customer satisfaction and loyalty, and the data analysis results show that a certain type of customer responds well to personalized services, then the intelligent decision-making strategy can be to launch exclusive personalized service packages for such customers and promote them through appropriate channels (such as SMS, email, APP push, etc.). For example, for high-value customers with a high preference for fashion products, provide them with exclusive fashion new product priority experience services and notify them through SMS and APP push.

[0113] In this embodiment, model training and optimization can be accomplished in the following ways:

[0114] Use historical data to train the selected model. By adjusting the parameters and hyperparameters of the model, the model can learn the laws and patterns in the data, so that it can accurately predict and analyze customer behavior and market trends. Taking the neural network model as an example, during the training process, the model performance is optimized by adjusting the number of hidden layers, the number of nodes, the learning rate, the regularization coefficient and other parameters and hyperparameters. The model is optimized using cross-validation, grid search and other technologies to improve the generalization ability and stability of the model and avoid overfitting. For example, when using the neural network model to predict customer demand, a 10-fold cross-validation is used to determine the appropriate network structure and parameter settings. , grid search is used to search in the parameter space of learning rate value range [0.0001,0.1], hidden layer node number range [5,100], regularization coefficient value range [0.001,0.1], etc., and model training and evaluation are performed for each parameter combination to find the parameter combination that minimizes the model loss function value on the validation set and has stable performance on the test set. For example, after optimization, it is determined that the hidden layer of the neural network model is 3 layers, the number of nodes is 30, 20, and 10 respectively, the learning rate is 0.001, and the regularization coefficient is 0.01. The model has high accuracy and stability in predicting customer purchase demand in the next month.

[0115] In one embodiment, the system management module is responsible for user management, authority management, system configuration and monitoring of the entire system; according to the job responsibilities and needs of different users, corresponding roles and permissions are assigned to them, and various parameters of the system are configured and adjusted to meet the personalized needs of the enterprise.

[0116] In this embodiment, regarding user and authority management, further optimization schemes are as follows:

[0117] The user management and permission management of the entire system assigns corresponding roles and permissions to different users according to their job responsibilities and needs. For example, a "data collector" role is created for data collectors, and they are given operation permissions for the data collection module, including data collection, preliminary sorting and other functions, such as allowing them to import data from various data sources, perform simple format checks and preliminary cleaning operations on the collected data; a "data analyst" role is created for data analysts to enable them to access the data analysis module to perform model training, data analysis and other operations, such as allowing them to select data mining algorithms, adjust model parameters, view analysis results, etc.; a "marketing / customer service specialist" role is created for marketing and customer service personnel, who can view customer portraits and intelligent decision-making results, but cannot modify core content such as model parameters and system configuration, such as They are allowed to view information such as consumption preferences and value characteristics in the customer portrait, as well as marketing suggestions and customer service strategies generated by the intelligent decision-making module, but they cannot retrain the data analysis model or modify the system's database configuration. This sophisticated user and permission management ensures the security of the system and the confidentiality of the data, while also enabling personnel in different positions to work efficiently within the scope of their authority. For example, data collectors focus on collecting high-quality data to ensure the integrity and accuracy of the data; data analysts can deeply explore the value of data and provide companies with valuable analysis results and models; marketers develop precise marketing activity plans based on customer portraits and intelligent decision-making strategies to improve marketing effectiveness; and customer service personnel provide personalized customer service based on customer portraits to improve customer satisfaction and loyalty.

[0118] In this embodiment, regarding system configuration and monitoring, further optimization schemes are as follows:

[0119] Configure and adjust various parameters of the system to meet the personalized needs of the enterprise. For example, the storage capacity and performance parameters of the database can be adjusted according to the enterprise's data storage needs. If the enterprise's business volume grows rapidly and the amount of data increases sharply, the system management module can dynamically increase the memory and disk space of the database server or adjust the database cache strategy to improve the efficiency of data storage and reading. Set the computing resource allocation of the cloud computing platform according to the real-time requirements of data analysis. For real-time data analysis tasks that require rapid response, such as real-time analysis of customer purchasing behavior by e-commerce platforms during promotional activities, the system management module can allocate more cloud computing resources to the data analysis module, such as increasing the number of CPU cores, improving memory bandwidth, etc., to ensure that data analysis can be completed in a timely manner and provide timely support for enterprise decision-making. Adjust the frequency and scope of data collection according to changes in the enterprise's business processes. If the enterprise launches a new business product line or enters a new market area, the system management module can adjust the configuration of the data collection module accordingly, increase the collection channels and frequency of new business-related data, and ensure that the system can obtain comprehensive and timely customer data.

[0120] At the same time, a system monitoring mechanism is established to monitor the operating status of the system in real time, including indicators such as data flow, server load, and model operation efficiency. For example, the data transmission flow between the data acquisition module and various data sources is monitored in real time through network monitoring tools to ensure the stability and continuity of data transmission. Server performance monitoring software is used to monitor the CPU utilization, memory utilization, disk I / O and other indicators of database servers, cloud computing servers, etc. When the server load exceeds the preset threshold (such as CPU utilization exceeds 80%), an alarm is issued in time and corresponding countermeasures are taken, such as automatically starting the load balancing mechanism to assign some tasks to other idle servers, or dynamically adjusting server resource configuration to increase the number of CPU cores or memory capacity. For model operation efficiency, the time consumption of model training and prediction is monitored. When the model runs for too long and affects the real-time performance of the system, the reasons are analyzed, which may be too large a data volume, too complex a model structure, or insufficient computing resources. Then targeted measures are taken, such as sampling data, optimizing the model structure, or increasing computing resources, to ensure the stable operation and efficient performance of the system.

[0121] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of various changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A customer relationship management data analysis method based on artificial intelligence, characterized in that: The following steps are involved: Data collection steps: Collect customer data through various channels to form a customer data matrix D, where D ij Represents the value of the jth data type of the i-th customer, i = 1, 2, ..., m, m is the number of customers, j = 1, 2, ..., n, n is the number of data types; Data storage step: Use a distributed storage system to store the collected customer data. The data structure after storage is S to ensure data security and scalability, and facilitate subsequent data processing and analysis; Data analysis and mining steps: Data preprocessing: Clean the collected data, remove noise data and outliers, and use data standardization formulas The data is standardized, where x ij is the original data, μ j is the mean value of the jth data type, σ j is the standard deviation of the jth data type, x' ij is the standardized data; Feature extraction: Use the principal component analysis (PCA) algorithm to extract features from the preprocessed data and solve the covariance matrix The eigenvalue λ k and the eigenvector v k , k = 1, 2, ..., n, select the first p eigenvectors in descending order of eigenvalue to form the projection matrix P, then the projection of the data in the new feature space is Y = XP, where Y ik Represents the projection value of the i-th customer on the k-th principal component; Customer behavior pattern recognition: The clustering algorithm is used to recognize customer behavior patterns on the extracted feature data. The clustering algorithm uses K-Means clustering. The K-Means algorithm divides the data points into K clusters with the goal of minimizing the sum of the distances of the data points within the cluster. The data point y i To cluster center μ k The distance is calculated using the Euclidean distance formula The cluster center and the cluster to which the data point belongs are updated by continuous iteration until the convergence condition is met; Customer portrait construction steps: Build a customer portrait based on the customer's basic information, transaction data, conversation records, social media interaction data, and behavior pattern recognition results. The customer portrait is represented by a vector P i =(a i1 ,a i2 ,…,a iq ), where a ij represents the attribute value of the i-th customer on the j-th portrait dimension, and q is the number of portrait dimensions; Intelligent decision support steps: Customer value assessment: Use the customer value assessment model to assess the value of customers. The customer value function is V i =f(P i ,Y i ), where Y i is the behavioral characteristic data of the i-th customer, f is the value evaluation function, and the linear regression model is used where β j and γ k is the regression coefficient, estimated through training data; Personalized recommendation: Use collaborative filtering algorithm to provide personalized recommendations for customers. Let the user-item rating matrix be R, where R ij Represents the score of the i-th customer on the j-th item. For the target customer u, find a set of k neighbor customers N similar to it. u , the similarity between customers uses the cosine similarity formula Recommend items or services that may be of interest to target customers based on their neighbor customers’ purchasing behavior and ratings.

2. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: The multiple channels include a customer's basic information channel, a transaction data channel, a conversation record channel, and a social media interaction data channel.

3. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: The distributed storage system includes a combination of structured and unstructured storage forms to meet the needs of rapid retrieval and update of data.

4. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: In the principal component analysis (PCA) algorithm, the calculation of the covariance matrix C is based on the original data matrix X after standardization.

5. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: In the K-Means clustering algorithm, cluster centers and the clusters to which data points belong are updated through continuous iterations, and the convergence condition is that the change in cluster centers is less than a preset threshold or the number of iterations reaches a preset number.

6. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: In the customer portrait construction step, portrait dimensions include but are not limited to the customer's age range, consumption preference level, purchase frequency range, etc.

7. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: The training data of the customer value assessment model comes from historical customer data, including customer portrait data and behavioral characteristic data.

8. The customer relationship management data analysis method based on artificial intelligence according to claim 1 is characterized in that: When making personalized recommendations, the collaborative filtering algorithm calculates the recommendation degree of the recommended item or service by using a weighted average or other method based on the similarity between the target customer and the neighbor customers and the purchase behavior and ratings of the neighbor customers.

9. A customer relationship management data analysis system based on artificial intelligence, characterized in that: include: Data collection module: used to collect customer data through various channels to form a customer data matrix D; Data storage module: used to store the collected customer data using a distributed storage system, and the data structure after storage is S; Data analysis and mining module: Data preprocessing submodule: used to clean the collected data, remove noise data and outliers, and use data standardization formula Standardize the data; Feature extraction submodule: It is used to extract features from preprocessed data using the principal component analysis (PCA) algorithm, by solving the covariance matrix The eigenvalue λ k and the eigenvector v k , select the first p eigenvectors in descending order of eigenvalue to form the projection matrix P, and obtain the projection Y=XP of the data in the new feature space; Customer behavior pattern recognition submodule: The clustering algorithm is used to recognize customer behavior patterns on the extracted feature data. The clustering algorithm uses K-Means clustering. i To cluster center μ k The distance is calculated using the Euclidean distance formula The cluster center and the cluster to which the data point belongs are updated by continuous iteration until the convergence condition is met; Customer portrait building module: used to build customer portrait P based on the customer's basic information, transaction data, conversation records, social media interaction data and behavior pattern recognition results. i =(a i1 ,a i2 ,…,a iq ); Intelligent decision support module: Customer value assessment submodule: Use the customer value assessment model to assess the value of customers. The customer value function is V i =f(P i ,Y i ), using a linear regression model Estimation through training data; Personalized recommendation submodule: Use collaborative filtering algorithm to provide personalized recommendations for customers. Let the user-item rating matrix be R. For the target customer u, find a set of k neighbor customers N similar to it. u , the similarity between customers uses the cosine similarity formula Recommend items or services that may be of interest to target customers based on their neighbor customers’ purchasing behavior and ratings.

10. The artificial intelligence-based customer relationship management data analysis system according to claim 9, characterized in that: The multiple channels of the data collection module include a customer basic information channel, a transaction data channel, a conversation record channel, and a social media interaction data channel.

11. The artificial intelligence-based customer relationship management data analysis system according to claim 9, characterized in that: The distributed storage system of the data storage module includes a combination of structured and unstructured storage forms to meet the needs of rapid retrieval and update of data.

12. The artificial intelligence-based customer relationship management data analysis system according to claim 9, characterized in that: In the principal component analysis (PCA) algorithm of the data analysis and mining module, the calculation of the covariance matrix C is based on the original data matrix X after standardization.

13. The customer relationship management data analysis system based on artificial intelligence according to claim 9, characterized in that: In the K-Means clustering algorithm of the data analysis and mining module, cluster centers and the clusters to which data points belong are updated through continuous iterations, and the convergence condition is that the change in the cluster center is less than a preset threshold or the number of iterations reaches a preset number.

14. The artificial intelligence-based customer relationship management data analysis system according to claim 9, characterized in that: The portrait dimensions of the customer portrait construction module include but are not limited to the customer's age range, consumption preference level, purchase frequency range, etc.

15. The customer relationship management data analysis system based on artificial intelligence according to claim 9, characterized in that: The training data of the customer value assessment submodule of the intelligent decision support module comes from historical customer data, including customer portrait data and behavior characteristic data.

16. The customer relationship management data analysis system based on artificial intelligence according to claim 9, characterized in that: When making personalized recommendations, the personalized recommendation submodule of the intelligent decision support module calculates the recommendation degree of the recommended item or service by using a weighted average or other method based on the similarity between the target customer and the neighbor customer and the purchase behavior and rating of the neighbor customer.

Images