Service area customer group portrait drawing method and system based on deep clustering

Abstract

The invention provides a service area customer group portraying method and system based on deep clustering. The method comprises the steps that S1, customer group information of a service area is acquired and preprocessed; s2, performing feature processing on the customer group information to obtain feature variables; s3, further extracting feature variables by using the trained model to obtain deep features; s4, performing clustering analysis on the customer group information of the service area, and outputting a clustering result; s5, calculating a contour coefficient to evaluate the clustering quality; optimizing and adjusting the model according to an evaluation result; and S6, according to a clustering result, dividing the customer group information of the service area into multiple types to construct customer group portraits. According to the method, the deep clustering algorithm is introduced into the customer group portrait of the high-speed service area, the limitation of a traditional method is effectively solved, different types of customer groups are identified, and the accurate customer group portrait is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of transportation systems, and in particular to a method and system for profiling customer groups in a service area based on deep clustering. Background Art

[0002] With the continuous expansion of the highway network and the rapid growth of the number of vehicles, the requirements for the management and service quality of highway service areas are increasing. As an important part of the highway, the service area provides drivers and passengers with a variety of services such as rest, refueling, catering, shopping, etc. Effectively managing the resources in the service area and optimizing the service quality are of great significance to improving user satisfaction and service area operation efficiency.

[0003] In the operation and management of highway service areas, vehicle information is an important source of data. By analyzing information such as vehicle type and dwell time, managers can understand the usage of service areas, optimize resource allocation, and improve service quality. However, traditional analysis methods mainly rely on simple statistics and rule analysis, which cannot fully explore the deep information in the data, resulting in inaccurate analysis results and difficulty in meeting the needs of modern management.

[0004] In recent years, deep learning technology has achieved remarkable results in the fields of image recognition, natural language processing and data analysis. Its powerful feature extraction and pattern recognition capabilities have enabled it to show great potential in complex data analysis. In particular, deep clustering algorithms, by combining deep learning with traditional clustering methods, can automatically extract deep features from data and perform efficient clustering analysis on data with high accuracy and robustness.

[0005] In the process of constructing customer portraits in highway service areas, traditional clustering methods such as K-means and hierarchical clustering mainly rely on shallow features and simple similarity measurements. Simple clustering analysis is performed by counting the basic information of the customer group, which has multiple limitations. Shallow features cannot fully capture the complex patterns and potential relationships in the data, resulting in inaccurate clustering results. At the same time, traditional clustering methods are sensitive to data noise and are easily disturbed, resulting in unstable clustering results. In addition, since the feature extraction and processing process relies on manual intervention, this not only increases time and labor costs, but also limits the real-time and automation level of the system.

[0006] In the Chinese patent document with the publication number CN110826886A, a method for constructing a power customer portrait based on a clustering algorithm and principal component analysis is disclosed, and the method includes the following steps: selecting information that affects power customer transactions as attributes for evaluating power customers; dividing all attributes into corresponding evaluation indicators according to different evaluation indicators to which each attribute belongs; reclassifying the continuous attributes based on the clustering algorithm, and determining the weights of each reclassification under the attribute; determining the evaluation index and its attribute weights based on the principal component analysis algorithm; calculating the comprehensive evaluation score of the power customer based on the attribute data of the power customer and the evaluation index, the attribute and the reclassification weights, and forming a power customer portrait. This patent document mainly relies on traditional clustering algorithms and principal component analysis to process power customer information, focusing on attribute reclassification and weight determination, which is suitable for the construction of an evaluation system, but its accuracy and robustness are not as good as deep clustering algorithms when processing high-dimensional and complex data. This application uses a deep learning Transformer model combined with reconstruction loss and clustering loss to generate a representation of deep features to automatically extract deep features from the customer information of highway service areas. It uses the K-means clustering algorithm to determine the initial clustering center, and continuously iterates and optimizes the model parameters and clustering results, which significantly improves the efficiency and accuracy of building customer portraits, and has strong scalability and business value.

[0007] In the Chinese patent document with the publication number CN114881678A, a high-precision logistics enterprise customer portrait method based on big data technology is disclosed. The user data is collected and pre-processed by combining the information of the database and the comprehensive method of multi-source heterogeneous crawlers; the hidden features are discovered according to the collected user data, and the user groups are classified using the density clustering algorithm. The classified users are scored and classified according to the acquired hidden features, and then the users are divided into three categories according to the results; the user's package is classified using the convolutional neural network optimized by the Salp Squid group algorithm, and the activation function is used and the network is modified according to the score of the network filter, and the package is classified according to three dimensions to construct the characteristic dimension of the user's package; according to the user's package and personal information, the user data is labeled and a high-precision customer portrait is generated; from the perspective of the logistics enterprise, the logistics customer is profiled in combination with the user's package, and a more accurate customer portrait is provided for the logistics enterprise. The patent document comprehensively collects user data through database and multi-source heterogeneous crawler technology, classifies users using the density clustering algorithm, and uses the optimized convolutional neural network to classify the user's package in multiple dimensions, and finally generates a customer portrait. This application places more emphasis on using the Transformer deep learning model to automatically extract customer features and to perform customer grouping using the K-means clustering algorithm. Compared with the density clustering algorithm, the deep clustering algorithm can better capture complex high-dimensional data structures and generate more accurate and detailed customer portraits. In addition, the Transformer model has advantages in processing time series data and long-distance dependency features. Compared with the convolutional neural network used by the former, it can better capture the complex patterns of customer behavior and is suitable for high-precision customer portrait analysis. Summary of the invention

[0008] In view of the defects in the prior art, the purpose of the present invention is to provide a method and system for service area customer profiling based on deep clustering.

[0009] A method for creating a customer profile of a service area based on deep clustering according to the present invention includes:

[0010] Step S1: Obtain customer group information of the service area and perform preprocessing;

[0011] Step S2: Process the preprocessed customer group information to obtain feature variables;

[0012] Encode and normalize the feature variables;

[0013] Step S3: Use the trained model to further extract feature variables to obtain deep features;

[0014] The loss function of the model includes reconstruction loss and clustering loss;

[0015] Step S4: Perform cluster analysis on the customer group information of the service area and output the clustering results;

[0016] Step S5: Complete the evaluation of clustering quality by calculating the silhouette coefficient to obtain the evaluation result; optimize and adjust the model according to the evaluation result, and repeat steps S3-S4 until the evaluation result meets the expected value;

[0017] Step S6: Classify the customer information of the service area into multiple types according to the clustering results to construct a customer profile.

[0018] Preferably, the customer information of the service area includes demographic characteristics, consumption behavior data, travel habit data and vehicle data; the preprocessing includes calculating the time each vehicle stays in the service area within a certain period of time based on the difference between two consecutive exit times and entrance times according to the records of vehicles entering and exiting the service area, and analyzing missing values, duplicate values ​​and outliers in the data.

[0019] Preferably, step S2 comprises:

[0020] For consumption behavior data, calculate the consumption interval, consumption frequency, total consumption amount, consumption commodity category and consumption time period of each customer within a certain time range;

[0021] For travel habit data, calculate the travel time and stay time of each customer in the service area within a certain time range;

[0022] For vehicle data, calculate the license plate type and vehicle type for each customer;

[0023] The encoding and normalization of the feature variables include frequency encoding of categorical variables, calculating the frequency of occurrence of each category and expressing it as a numerical feature; and normalizing continuous variables to eliminate dimensional differences between different features.

[0024] Preferably, step S3 comprises:

[0025] Step S3.1: Build a Transformer model, define the error between the output of the decoding stage and the original data as the reconstruction loss, and minimize the reconstruction loss; use the clustering loss to guide the representation learning process and minimize the clustering loss;

[0026] Step S3.2: Use the constructed Transformer model to perform preliminary feature extraction to obtain deep features; the deep features can capture complex patterns and relationships in the data;

[0027] Step S3.3: Train the Transformer model and optimize the reconstruction loss and clustering loss. In each iteration, adjust the model parameters according to the current clustering results and use the feedback mechanism to optimize the model performance.

[0028] Preferably, step S4 includes, during the K-means clustering process, the algorithm gradually adjusts the position of each cluster center by iterative optimization, and assigns each data point to the cluster center closest to it.

[0029] Preferably, the silhouette coefficient measures the clustering effect by comparing the closeness of each data point to its cluster and the degree of separation from other clusters; for a certain data point, the calculation formula of the silhouette coefficient is:

[0030]

[0031] Among them, a is the average distance from the data point to other data points in its cluster, and b is the average distance from the data point to the nearest other cluster. The closer the silhouette coefficient is to 1, the closer the data point is to other data points in its cluster and the more separated it is from data points in other clusters. If the silhouette coefficient is close to -1, it indicates that the data point may be incorrectly assigned to the current cluster.

[0032] Preferably, in step S6, the consumption behaviors, stay habits and vehicle information of the various types of customer groups are not completely the same.

[0033] A system for service area customer group portrait based on deep clustering provided by the present invention includes:

[0034] Module M1: Obtain customer information of the service area and perform preprocessing;

[0035] Module M2: Process the pre-processed customer group information to obtain feature variables;

[0036] Encode and normalize the feature variables;

[0037] Module M3: Use the trained model to further extract feature variables and obtain deep features;

[0038] The loss function of the model includes reconstruction loss and clustering loss;

[0039] Module M4: cluster analysis of customer information in the service area and output of clustering results;

[0040] Module M5: Complete the clustering quality evaluation by calculating the silhouette coefficient and obtain the evaluation result; optimize and adjust the model according to the evaluation result, and repeat modules M3-M4 until the evaluation result meets the expected value;

[0041] Module M6: Classify the customer information of the service area into multiple types based on the clustering results and construct a customer profile.

[0042] Preferably, the customer information of the service area includes demographic characteristics, consumption behavior data, travel habit data and vehicle data; the preprocessing includes calculating the time each vehicle stays in the service area within a certain period of time based on the difference between two consecutive exit times and entrance times according to the records of vehicles entering and exiting the service area, and analyzing missing values, duplicate values ​​and outliers in the data.

[0043] Preferably, the module M2 comprises:

[0044] For consumption behavior data, calculate the consumption interval, consumption frequency, total consumption amount, consumption commodity category and consumption time period of each customer within a certain time range;

[0045] For travel habit data, calculate the travel time and stay time of each customer in the service area within a certain time range;

[0046] For vehicle data, calculate the license plate type and vehicle type for each customer;

[0047] The encoding and normalization of the feature variables include frequency encoding of categorical variables, calculating the frequency of occurrence of each category and expressing it as a numerical feature; and normalizing continuous variables to eliminate dimensional differences between different features.

[0048] Preferably, the module M3 includes:

[0049] Module M3.1: Build a Transformer model, define the error between the output of the decoding stage and the original data as the reconstruction loss, and minimize the reconstruction loss; use the clustering loss to guide the representation learning process and minimize the clustering loss;

[0050] Module M3.2: Use the built Transformer model to perform preliminary feature extraction to obtain deep features; the deep features can capture complex patterns and relationships in the data;

[0051] Module M3.3: Train the Transformer model and optimize the reconstruction loss and clustering loss. In each iteration, adjust the model parameters according to the current clustering results and use the feedback mechanism to optimize the model performance.

[0052] Preferably, the module M4 includes that during the K-means clustering process, the algorithm gradually adjusts the position of each cluster center through iterative optimization, and assigns each data point to the cluster center closest to it.

[0053] Preferably, the silhouette coefficient measures the clustering effect by comparing the closeness of each data point to its cluster and the degree of separation from other clusters; for a certain data point, the calculation formula of the silhouette coefficient is:

[0054]

[0055] Among them, a is the average distance from the data point to other data points in its cluster, and b is the average distance from the data point to the nearest other cluster. The closer the silhouette coefficient is to 1, the closer the data point is to other data points in its cluster and the more separated it is from data points in other clusters. If the silhouette coefficient is close to -1, it indicates that the data point may be incorrectly assigned to the current cluster.

[0056] Preferably, in the module M6, the consumption behaviors, stay habits and vehicle information of the various types of customer groups are not completely the same.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. The present invention introduces a deep clustering algorithm into the customer portrait of the highway service area, which can effectively solve the limitations of traditional methods, identify different types of customer groups, and build accurate customer portraits; through these customer portraits, managers can better understand the behavioral characteristics and needs of different customer groups, thereby formulating more accurate management and service strategies, and improving the overall operational efficiency and user experience of the service area.

[0059] 2. The deep clustering algorithm of the present invention shows significant advantages in constructing customer portraits in highway service areas. The deep learning model automatically extracts deep features from customer information, significantly improving the efficiency of customer portrait construction. The deep learning model can automatically extract high-dimensional and discriminative deep features from raw data, greatly reducing manual intervention and significantly improving the degree of automation of data processing.

[0060] 3. The present invention utilizes deep clustering algorithms to perform customer segmentation, which can more accurately identify customer groups and improve the accuracy of customer portraits, enabling companies to gain an in-depth understanding of the behavioral characteristics and needs of different customer groups, develop more accurate marketing strategies, improve user experience and the profitability of service areas, help companies optimize resource allocation and service processes, and improve operational efficiency.

[0061] Other beneficial effects of the present invention will be explained in the specific implementation manner through the introduction of specific technical features and technical solutions. Through the introduction of these technical features and technical solutions, those skilled in the art should be able to understand the beneficial technical effects brought about by the technical features and technical solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0063] Figure 1 The figure is a flow chart of the method of the present invention.

[0064] Figure 2 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0065] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0066] The purpose of the present invention is to solve many problems existing in the current analysis of customer groups in highway service areas, including insufficient accuracy of customer group categories and insufficient in-depth characterization of complex customer group behaviors. By introducing a deep clustering algorithm, efficient extraction and accurate classification of customer group characteristics can be achieved, thereby providing more scientific and comprehensive support for service area operation management.

[0067] Specifically, the objects of the present invention include the following aspects:

[0068] 1. Improve clustering accuracy

[0069] Traditional clustering algorithms often show poor clustering effects when faced with complex high-dimensional data. The present invention, by introducing a deep neural network, can effectively extract the deep features of the data, thereby improving the accuracy of clustering and making the differences between different customer groups more obvious.

[0070] 2. Dealing with nonlinear relationships

[0071] The behavioral characteristics of the customer groups in the service area often have complex nonlinear relationships, and traditional linear models are difficult to accurately describe these relationships. The deep clustering algorithm proposed in the present invention can better capture the nonlinear patterns in the data, thereby achieving more accurate clustering results.

[0072] 3. Refine customer profile

[0073] Through deep clustering algorithms, more refined and multi-dimensional customer profiles can be generated. These profiles can help service area managers better understand customer characteristics and provide a scientific basis for optimizing services.

[0074] 4. Improve service quality

[0075] Deep clustering algorithms can identify different types of customer groups, making it easier for service area managers to target different customer groups.

[0076] Provide personalized services to improve the overall service quality and customer satisfaction in the service area.

[0077] 5. Support operational decisions

[0078] Through accurate customer portraits, service area managers can develop more targeted marketing strategies and resource allocation plans based on the needs and behavior patterns of different customer groups, thereby improving operational efficiency and economic benefits.

[0079] Reference Figure 1 As shown, a method for creating a customer profile of a service area based on deep clustering includes:

[0080] Step S1: Obtain the customer information of the service area and perform preprocessing; the customer information of the service area includes demographic characteristics, consumption behavior data, travel habit data and vehicle data; the preprocessing includes calculating the time each vehicle stays in the service area within a certain period of time based on the difference between two consecutive exit times and entrance times according to the records of vehicles entering and leaving the service area, and analyzing missing values, duplicate values ​​and outliers in the data.

[0081] Step S2: Process the preprocessed customer group information to obtain feature variables;

[0082] Encode and normalize the feature variables;

[0083] For consumption behavior data, calculate the consumption interval, consumption frequency, total consumption amount, consumption commodity category and consumption time period of each customer within a certain time range;

[0084] For travel habit data, calculate the travel time and stay time of each customer in the service area within a certain time range;

[0085] For vehicle data, calculate the license plate type and vehicle type for each customer;

[0086] The encoding and normalization of the feature variables include frequency encoding of categorical variables, calculating the frequency of occurrence of each category and expressing it as a numerical feature; and normalizing continuous variables to eliminate dimensional differences between different features.

[0087] Step S3: Use the trained Transformer model to further extract feature variables and obtain deep features;

[0088] The loss function of the Transformer model includes reconstruction loss and clustering loss;

[0089] Specific:

[0090] Step S3.1: Build a Transformer model, define the error between the output of the decoding stage and the original data as the reconstruction loss, and minimize the reconstruction loss; use the clustering loss to guide the representation learning process and minimize the clustering loss;

[0091] Step S3.2: Use the constructed Transformer model to perform preliminary feature extraction to obtain deep features; the deep features can capture complex patterns and relationships in the data;

[0092] Step S3.3: Train the Transformer model and optimize the reconstruction loss and clustering loss. In each iteration, adjust the model parameters according to the current clustering results and use the feedback mechanism to optimize the model performance.

[0093] Step S4: Use the K-means algorithm to perform cluster analysis on the customer information of the service area and output the clustering results; during the K-means clustering process, the algorithm gradually adjusts the position of each cluster center through iterative optimization and assigns each data point to the cluster center closest to it.

[0094] Step S5: Complete the evaluation of clustering quality by calculating the silhouette coefficient to obtain the evaluation result; optimize and adjust the Transformer model according to the evaluation result, and repeat steps S3-S4 until the evaluation result meets the expected value;

[0095] The silhouette coefficient measures the effect of clustering by comparing the closeness of each data point to its cluster and the degree of separation from other clusters. For a certain data point, the calculation formula of the silhouette coefficient is:

[0096]

[0097] Among them, a is the average distance from the data point to other data points in its cluster, and b is the average distance from the data point to the nearest other cluster. The closer the silhouette coefficient is to 1, the closer the data point is to other data points in its cluster and the more separated it is from data points in other clusters. If the silhouette coefficient is close to -1, it indicates that the data point may be incorrectly assigned to the current cluster.

[0098] Step S6: Categorize the customer information of the service area into multiple types according to the clustering results to construct customer profiles. The consumption behaviors, stay habits and vehicle information of the multiple types of customers are not completely the same.

[0099] The above are basic embodiments of the present invention. The technical solution of the present invention is further described below through two preferred embodiments.

[0100] Example 1

[0101] Reference Figure 2 As shown, a method for creating a customer profile of a service area based on deep clustering includes:

[0102] 1. Data collection and preprocessing

[0103] Obtain service area customer information from cameras and other data sources in highway service areas, including demographic characteristics (customer age, gender, etc.), consumer behavior data (category, quantity, amount, payment method, frequency of consumption, etc. of purchased goods), travel habit data (customer travel time, length of stay, etc.), and vehicle data (license plate number, license plate type, license plate color, vehicle type, whether it is a hazardous chemical vehicle, etc.). According to the vehicle entry and exit records, use the difference between two consecutive exit times and entry times to calculate the stay time of each vehicle in the service area within a certain time range (such as one month). Analyze missing values, duplicate values, and outliers in the data. Missing values ​​are processed by post-value interpolation to ensure the continuity and integrity of the data. For outliers, filter data with a stay time greater than 0 to exclude possible entry errors and abnormal records.

[0104] 2. Feature Engineering

[0105] Perform feature processing on consumption behavior, travel habits, and vehicle information data of service area customers.

[0106] For consumption behavior, calculate each customer's consumption interval, consumption frequency, total consumption amount, main consumption commodity categories, consumption time period, etc. within a certain time range (e.g., one month);

[0107] For travel habits, calculate the travel time (such as whether it is a weekend or holiday) and the length of stay of each customer in the service area within a certain time range; for vehicle data, calculate the license plate type (such as: domestic, non-local) and vehicle type (such as: new energy or fuel vehicle) of each customer. After feature processing, encode and normalize the variables.

[0108] Among them, frequency encoding is performed on categorical variables. Frequency encoding is a method of converting categorical variables into numerical features. By calculating the frequency of each category, it is represented as a numerical feature. This method can effectively retain the information of categorical variables and make it suitable for subsequent deep learning models. Continuous variables are normalized to eliminate the dimensional differences between different features and prevent certain features from having too much impact on model training.

[0109] 3. Transformer model optimization

[0110] Build a Transformer model to extract deep features of the customer base in the service area.

[0111] The loss function of the model consists of reconstruction loss and clustering loss. Reconstruction loss is the error between the result output by the Transformer model in the decoding phase and the original data. By minimizing the reconstruction loss, the model can remove noise from the data and reveal its underlying structure, thereby improving the quality of feature representation. Clustering loss is used to guide the representation learning process to make samples more distinguishable in the latent space. By minimizing the clustering loss, the model can learn more discriminative features, making samples of the same type more clustered in the latent space, thereby improving the clustering effect.

[0112] 4. Feature extraction and determination of initial cluster centers

[0113] The constructed Transformer model is used to perform preliminary feature extraction to obtain the representation of deep features. These deep features can better capture the complex patterns and relationships in the data and have higher representation capabilities than traditional shallow features. Then, the K-means algorithm is applied to determine the initial clustering center in the extracted feature space.

[0114] 5. Transformer model training and hyperparameters

[0115] The Transformer model optimized in step 3 is used for training, and the reconstruction loss and clustering loss are optimized at the same time. In each iteration, the model parameters are adjusted according to the current clustering results, and the model performance is continuously optimized using the feedback mechanism. In each round of training, the K-means algorithm is used to update the cluster center, so that the model can gradually improve the clustering accuracy during the continuous optimization process. At the same time, hyperparameter optimization is performed to select the best model parameter settings to improve the model effect.

[0116] 6. Clustering with K-means model

[0117] After deep feature extraction is completed, the K-means algorithm is used to perform cluster analysis on customer information. In the process of K-means clustering, the algorithm gradually adjusts the position of each cluster center through iterative optimization, and assigns each data point to the cluster center closest to it. Specifically, in each iteration, the K-means algorithm first calculates the Euclidean distance from each data point to each cluster center, and then assigns the data point to the cluster center closest to it. Then, the algorithm updates the position of each cluster center to make it equal to the mean of all data points in the current cluster. This process is repeated until the change of the cluster center tends to stabilize, that is, when the position of the cluster center no longer changes significantly in several consecutive iterations, the algorithm stops iterating and outputs the final clustering result.

[0118] 7. Evaluation of clustering effect

[0119] After completing K-means clustering, the clustering quality is evaluated by calculating the silhouette coefficient. The silhouette coefficient is a commonly used clustering quality evaluation indicator with a value range of -1 to 1. Specifically, the silhouette coefficient measures the clustering effect by comparing the closeness of each data point to its cluster and the degree of separation from other clusters. For a certain data point, the calculation formula of the silhouette coefficient is:

[0120]

[0121] Among them, a is the average distance from the data point to other data points in its cluster, and b is the average distance from the data point to the nearest other cluster. The closer the silhouette coefficient is to 1, the closer the data point is to other data points in its cluster and the more separated it is from data points in other clusters, indicating a better clustering effect; on the contrary, if the silhouette coefficient is close to -1, it indicates that the data point may be incorrectly assigned to the current cluster, and the clustering effect is poor. The silhouette coefficient is not only used to evaluate the final quality of clustering, but also to guide model optimization and parameter adjustment.

[0122] 8. Construction of customer profiles in service areas

[0123] According to the clustering results, the customer information of the service area is divided into multiple types (such as 3 types). Each type of customer group has differences in consumption behavior (such as consumption amount and frequency), stay habits (such as stay time), vehicle information (such as license plate number, license plate type, license plate color, etc.). Through in-depth analysis of these differences, detailed customer group portraits can be constructed. These portraits can help managers better understand the needs and behavioral characteristics of different customer groups, so as to formulate more accurate management and service strategies.

[0124] The deep clustering algorithm of the present invention shows significant advantages in constructing customer portraits in highway service areas. It automatically extracts deep features from customer information through a deep learning model, significantly improving the efficiency of customer portrait construction. The deep learning model can automatically extract high-dimensional and discriminative deep features from raw data, greatly reducing manual intervention and significantly improving the degree of automation in data processing.

[0125] Example 2

[0126] A system for profiling customer groups in service areas based on deep clustering, comprising:

[0127] Module M1: Obtain customer information of the service area and perform preprocessing;

[0128] Module M2: Process the pre-processed customer group information to obtain feature variables;

[0129] Encode and normalize the feature variables;

[0130] Module M3: Use the trained model to further extract feature variables and obtain deep features;

[0131] The loss function of the model includes reconstruction loss and clustering loss;

[0132] Module M4: cluster analysis of customer information in the service area and output of clustering results;

[0133] Module M5: Complete the clustering quality evaluation by calculating the silhouette coefficient and obtain the evaluation result; optimize and adjust the model according to the evaluation result, and repeat modules M3-M4 until the evaluation result meets the expected value;

[0134] Module M6: Classify the customer information of the service area into multiple types based on the clustering results and construct a customer profile.

[0135] The customer information of the service area includes demographic characteristics, consumption behavior data, travel habit data and vehicle data; the preprocessing includes calculating the time each vehicle stays in the service area within a certain period of time based on the difference between two consecutive exit times and entrance times according to the records of vehicles entering and leaving the service area, and analyzing missing values, duplicate values ​​and outliers in the data.

[0136] The module M2 comprises:

[0137] For consumption behavior data, calculate the consumption interval, consumption frequency, total consumption amount, consumption commodity category and consumption time period of each customer within a certain time range;

[0138] For travel habit data, calculate the travel time and stay time of each customer in the service area within a certain time range;

[0139] For vehicle data, calculate the license plate type and vehicle type for each customer;

[0140] The encoding and normalization of the feature variables include frequency encoding of categorical variables, calculating the frequency of occurrence of each category and expressing it as a numerical feature; and normalizing continuous variables to eliminate dimensional differences between different features.

[0141] The module M3 comprises:

[0142] Module M3.1: Build a Transformer model, define the error between the output of the decoding stage and the original data as the reconstruction loss, and minimize the reconstruction loss; use the clustering loss to guide the representation learning process and minimize the clustering loss;

[0143] Module M3.2: Use the built Transformer model to perform preliminary feature extraction to obtain deep features; the deep features can capture complex patterns and relationships in the data;

[0144] Module M3.3: Train the Transformer model and optimize the reconstruction loss and clustering loss. In each iteration, adjust the model parameters according to the current clustering results and use the feedback mechanism to optimize the model performance.

[0145] The module M4 includes that during the K-means clustering process, the algorithm gradually adjusts the position of each cluster center through iterative optimization and assigns each data point to the cluster center closest to it.

[0146] The silhouette coefficient measures the effect of clustering by comparing the closeness of each data point to its cluster and the degree of separation from other clusters. For a certain data point, the calculation formula of the silhouette coefficient is:

[0147]

[0148] Among them, a is the average distance from the data point to other data points in its cluster, and b is the average distance from the data point to the nearest other cluster. The closer the silhouette coefficient is to 1, the closer the data point is to other data points in its cluster and the more separated it is from data points in other clusters. If the silhouette coefficient is close to -1, it indicates that the data point may be incorrectly assigned to the current cluster.

[0149] In the module M6, the consumption behaviors, stay habits and vehicle information of various types of customer groups are not completely the same.

[0150] The present invention utilizes deep clustering algorithms to perform customer segmentation, which can more accurately identify customer groups and improve the accuracy of customer portraits, enabling companies to gain an in-depth understanding of the behavioral characteristics and needs of different customer groups, formulate more accurate marketing strategies, enhance user experience and the profitability of service areas, help companies optimize resource allocation and service processes, and improve operational efficiency.

[0151] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0152] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for creating a customer profile of a service area based on deep clustering, characterized in that: include: Step S1: Obtain customer group information of the service area and perform preprocessing; Step S2: Process the preprocessed customer group information to obtain feature variables; Encode and normalize the feature variables; Step S3: Use the trained model to further extract feature variables to obtain deep features; The loss function of the model includes reconstruction loss and clustering loss; Step S4: Perform cluster analysis on the customer group information of the service area and output the clustering results; Step S5: Complete the evaluation of clustering quality by calculating the silhouette coefficient to obtain the evaluation result; optimize and adjust the model according to the evaluation result, and repeat steps S3-S4 until the evaluation result meets the expected value; Step S6: Classify the customer information of the service area into multiple types according to the clustering results to construct a customer profile.

2. The method for creating a customer profile of a service area based on deep clustering according to claim 1, characterized in that: The customer information of the service area includes demographic characteristics, consumption behavior data, travel habit data and vehicle data; the preprocessing includes calculating the time each vehicle stays in the service area within a certain period of time based on the difference between two consecutive exit times and entrance times according to the records of vehicles entering and leaving the service area, and analyzing missing values, duplicate values ​​and outliers in the data.

3. The method for creating a customer profile of a service area based on deep clustering according to claim 2, characterized in that: The step S2 comprises: For consumption behavior data, calculate the consumption interval, consumption frequency, total consumption amount, consumption commodity category and consumption time period of each customer within a certain time range; For travel habit data, calculate the travel time and stay time of each customer in the service area within a certain time range; For vehicle data, calculate the license plate type and vehicle type for each customer; The encoding and normalization of the feature variables include frequency encoding of categorical variables, calculating the frequency of occurrence of each category and expressing it as a numerical feature; and normalizing continuous variables to eliminate dimensional differences between different features.

4. The method for creating a customer profile of a service area based on deep clustering according to claim 1, characterized in that: The step S3 comprises: Step S3.1: Build a Transformer model, define the error between the output of the decoding stage and the original data as the reconstruction loss, and minimize the reconstruction loss; use the clustering loss to guide the representation learning process and minimize the clustering loss; Step S3.2: Use the constructed Transformer model to perform preliminary feature extraction to obtain deep features; the deep features can capture complex patterns and relationships in the data; Step S3.3: Train the Transformer model and optimize the reconstruction loss and clustering loss. In each iteration, adjust the model parameters according to the current clustering results and use the feedback mechanism to optimize the model performance.

5. The method for creating a customer profile of a service area based on deep clustering according to claim 4, characterized in that: The step S4 includes that during the K-means clustering process, the algorithm gradually adjusts the position of each cluster center through iterative optimization, and assigns each data point to the cluster center closest to it.

6. The method for creating a customer profile of a service area based on deep clustering according to claim 5, characterized in that: The silhouette coefficient measures the effect of clustering by comparing the closeness of each data point to its cluster and the degree of separation from other clusters. For a certain data point, the calculation formula of the silhouette coefficient is: Among them, a is the average distance from the data point to other data points in its cluster, and b is the average distance from the data point to the nearest other cluster. The closer the silhouette coefficient is to 1, the closer the data point is to other data points in its cluster and the more separated it is from data points in other clusters. If the silhouette coefficient is close to -1, it indicates that the data point may be incorrectly assigned to the current cluster.

7. The method for creating a customer profile of a service area based on deep clustering according to claim 1, characterized in that: In step S6, the consumption behaviors, stay habits and vehicle information of the various types of customer groups are not completely the same.

8. A system for creating customer profiles in service areas based on deep clustering, characterized in that: include: Module M1: Obtain customer information of the service area and perform preprocessing; Module M2: Process the pre-processed customer group information to obtain feature variables; Encode and normalize the feature variables; Module M3: Use the trained model to further extract feature variables and obtain deep features; The loss function of the model includes reconstruction loss and clustering loss; Module M4: cluster analysis of customer information in the service area and output of clustering results; Module M5: Complete the clustering quality evaluation by calculating the silhouette coefficient and obtain the evaluation result; optimize and adjust the model according to the evaluation result, and repeat modules M3-M4 until the evaluation result meets the expected value; Module M6: Classify the customer information of the service area into multiple types based on the clustering results and construct a customer profile.

9. The system for service area customer profile based on deep clustering according to claim 8, characterized in that: The customer information of the service area includes demographic characteristics, consumption behavior data, travel habit data and vehicle data; the preprocessing includes calculating the time each vehicle stays in the service area within a certain period of time based on the difference between two consecutive exit times and entrance times according to the records of vehicles entering and leaving the service area, and analyzing missing values, duplicate values ​​and outliers in the data.

10. The system for service area customer profile based on deep clustering according to claim 9, characterized in that: The module M2 comprises: For consumption behavior data, calculate the consumption interval, consumption frequency, total consumption amount, consumption commodity category and consumption time period of each customer within a certain time range; For travel habit data, calculate the travel time and stay time of each customer in the service area within a certain time range; For vehicle data, calculate the license plate type and vehicle type for each customer; The encoding and normalization of the feature variables include frequency encoding of categorical variables, calculating the frequency of occurrence of each category and expressing it as a numerical feature; and normalizing continuous variables to eliminate dimensional differences between different features.

Images