Personalized service management system and method based on user portrait analysis

Abstract

The invention discloses a personalized service management system and method based on user portrait analysis, and relates to the related technical field of big data analysis, machine learning and personalized services, and the system is composed of a user information collection module, a data preprocessing module, a user portrait construction module, a service database module, a service recommendation module, a service providing module and a service evaluation module. The user information collection module collects identity, behavior and feedback information; the data preprocessing module performs cleaning, normalization and feature extraction; the user portrait construction module constructs and updates a user portrait by using an auto-encoder; the service database module stores and retrieves service information; the service recommendation module is combined with a recommendation algorithm based on content and collaborative filtering to provide personalized recommendation and dynamically adjust algorithm details; the service providing module schedules and executes services; the service evaluation module collects evaluation information, calculates a comprehensive score and predicts an evaluation trend. All the modules operate cooperatively, personalized services are provided for users, and continuous optimization is achieved.

Description

Technical Field

[0001] The present invention relates to the technical fields of big data analysis, machine learning, and personalized services, and specifically, to a personalized service management system and method based on user portrait analysis. Background Art

[0002] In the current situation of the deep integration of digitization and informatization, the popularization of the Internet has enabled various platforms to accumulate a large amount and diverse types of user data. How to effectively mine and utilize this data to provide accurate and personalized services to users has become a key issue actively explored in many industries.

[0003] When facing the increasing personalized needs of users, the traditional general service model seems powerless. This model cannot accurately adapt to the unique preferences, behavior habits, and consumption capabilities of different users, resulting in a poor user experience and a significant reduction in service effects. Taking the e-commerce industry as an example, the one-size-fits-all product recommendation strategy often fails to meet the actual needs of different users, making it difficult for users to quickly find their favorite products among a vast number of goods. This not only reduces the efficiency and satisfaction of users' shopping but also may cause user loss.

[0004] In addition, with the increasingly fierce market competition, enterprises are urgently in need of enhancing user stickiness and expanding market share by providing personalized services. However, the current personalized service technologies still have limitations in many aspects. For example, in the data processing link, some technologies fail to fully explore the potential value of user data, resulting in an insufficient understanding of users; in the model construction aspect, some models cannot effectively cope with the dynamic changes of user behavior, making the recommendation results lag behind the evolution of user interests and reducing the accuracy and timeliness of recommendations.

[0005] In summary, developing a management system that can deeply analyze user portraits and accurately provide personalized services is of great practical significance and urgency for meeting the increasing personalized needs of users, enhancing the competitiveness of enterprises, and promoting the sustainable development of the industry.

[0006] To solve the above problems, the present invention proposes a personalized service management system and method based on user portrait analysis to more effectively process user data, accurately construct user portraits, provide personalized services, meet the diverse needs of users, and enhance the competitiveness of enterprises. Summary of the Invention

[0007] The purpose of the present invention is to provide a personalized service management system and method based on user portrait analysis to solve the problems raised in the prior art.

[0008] To achieve the above purpose, the present invention provides the following technical solutions:

[0009] A personalized service management system based on user profile analysis, including a user information collection module, a data preprocessing module, a user profile construction module, a service database module, a service recommendation module, a service provision module, and a service evaluation module; the user information collection module is used to collect the user's identity information, behavior data, and user feedback information; the data preprocessing module is used to perform data cleaning, normalization processing, and feature extraction on the collected user information; the user profile construction module constructs a user profile using the preprocessed user information; the service database module is used to store all the information provided by the system; the service recommendation module is used to provide personalized services to users based on the user profile and service database information; the service provision module is used to provide the recommended services to users; the service evaluation module is used to collect the user's evaluation information on the services.

[0010] The user information collection module includes a user identity information collection unit, a user behavior data collection unit, and a user feedback collection information unit;

[0011] The user identity information collection unit is used to collect the user's basic identification information, including name, gender, age, contact information, and address when the user registers the system; through this information, the system preliminarily classifies the user types according to age and gender factors, providing a basic framework for subsequent service recommendations;

[0012] The user behavior data collection unit is used to monitor and record the user's behavior in the system in real time. By collecting the user's browsing history, the system will push content according to the user's interest focus;

[0013] The user feedback information collection unit is used to collect the user's views on the system services in the form of evaluations, comments, and scores.

[0014] The data preprocessing module includes a data cleaning unit, a data normalization unit, and a feature extraction unit;

[0015] The data cleaning unit is used to set the reasonable range and format rules of the data, and identify and process the noise data, error data, and missing data in the collected user information;

[0016] The data normalization unit is used to convert data with different dimensions and value ranges to the same scale, making different features equally important in subsequent analysis; the calculation method is as follows:

[0017] For each data point x, according to the given data column [x 1 , x 2 ,..., x n , find the maximum value x max and the minimum value x min, and then for each x i Normalize it, and calculate the normalized value x according to the formula norm , and the calculation formula is as follows:

[0018]

[0019] The feature extraction unit uses the normalized data matrix to calculate its covariance matrix [m×m], revealing the correlation between features; then by solving the eigenvalues and eigenvectors of the covariance matrix, where the total number of eigenvalues is m, find the main directions of data change and the variance magnitudes in these directions; according to the magnitudes of the eigenvalues, select the first k eigenvalues and their corresponding eigenvectors as the main directions of data change, where k < m; project the normalized data matrix onto the selected principal components to convert high-dimensional data into low-dimensional data, realizing dimensionality reduction processing of the data and extracting the main features of user information.

[0020] The user portrait construction module includes an autoencoder construction and training unit, a user portrait generation unit, and an incremental learning and updating unit;

[0021] The autoencoder construction and training unit is used to initialize the autoencoder network structure and train it using the preprocessed data, learning the key features of user information by minimizing the reconstruction error; first receive the user data X that has been cleaned, normalized, and feature-extracted output by the data preprocessing module, build an autoencoder network, the encoder converts the input data X into a low-dimensional hidden layer representation h, and the decoder reconstructs the hidden layer representation h into an approximation of the original data Then set the network parameters, including the number of neurons in each layer, activation function, learning rate, optimizer, and use the mean square error as the loss function L, and the loss function is expressed as

[0022]

[0023] where xi is the i-th element of the original user information X, is the i-th element of the reconstructed information ;

[0024] Then divide the dataset, divide the data X into a training set, a validation set, and a test set according to a preset ratio, use the training set to train the autoencoder, input the training set data into the encoder to obtain the hidden layer representation h, and then obtain the reconstructed data through the decoder Calculate the reconstruction error loss function L, and use the optimizer to update the autoencoder parameters according to the loss function to minimize the reconstruction error;

[0025] The user profile generation unit converts the input data into a user profile vector containing multi-dimensional user features based on the trained autoencoder. First, the trained autoencoder is used to separate its encoder part as a tool for generating the user profile, and the user profile vector P is obtained. The user profile vector P = [p 1 , p 2 ,..., p n . Each element p i in it represents the feature value of the user in the i-th dimension, and these feature values reflect the multi-dimensional information of the user. The original data is converted into the user profile, and the expression is P = f(x), where f is the mapping function from the original data to the user profile.

[0026] The incremental learning and updating unit updates the autoencoder parameters and the user profile without retraining the entire model when new user information is added. First, the new user information X new is obtained and input into the encoder part of the trained autoencoder to get the new hidden layer representation h new . Then, only the decoder part is adjusted to construct a new model that only contains the decoder. The input is the hidden layer representation h new of the new data, and the output is the reconstructed new data. Then, the reconstruction error of the new data is used as the update basis. Finally, by methods such as mini-batch gradient descent, only some parameters of the decoder are updated to minimize the new reconstruction error, realizing the update of the user profile.

[0027] The service database module includes a service information storage unit and a service information retrieval and screening unit.

[0028] The service information storage unit is used to store all the information provided by the system into the data storage structure.

[0029] The service information retrieval and screening unit is used to retrieve the service information that meets the user's needs from the stored service information according to different screening conditions.

[0030] The service recommendation module is connected to the user profile construction module and the service database module, and recommends personalized services for users according to the user profile and service database information. It includes a hybrid recommendation unit and a recommendation algorithm adjustment unit.

[0031] The hybrid recommendation unit combines the content-based recommendation algorithm and the collaborative filtering recommendation algorithm to generate personalized service recommendations for users. First, the content-based recommendation algorithm and the collaborative filtering recommendation algorithm are run simultaneously. The content-based recommendation algorithm will look for similar service content in the service database according to the information in the user profile and generate the content-based recommendation result R cb, and then, in conjunction with the collaborative filtering recommendation algorithm, based on the similarity between the user profile and other user profiles, identify the services liked by similar users to obtain the collaborative filtering recommendation result R cf ; Then, according to the user activity A u dynamically adjust their weights w(A u ), and finally add R cb and R cf according to the calculated weights to obtain the final recommendation result R. The calculation formula is as follows:

[0032] R = w(A u ) * R cb + (1 - w(A u )) * R cf ;

[0033] The recommendation algorithm adjustment unit is used to dynamically adjust the specific implementation details of the content-based recommendation and collaborative filtering recommendation algorithms according to different service categories and user preferences, and adopt the method of optimizing the weight function according to the characteristics in the user profile and the characteristics of the service to optimize the content similarity calculation method in the content-based recommendation and the neighbor selection criteria in the collaborative filtering recommendation.

[0034] The service providing module includes a service scheduling unit and a service execution unit;

[0035] The service scheduling unit is used to receive the recommended service information transmitted by the service recommendation module and the user demand information obtained from the user profile, and call the resources required for service provision according to the service characteristics and user demands;

[0036] The service execution unit obtains the detailed instructions for service execution from the service scheduling unit, including the required resources, service provision speed, and specific content of the service, and then integrates and configures the required resources according to the scheduling instructions.

[0037] The service evaluation module includes an evaluation information collection unit and an evaluation analysis unit;

[0038] The evaluation information collection unit first provides a feedback channel for user evaluation information, sets an evaluation form on the service completion page, where users can rate the satisfaction of the service in the form and fill in the service completion degree; at the same time, it can also collect users' comments and suggestion information; then standardize the evaluation information input by users, summarize the evaluation information of different users into an evaluation information set V, and finally update the evaluation information set in a timely manner when new user evaluation information is generated to ensure the timeliness and integrity of the evaluation information;

[0039] The evaluation and analysis unit calculates evaluation information using the evaluation function V = h based on the user's behavior and user information after using the recommendation service, where h represents the user's satisfaction score and service completion rate, and then calculates the comprehensive evaluation score using the weighted average evaluation formula. The calculation formula is as follows:

[0040] S a = w 1 *S + w 2 *C

[0041] Where S is the satisfaction score, C is the service completion rate, w 1 and w 2 are weight coefficients, and w 1 + w 2 = 1; The values of w 1 and w 2 can be dynamically adjusted according to different service types and user groups.

[0042] A personalized service management method based on user portrait analysis is applied to a personalized service management system according to any one of claims 1-8. It is characterized by the following steps:

[0043] S1. Through the user identity information collection unit, user behavior data collection unit, and user feedback information collection unit of the user information collection module, collect the basic identification information when the user registers, the behavior data within the system, and the evaluation, comment, and scoring information of the user on the system service respectively;

[0044] S2. The data cleaning unit, data normalization unit, and feature extraction unit of the data preprocessing module process the collected user information, including setting data rules, processing error and missing data, unifying the data scale, and performing dimensionality reduction processing using the covariance matrix and eigenvectors;

[0045] S3. The autoencoder construction and training unit of the user portrait construction module receives the preprocessed data, constructs and trains the autoencoder, learns key features by minimizing the reconstruction error, the user portrait generation unit uses the trained autoencoder to convert the data into a user portrait vector containing multi-dimensional features, and the incremental learning update unit updates the autoencoder parameters and user portrait when new user information is added;

[0046] S4. The service information storage unit of the service database module stores the system service information, and the service information retrieval and screening unit retrieves and screens the service information according to the requirements;

[0047] S5. The hybrid recommendation unit of the service recommendation module combines content-based recommendation and collaborative filtering recommendation algorithms. Based on the user profile and service database information, it adjusts the weights according to user activity to obtain the final recommendation result. The recommendation algorithm adjustment unit dynamically adjusts the specific details of the two recommendation algorithms.

[0048] S6. The service scheduling unit of the service provision module schedules the resources required for the service according to the service recommendation information and user requirements. The service execution unit executes the service, integrates and configures the resources, and delivers the service.

[0049] S7. The evaluation information collection unit of the service evaluation module collects information such as user evaluations, comments, and suggestions on the service completion page and organizes and updates the evaluation set. The evaluation analysis unit calculates the comprehensive evaluation score using the evaluation function and weighted average formula.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] 1. Precise recommendation and improved user experience: The present invention collects user information from multiple dimensions, constructs an accurate user profile using an autoencoder, and adopts a hybrid recommendation algorithm that combines content-based and collaborative filtering recommendations, dynamically adjusting the weights according to user activity. This enables the service recommendation module to deeply understand user needs and provide personalized services that highly match their preferences for users.

[0052] 2. Efficient data processing and optimized resource allocation: The data preprocessing module cleans, normalizes, and extracts features from massive user data, effectively removing noise and redundant information and improving data quality. The user profile construction module uses an autoencoder to learn key features, achieving data dimensionality reduction and efficient utilization. The service scheduling unit reasonably allocates resources according to service characteristics and user requirements to ensure efficient service provision.

[0053] 3. Continuous optimization and enhanced service competitiveness: The service evaluation module comprehensively collects information such as user satisfaction scores and service completion degrees, calculates the comprehensive evaluation score using the evaluation function and weighted average formula, and dynamically adjusts the weights according to different service types and user groups. The system can optimize the service recommendation strategy in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is an organizational structure diagram of a personalized service management system based on user profile analysis according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0055] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0056] Embodiment: As Figure 1 shown, the present invention provides a technical solution,

[0057] A personalized service management system based on user portrait analysis, including a user information collection module, a data preprocessing module, a user portrait construction module, a service database module, a service recommendation module, a service provision module, and a service evaluation module; the user information collection module is used to collect the user's identity information, behavior data, and user feedback information; the data preprocessing module is used to perform data cleaning, normalization processing, and feature extraction on the collected user information; the user portrait construction module uses the preprocessed user information to construct a user portrait; the service database module is used to store all the information provided by the system; the service recommendation module is used to provide personalized services for users based on the user portrait and service database information; the service provision module is used to provide the recommended services to users; the service evaluation module is used to collect the evaluation information of users on the services.

[0058] The user information collection module includes a user identity information collection unit, a user behavior data collection unit, and a user feedback collection information unit;

[0059] The user identity information collection unit is used to collect the user's basic identification information, including name, gender, age, contact information, and address when the user registers the system; through this information, the system initially classifies the user types according to age and gender factors, providing a basic framework for subsequent service recommendations;

[0060] The user behavior data collection unit is used to monitor and record the user's behavior in the system in real time. By collecting the user's browsing history, the system will push content according to the user's interest focus;

[0061] The user feedback information collection unit is used to collect the user's views on the system services in the form of evaluations, comments, and scores.

[0062] The data preprocessing module includes a data cleaning unit, a data normalization unit, and a feature extraction unit;

[0063] The data cleaning unit is used to set the reasonable range and format rules of the data, and identify and process the noise data, error data, and missing data in the collected user information;

[0064] The data normalization unit is used to convert data with different dimensions and value ranges to the same scale, so that different features have the same importance in subsequent analysis. The calculation method is as follows:

[0065] For each data point x, according to the given data column [x 1 , x 2 ,..., x n , find the maximum value x max and the minimum value x min of this column. Then, normalize each x i , and calculate the normalized value x norm according to the formula. The calculation formula is as follows:

[0066]

[0067] The feature extraction unit uses the normalized data matrix to calculate its covariance matrix [m×m], revealing the correlation between features. Then, by solving the eigenvalues and eigenvectors of the covariance matrix, where the total number of eigenvalues is m, find the main directions of data change and the variance magnitudes in these directions. According to the magnitudes of the eigenvalues, select the top k eigenvalues and their corresponding eigenvectors as the main directions of data change, where k < m. Project the normalized data matrix onto the selected principal components to convert high-dimensional data into low-dimensional data, achieving dimensionality reduction of the data and extracting the main features of user information.

[0068] The user portrait construction module includes an autoencoder construction and training unit, a user portrait generation unit, and an incremental learning and updating unit;

[0069] The autoencoder construction and training unit is used to initialize the autoencoder network structure and train it using the preprocessed data, learning the key features of user information by minimizing the reconstruction error. First, receive the user data X that has been cleaned, normalized, and feature-extracted output by the data preprocessing module, and build an autoencoder network. The encoder converts the input data X into a low-dimensional hidden layer representation h, and the decoder reconstructs the hidden layer representation h into an approximation of the original data Then, set the network parameters, including the number of neurons in each layer, activation function, learning rate, optimizer, and use the mean squared error as the loss function L. The loss function is expressed as

[0070]

[0071] where xi is the i-th element of the original user information X, is the i-th element of the reconstructed information ;

[0072] Then divide the dataset, split the data X into a training set, a validation set, and a test set according to a preset ratio, use the training set to train the autoencoder, input the training set data into the encoder to obtain the hidden layer representation h, and then obtain the reconstructed data through the decoder. Calculate the reconstruction error loss function L, and use the optimizer to update the autoencoder parameters according to the loss function to minimize the reconstruction error.

[0073] The user profile generation unit, based on the trained autoencoder, converts the input data into a user profile vector containing multi-dimensional features of the user; first, use the trained autoencoder, separate its encoder part as a tool for generating the user profile, and obtain the user profile vector P; the user profile vector P = [p 1 , p 2 ,..., p n . Each element p i in it represents the feature value of the user in the i-th dimension, and these feature values reflect the multi-dimensional information of the user; convert the original data into a user profile, and the expression is P = f(x), where f is the mapping function from the original data to the user profile.

[0074] The incremental learning update unit, when new user information is added, updates the autoencoder parameters and the user profile without retraining the entire model; first, obtain the new user information X new , input it into the encoder part of the trained autoencoder, and obtain the new hidden layer representation h new ; then only adjust the decoder part, construct a new model that only contains the decoder, input the hidden layer representation h of the new data new , and the output is the reconstructed new data; then use the reconstruction error of the new data as the update basis; finally, through methods such as mini-batch gradient descent, only update some parameters of the decoder to minimize the new reconstruction error and achieve the update of the user profile.

[0075] The service database module includes a service information storage unit and a service information retrieval and screening unit.

[0076] The service information storage unit is used to store all the information provided by the system into the data storage structure.

[0077] The service information retrieval and screening unit is used to retrieve the service information that meets the user's needs from the stored service information according to different screening conditions.

[0078] The service recommendation module is connected to the user profile construction module and the service database module, and recommends personalized services for users according to the user profile and service database information; it includes a hybrid recommendation unit and a recommendation algorithm adjustment unit.

[0079] The hybrid recommendation unit combines the content-based recommendation algorithm and the collaborative filtering recommendation algorithm to generate personalized service recommendations for users. First, the content-based recommendation algorithm and the collaborative filtering recommendation algorithm are run simultaneously. The content-based recommendation algorithm will, according to the information in the user profile, search for similar service content in the service database to generate the content-based recommendation result R cb , and then the collaborative filtering recommendation algorithm will, according to the similarity between the user profile and other user profiles, find the services liked by similar users to obtain the collaborative filtering recommendation result R cf ; Then, according to the user activity A u , dynamically adjust their weights w(A u ), and finally add R cb and R cf according to the calculated weights to obtain the final recommendation result R. The calculation formula is as follows:

[0080] R = w(A u ) * R cb + (1 - w(A u )) * R cf ;

[0081] The recommendation algorithm adjustment unit is used to dynamically adjust the specific implementation details of the content-based recommendation and the collaborative filtering recommendation algorithm according to different service categories and user preferences, and optimize the content similarity calculation method in the content-based recommendation and the neighbor selection criteria in the collaborative filtering recommendation by using the method of optimizing the weight function according to the characteristics in the user profile and the characteristics of the service.

[0082] The service providing module includes a service scheduling unit and a service execution unit;

[0083] The service scheduling unit is used to receive the recommended service information transmitted by the service recommendation module and the user demand information obtained from the user profile, and call the resources required for service provision according to the service characteristics and user demands;

[0084] The service execution unit obtains the detailed instructions for service execution from the service scheduling unit, including the required resources, service providing speed, and the specific content of the service, and then integrates and configures the required resources according to the scheduling instructions.

[0085] The service evaluation module includes an evaluation information collection unit and an evaluation analysis unit;

[0086] The evaluation information collection unit first provides a feedback channel for users to evaluate information, sets up an evaluation form on the service completion page, where users can rate the satisfaction of the service and fill in the service completion degree in the form. At the same time, it can also collect users' comments and suggestions. Then, it standardizes the evaluation information input by users, aggregates the evaluation information of different users into an evaluation information set V. Finally, when new user evaluation information is generated, it updates the evaluation information set in a timely manner to ensure the timeliness and integrity of the evaluation information.

[0087] The evaluation analysis unit calculates the evaluation information using the evaluation function V = h based on the user's behavior after using the recommended service and the user information, where h represents the user's satisfaction score and service completion degree. Then, it calculates the comprehensive evaluation score using the weighted average evaluation formula. The calculation formula is as follows:

[0088] S a =w 1 *S + w 2 *C

[0089] Where S is the satisfaction score, C is the service completion degree, w 1 and w 2 are weight coefficients, and w 1 + w 2 =1; The values of w 1 and w 2 can be dynamically adjusted according to different service types and user groups.

[0090] A personalized service management method based on user portrait analysis is applied to a personalized service management system according to any one of claims 1 - 8. It is characterized by including the following steps:

[0091] S1. Through the user identity information collection unit, user behavior data collection unit, and user feedback information collection unit of the user information collection module, respectively collect the basic identification information when the user registers, the behavior data within the system, and the evaluation, comments, and scoring information of the user on the system service.

[0092] S2. The data cleaning unit, data normalization unit, and feature extraction unit of the data preprocessing module process the collected user information, including setting data rules, processing error and missing data, unifying the data scale, and performing dimensionality reduction processing using the covariance matrix and eigenvalue eigenvector.

[0093] S3. The autoencoder construction and training unit of the user profile construction module receives the preprocessed data, constructs and trains an autoencoder, learns key features by minimizing the reconstruction error. The user profile generation unit uses the trained autoencoder to convert the data into a user profile vector containing multi-dimensional features. The incremental learning and update unit updates the autoencoder parameters and user profile when new user information is added;

[0094] S4. The service information storage unit of the service database module stores the system service information. The service information retrieval and screening unit retrieves and screens the service information according to the requirements;

[0095] S5. The hybrid recommendation unit of the service recommendation module combines content-based recommendation and collaborative filtering recommendation algorithms. According to the user profile and service database information, it adjusts the weights based on user activity to obtain the final recommendation result. The recommendation algorithm adjustment unit dynamically adjusts the specific details of the two recommendation algorithms;

[0096] S6. The service scheduling unit of the service provision module schedules the resources required for the service according to the service recommendation information and user requirements. The service execution unit executes the service, integrates and configures the resources and delivers the service;

[0097] S7. The evaluation information collection unit of the service evaluation module collects information such as user evaluations, comments, and suggestions on the service completion page and organizes and updates the evaluation set. The evaluation analysis unit calculates the comprehensive evaluation score using the evaluation function and weighted average formula.

[0098] Suppose there is an online education platform. When user Li registers on the online education platform, the system collects Li's identity information, assumed to be a 22-year-old male, contact information, and address. Based on the age, the system initially determines that Li is in the university study stage. Being male, he may have more preferences for science and engineering courses, laying a foundation for subsequent recommendations;

[0099] Li browsed multiple artificial intelligence-related courses on the platform and frequently watched teaching videos on machine learning algorithms. The behavior data collection unit recorded these browsing histories, indicating that Li is highly interested in the field of artificial intelligence, especially in the direction of machine learning.

[0100] After Li completed a Python programming course, he gave a satisfaction score of 8 in the evaluation form and mentioned in the comments that the course content was rich, but the guidance on practical projects could be strengthened. The feedback information collection unit included these evaluations and suggestions.

[0101] Among the user information collected by the platform, some age data had outliers, such as negative numbers. The data cleaning unit screened according to the reasonable age range (0 - 120 years old) and corrected the incorrect data. At the same time, it unified and standardized the contact information format to ensure data quality.

[0102] The course data of the platform includes data with different dimensions such as course difficulty scores (1 - 5 points) and course durations (hours). For the course difficulty score column, assuming the maximum value is 5 and the minimum value is 1, the course difficulty data is normalized so that it has the same importance as other data such as course duration in subsequent analyses.

[0103] The feature extraction unit constructs the normalized user data into a matrix, calculates the covariance matrix, and reveals the correlations between different features (such as age, course preferences, learning duration, etc.). By solving the eigenvalues and eigenvectors of the covariance matrix, the eigenvectors corresponding to the first few key eigenvalues are selected to convert the high - dimensional data into low - dimensional data, and the main features of Xiaoli in terms of learning preferences, learning ability, etc. are extracted.

[0104] Then the system receives the pre - processed data of Xiaoli and builds an auto - encoder network. Set the network parameters, use ReLU as the activation function and Adam as the optimizer. Divide the data into a training set, a validation set, and a test set in the ratio of 70%, 15%, and 15%. Use the training set to train the auto - encoder, and by continuously adjusting the parameters, minimize the reconstruction error.

[0105] After the training is completed, the user portrait generation unit separates the encoder part of the auto - encoder, converts the input data of Xiaoli, and obtains the user portrait vector. This vector contains the eigenvalue features of Xiaoli in multiple dimensions such as course difficulty preferences and subject area preferences. For example, the interest score in the field of artificial intelligence is relatively high, and the acceptance of medium - to - high - difficulty courses is relatively high.

[0106] The incremental learning update unit retrieves that Xiaoli has subsequently browsed courses related to deep learning, and new user information X new is generated. Input it into the trained encoder part of the auto - encoder to obtain a new hidden layer representation h new . Build a new model for the decoder part, use the reconstruction error of the new data, and update the parameters of the decoder part through mini - batch gradient descent to realize the update of Xiaoli's user portrait, making it more accurately reflect the changes in Xiaoli's learning interests.

[0107] The service information storage unit stores all course information on the platform, including course names, lecturers, course content introductions, course difficulties, applicable populations, etc. The service information retrieval and screening unit filters out the courses suitable for him from the database according to Xiaoli's user portrait.

[0108] The hybrid recommendation unit in the service recommendation module, based on the content - based recommendation algorithm, finds similar deep - learning courses in the service database according to Xiaoli's preferences for artificial intelligence and machine learning in the user portrait, and generates content - based recommendation results R cb。The collaborative filtering recommendation algorithm finds other users similar to Xiao Li's profile, discovers some cutting-edge artificial intelligence research courses they like, and obtains the collaborative filtering recommendation result R cf 。Dynamically adjust the weights according to Xiao Li's activity on the platform (such as recent login frequency, learning duration, etc.), and combine R cb and R cf Add them according to the weights to recommend a series of advanced deep learning courses and cutting-edge research courses to Xiao Li.

[0109] Recommendation algorithm adjustment unit: For the online education service category, aiming at Xiao Li's preference for courses that combine theory with practice, optimize the content similarity calculation method in content-based recommendation, and pay more attention to the matching degree of the course practice session settings. At the same time, in collaborative filtering recommendation, adjust the neighbor selection criteria, and preferentially select users who are similar to Xiao Li in terms of learning progress and learning goals to further improve the accuracy of recommendation.

[0110] The service scheduling unit, according to the recommended course information and Xiao Li's user needs (such as hoping to learn as soon as possible and having high requirements for course quality), the scheduling unit allocates high-quality teacher resources and arranges the course to be taught live on a high-bandwidth server or provides high-definition recorded broadcast resources.

[0111] The service execution unit, according to the scheduling instructions, integrates the teaching materials, video resources, etc. required for the course, and provides Xiao Li with a smooth course learning service to ensure that he can smoothly learn the recommended courses.

[0112] After Xiao Li completes the deep learning course, he gives a satisfaction score of 9 in the evaluation form, the service completion rate is 0.95, and he suggests adding more case analyses. The evaluation information collection unit standardizes this information and updates it to the evaluation information set.

[0113] The evaluation analysis unit uses the evaluation function to calculate the evaluation information by combining Xiao Li's satisfaction score and service completion rate. Adopt the weighted average evaluation formula, and dynamically adjust the weights (such as w1 = 0.6, w2 = 0.4) according to the type of online education service and the characteristics of young student groups like Xiao Li to calculate the comprehensive evaluation score. Based on the evaluation trend analysis of a large number of users, the platform predicts that the demand for practical cases in the student group will increase, and adjusts the course settings and recommendation strategies in advance to add courses with rich cases in subsequent recommendations.

[0114] It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Accordingly, all changes that fall within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. A personalized service management system based on user portrait analysis, characterized by: It includes a user information collection module, a data preprocessing module, a user portrait construction module, a service database module, a service recommendation module, a service provision module, and a service evaluation module; The user information collection module is used to collect the user's identity information, behavior data, and user feedback information; The data preprocessing module is used to perform data cleaning, normalization processing, and feature extraction on the collected user information; The user portrait construction module constructs a user portrait using the preprocessed user information; The service database module is used to store all the information provided by the system; the service recommendation module is used to provide personalized services to users based on the user portrait and service database information; The service provision module is used to provide the recommended services to users; The service evaluation module is used to collect the evaluation information of users on the services.

2. A personalized service management system based on user portrait analysis according to claim 1, characterized in that: The user information collection module includes a user identity information collection unit, a user behavior data collection unit, and a user feedback collection information unit; The user identity information collection unit is used to collect the user's basic identification information, including name, gender, age, contact information, and address, when the user registers the system; Based on these information, the system preliminarily classifies the user types according to age and gender factors, providing a basic framework for subsequent service recommendations; The user behavior data collection unit is used to monitor and record the user's behavior in the system in real time. By collecting the user's browsing history, the system will push content according to the user's interest focus; The user feedback information collection unit is used to collect the user's views on the system services in the form of evaluations, comments, and scores.

3. The personalized service management system based on user portrait analysis according to claim 1, characterized in that: The data preprocessing module includes a data cleaning unit, a data normalization unit, and a feature extraction unit; The data cleaning unit is used to set the reasonable range and format rules of the data, and identify and process the noise data, error data, and missing data in the collected user information; The data normalization unit is used to convert data with different dimensions and value ranges to the same scale, making different features equally important in subsequent analysis; the calculation method is as follows: For each data point x, according to the given data column [x1,x2,...,x n ], find the maximum value x in the column max and the minimum value x min , then for each x i Normalization, calculate the normalized value x according to the formula norm , the calculation formula is as follows: The feature extraction unit uses the normalized data matrix to calculate its covariance matrix [m×m] to reveal the correlation between features; then, by solving the eigenvalues and eigenvectors of the covariance matrix, where the total number of eigenvalues is m, the main directions of data change and the variance sizes in these directions are found; according to the sizes of the eigenvalues, the first k eigenvalues and their corresponding eigenvectors are selected as the main directions of data change, where k < m; project the normalized data matrix onto the selected principal components to convert the high-dimensional data into low-dimensional data, realizing the dimensionality reduction processing of the data and extracting the main features of the user information.

4. The personalized service management system based on user portrait analysis according to claim 1, characterized in that: The user portrait construction module includes an autoencoder construction and training unit, a user portrait generation unit, and an incremental learning and updating unit; The autoencoder construction and training unit is used to initialize the autoencoder network structure and train it with preprocessed data to learn the key features of user information by minimizing the reconstruction error. First, the user data X after cleaning, normalization and feature extraction output by the data preprocessing module is received, and the autoencoder network is built. The encoder converts the input data X into a low-dimensional hidden layer representation h, and the decoder reconstructs the hidden layer representation h into a near-original data. Then set the network parameters, including the number of neurons in each layer, activation function, learning rate, optimizer, and mean square error as the loss function L. The loss function is expressed as Where xi is the i-th element of the original user information X, It is reconstruction information The i-th element of ; Then divide the data set, convert the data X into training set, validation set and test set according to the preset ratio, use the training set to train the autoencoder, input the training set data into the encoder to obtain the hidden layer representation h, and then pass it through the decoder to obtain the reconstructed data Calculate the reconstruction error loss function L, and use the optimizer to update the autoencoder parameters according to the loss function to minimize the reconstruction error; The user portrait generation unit, based on the trained autoencoder, converts the input data into a user portrait vector containing multi-dimensional features of the user; first, use the trained autoencoder, separate its encoder part as a tool for generating the user portrait, and obtain the user portrait vector P; User portrait vector P = [p1, p2, ..., p n Each element p in ] i Represents the feature value of the user in the i-th dimension, which reflects the user's multi-dimensional information; converts the original data to the user portrait, the expression is P = f(x), where f is the mapping function from the original data to the user portrait; The incremental learning update unit, when new user information is added, updates the autoencoder parameters and the user profile without retraining the entire model. First, the new user information X is obtained. new , input it into the encoder part of the trained autoencoder to obtain the new hidden layer representation h new ; Then only the decoder part is adjusted, and a new model containing only the decoder is constructed, with the input being the hidden layer representation h of the new data new , the output is the reconstructed new data; the reconstruction error of the new data is used as the update basis; finally, through methods such as small batch gradient descent, only some parameters of the decoder are updated to minimize the new reconstruction error and realize the update of the user portrait.

5. The personalized service management system based on user portrait analysis according to claim 1, characterized in that: The service database module includes a service information storage unit and a service information retrieval and screening unit; The service information storage unit is used to store all information provided by the system into a data storage structure; The service information retrieval and screening unit is used to retrieve service information that meets user needs from the stored service information according to different screening conditions.

6. The personalized service management system based on user portrait analysis according to claim 1, characterized in that: The service recommendation module is connected to the user portrait construction module and the service database module, and recommends personalized services to users based on the user portrait and service database information; Including a hybrid recommendation unit and a recommendation algorithm adjustment unit; The hybrid recommendation unit combines the content-based recommendation algorithm and the collaborative filtering recommendation algorithm to generate personalized service recommendations for users. First, the content-based recommendation algorithm and the collaborative filtering recommendation algorithm are run simultaneously. The content-based recommendation algorithm searches for similar service content in the service database based on the information in the user portrait and generates a content-based recommendation result R. cb , and then the collaborative filtering recommendation algorithm, based on the similarity between the user profile and other user profiles, finds out the services that similar users like, and obtains the collaborative filtering recommendation result R cf ; Then according to the user's activity A u Dynamically adjust their weights w(A u ), and finally R cb and R cf Add the calculated weights to get the final recommendation result R, which is calculated as follows: R=w(A u )*R cb +(1-w(A u ))*R cf ; The recommendation algorithm adjustment unit is used to dynamically adjust the specific implementation details of the content-based recommendation and collaborative filtering recommendation algorithms according to different service categories and user preferences, and adopts a weight function optimization method based on the features in the user portrait and the characteristics of the service to optimize the content similarity calculation method in the content-based recommendation and the neighbor selection criteria in the collaborative filtering recommendation.

7. The personalized service management system based on user portrait analysis according to claim 1, characterized in that: The service providing module includes a service scheduling unit and a service execution unit; The service scheduling unit is used to receive the recommended service information transmitted by the service recommendation module and the user demand information obtained from the user portrait, and call the service to provide the required scheduling resources according to the service characteristics and user needs; The service execution unit obtains detailed instructions for service execution from the service scheduling unit, including required resources, service provision speed, and specific content of the service, and then integrates and configures the required resources according to the scheduling instructions.

8. The personalized service management system based on user portrait analysis according to claim 1, characterized in that: The service evaluation module includes an evaluation information collection unit and an evaluation analysis unit; The evaluation information collection unit first provides users with a feedback channel for evaluation information, sets an evaluation form on the service completion page, and users can rate the service satisfaction in the form and fill in the service completion degree; at the same time, it can also collect user comments and suggestions; then standardize the evaluation information input by the user, aggregate the evaluation information of different users into an evaluation information set V, and finally, when new user evaluation information is generated, update the evaluation information set in time to ensure the timeliness and completeness of the evaluation information; The evaluation analysis unit calculates the evaluation information using the evaluation function V=h according to the user's behavior after using the recommended service and the user information, where h represents the user's satisfaction score and the service completion degree, and then uses the weighted average evaluation formula to calculate the comprehensive evaluation score, which is as follows: S a =w1*S+w2*C Among them, S is the satisfaction score, C is the service completion, w1 and w2 are weight coefficients, and w1+w2=1; the values ​​of w1 and w2 can be dynamically adjusted according to different service types and user groups.

9. A personalized service management method based on user portrait analysis, applied to a personalized service management system based on user portrait analysis as claimed in any one of claims 1 to 8, characterized in that: The following steps are involved: S1, through the user identity information collection unit, user behavior data collection unit and user feedback information collection unit of the user information collection module, respectively collect the basic identification information of the user during registration, the behavior data in the system and the user's evaluation, comments and scoring information on the system service; S2, the data cleaning unit, data normalization unit and feature extraction unit of the data preprocessing module process the collected user information, including setting data rules, processing errors and missing data, unifying data scales, and using covariance matrix and eigenvalue eigenvector for dimensionality reduction; S3, the autoencoder construction and training unit of the user portrait construction module receives the preprocessed data, constructs and trains the autoencoder, learns the key features by minimizing the reconstruction error, the user portrait generation unit uses the trained autoencoder to convert the data into a user portrait vector containing multi-dimensional features, and the incremental learning update unit updates the autoencoder parameters and user portrait when new user information is added; S4, the service information storage unit of the service database module stores system service information, and the service information retrieval and screening unit retrieves and screens service information according to requirements; S5. The hybrid recommendation unit of the service recommendation module combines content-based recommendation and collaborative filtering recommendation algorithms, adjusts the weights according to user portraits and service database information, and obtains the final recommendation results based on user activity. The recommendation algorithm adjustment unit dynamically adjusts the specific details of the two recommendation algorithms. S6. The service scheduling unit of the service provision module schedules the resources required for the service according to the service recommendation information and user needs, and the service execution unit executes the service, integrates and configures the resources and delivers the service; S7. The evaluation information collection unit of the service evaluation module collects the user's evaluation, comments, suggestions and other information on the service completion page and organizes and updates the evaluation set. The evaluation analysis unit uses the evaluation function and weighted average formula to calculate the comprehensive evaluation score.

Images