Multi-dimensional user portrait efficient calculation method

Abstract

The invention relates to a multi-dimensional user portrait efficient calculation method, and relates to the technical field of user portrait calculation, and the method comprises the steps: delineating a target crowd package through a user interface, and splitting a delineated crowd into a plurality of independent sub-crowd packages through an advanced delineation analysis algorithm, thereby achieving the deep analysis of user attributes and behaviors; then, the bitmap algorithm is used for carrying out compression and duplicate removal on the data, and the calculation efficiency is remarkably improved; and finally, the merged user portrait result and other user tags are subjected to intersection operation to generate personalized recommendation content. The method is suitable for private domain user label management and grouping management, the user portrait construction efficiency in a large-scale user data environment is improved, user behavior changes can be adapted in real time, and therefore accurate support is provided for marketing and personalized recommendation.

Description

Technical Field

[0001] This application relates to the field of user profile calculation, and particularly to a method for efficiently tagging and clustering users in the management of private domain traffic. Background Art

[0002] In today's digital age, user profiles have become an important tool for enterprises to understand user needs and optimize products and services. A user profile is a set of user characteristics generated through analysis and modeling based on user behavior data, attribute information, and interaction records. However, with the increase in the number of users, especially in industries such as e-commerce and social media, traditional methods for constructing user profiles often face the following challenges: Huge data volume: User data is growing exponentially, and how to effectively process and analyze this vast amount of data is an urgent problem to be solved; Low computing efficiency: Traditional methods often rely on complex calculations when processing user profiles, resulting in long response times and affecting the user experience; Dynamic user behavior: Users' interests and behaviors change over time, and how to update user profiles in real time to reflect these changes is another challenge; Diversity of personalized needs: Users' personalized needs are becoming increasingly diverse, and how to provide accurate services and recommendations according to different user groups has become a major problem for enterprises.

[0003] To address these challenges, there is an urgent need for an efficient, flexible, and dynamically adjustable method for constructing user profiles to enhance the user experience, improve customer satisfaction, and achieve higher conversion rates. Summary of the Invention

[0004] The purpose of this application is to provide an efficient multi-dimensional user profile calculation method that can support flexible construction of user profiles under large-scale user data, especially having significant advantages in private domain user tag management and clustering management.

[0005] The efficient multi-dimensional user profile calculation method provided by this application adopts the following technical solutions: The core steps of this method are as follows: S1. Define the population package to be selected this time through the user interface. The user interface supports multi-level screening and visual display to enhance the user interaction experience; S2. Use the selected parsing algorithm to split the population package into several independent sub-population packages, where the algorithm is optimized based on a machine learning model of user behavior patterns; S3. Divide the sub-population packages into attribute-based population packages, behavior-based population packages, and dynamic asset-based population packages, and extract features from them to improve the accuracy of the profile; S4. Efficiently compress and deduplicate multiple population packages using a bitmap algorithm, and perform efficient NAND operations to obtain the merged population package result; S5. Perform an intersection operation on the merged population package with an attribute package, a behavior package, a model label package, etc. to obtain the final user profile result, which is used for real-time calculation of personalized recommendations.

[0006] Preferably, the method for defining a population package includes the following steps: User interface design: The user interface is designed to be intuitive and easy to operate, supporting multi-level filtering so that users can conveniently select the required population packages; the user interface supports graphical display of user characteristics, using visual methods such as heat maps and bar charts to enhance the user's interaction experience; Multi-dimensional filtering conditions: Preferred filtering conditions include but are not limited to demographic characteristics (such as gender, age, geographical location, occupation), behavioral characteristics (such as purchase frequency, consumption amount, access time period), and interest tags (generated based on user historical behavior); users can customize filtering conditions according to their needs to meet the requirements of different business scenarios; Dynamic update: When the user selects filtering conditions, the system updates the optional user list in real time to ensure that users can see the latest data that meets the conditions; users can view the currently defined population package through the preview area and modify it according to actual needs.

[0007] Confirmation and feedback of the definition process: After the user is satisfied with the defined result, confirm the defined result through the "Confirm" button; the system provides the user with statistical information on the defined result, including the number of defined people, main characteristics, etc., for the user to conduct further analysis; Support for complex condition combinations: The preferred definition method supports users to set complex combination conditions, such as "aged between 25 and 35 years old and the consumption amount exceeds 1000 yuan", to achieve more refined population package division.

[0008] User experience optimization: The interface design focuses on the user experience, using icons to represent different filtering conditions to ensure simple and clear operations; providing a user feedback channel to allow users to make suggestions for the definition process, thereby continuously optimizing the system performance.

[0009] Preferably, the selected parsing algorithm uses a deep learning model based on user behavior, and improves the parsing ability of complex user profiles through training.

[0010] Preferably, the bitmap algorithm adopts a compression storage technology, including a compression method based on column storage, to reduce the consumption of computing resources and speed up the response speed. The compression method is as follows: By merging consecutive identical status values, the occupation of storage space is reduced. For example, "11110000" is represented as "4 ones and 4 zeros"; Adopt this encoding method to further compress the data with consecutive identical values, so as to reduce the storage requirement and improve the access speed; Allocate short encodings to the frequently occurring status values to enhance the storage compactness and processing efficiency.

[0011] Preferably, the user interface supports multiple interaction methods, including but not limited to mouse clicks, dragging, touch operations and speech recognition.

[0012] Preferably, the attribute class, behavior class and dynamic asset class of the population package can all be updated according to real-time data and have the ability of self-learning to adapt to market changes.

[0013] Preferably, the NAND operation of the merged population package adopts parallel computing technology, and the computing efficiency is improved through multi-threaded processing to support the generation of real-time user portraits. The implementation steps of parallel computing are as follows: Task decomposition: Decompose the NAND operation task into multiple independent subtasks. For example, for the bitmap of each user, the bitmap can be bit-split and the NAND logic of each bit can be processed separately; Multi-threaded processing: Utilize multi-threaded technology to allocate the decomposed subtasks to multiple threads for parallel execution. Each thread independently processes a part of the bitmap data to complete the NAND operation; Result merging: After each thread completes its subtask, the results are summarized to form the final merged result. This process can be achieved through thread-safe data structures to ensure data consistency.

[0014] Preferably, the final user portrait result includes personalized recommendation content, potential interest tags and behavior prediction of the user, and the accuracy of the portrait can be continuously optimized through a feedback mechanism. Specifically, it includes the following steps: Feedback type: The system collects explicit and implicit feedback information to evaluate the accuracy of the user portrait. Explicit feedback includes ratings, comments or satisfaction surveys actively provided by users; Implicit feedback is based on user behavior data, such as click-through rate, purchase behavior and page stay time; Feedback collection process: Data monitoring: The system continuously monitors the interaction between users and the platform and automatically collects implicit feedback data; User interaction: Prompt users to provide explicit feedback at an appropriate time, asking about their satisfaction or suggestions for the recommended content; Feedback storage: Store the collected feedback data in the database for subsequent analysis; Feedback Data Analysis: Data Mining: Use data mining techniques to analyze feedback data and identify changes and trends in user preferences; Model Adjustment: Optimize the feature weights and parameter settings of the user profile generation model according to the feedback data to improve the adaptability of the model; Real-time Update Mechanism: Dynamic Adjustment: The system updates the user profile in real time according to user feedback, enabling it to continuously reflect the latest behaviors and preferences of users; Automated Learning: Adopt incremental learning algorithms to quickly update the user profile without the need for a full recalculation, reducing resource consumption; Advantages of the Feedback Mechanism: Improve Accuracy: Ensure that the user profile promptly reflects the true needs of users through continuous feedback collection and analysis; Enhance User Stickiness: Improve user satisfaction and loyalty and enhance interaction with the platform; Adapt to Market Changes: Respond promptly to changes in the market and user behaviors to maintain a competitive edge.

[0015] Preferably, the method can dynamically adjust the algorithm parameters of the selected circle according to the historical behavior data and real-time interaction information of users to improve the adaptability and real-time performance of the user profile.

[0016] Preferably, the construction process of the user profile adopts an incremental learning method, which can quickly update the user profile when new data arrives without the need for a full recalculation. By gradually learning the newly arrived data, it avoids a full recalculation of the user profile, can effectively reduce computational resource consumption, and improve the system response speed; The implementation steps of incremental learning are as follows: Data Acquisition: The system collects user behavior data in real time, including user click records, purchase behaviors, and user opinions obtained through the feedback mechanism; Data Preprocessing: Clean and preprocess the newly collected data to ensure data quality, remove noise and irrelevant information for subsequent learning.

[0017] Model Update: Incremental Training: After receiving new data, the system uses incremental learning algorithms to update the existing user profile model. Incremental training adjusts the model weights and parameters to enable it to better adapt to the new data; Adaptive Adjustment: Dynamically adjust the learning rate and feature importance of the model according to the characteristics of the new data and changes in user behaviors to improve the adaptability and accuracy of the model.

[0018] Preferably, the visualization result of the user profile supports multi-dimensional display, including charts (such as pie charts, bar charts), heat maps (showing user activity), and distribution maps (displaying user feature distributions). Users can, through interactive operations, deeply understand each dimension of the user profile to improve the effectiveness of decision-making.

[0019] In summary, the present application includes at least one of the following beneficial technical effects: 1. The method of this application includes delineating a target population package through a user interface, splitting it into independent sub-population packages using a selection parsing algorithm, applying a bitmap algorithm for efficient compression and deduplication, and finally generating a personalized user profile. It is applicable to private domain user label management and group management, and can adapt to changes in user behavior in real time, thereby providing precise support for marketing and personalized recommendations, enabling the platform to deeply understand the preferences and needs of each user, and thus providing more personalized content and product recommendations; 2. The method of this application can achieve precise positioning and personalized services for users through effective user label management and group management. This method supports real-time updating of user labels to ensure the effectiveness and accuracy of the labels. User labels can be dynamically adjusted based on behavior changes to ensure that the system can instantly push relevant content, greatly enhancing user engagement and satisfaction; 3. The method of this application can enhance user stickiness. Personalized recommendations not only improve the user experience but also encourage users to interact with the platform more frequently, increasing user stickiness. This interaction helps to improve user retention rate, thereby promoting the long-term development and revenue growth of the platform.

[0020] 4. Through the incremental learning mechanism of the method of this application, the system can quickly update the user profile when user behavior changes without retraining the model. This efficient update process ensures the timeliness of the user profile, enabling the recommendation system to respond to user needs within the shortest time.

[0021] 5. By optimizing the use of computing resources, the system of the method of this application can operate stably under high concurrency conditions, avoiding resource waste. This efficiency not only reduces operating costs but also improves the user experience, ensuring that users can obtain fast and accurate feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is the flowchart of the method steps of this application; Figure 2 is the diagram of the user label management page of this application; Figure 3 is the diagram of the group management page of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The following will further elaborate on this application in conjunction with the attached Figure 1 - attached Figure 3 , to provide a more detailed description of this application.

[0024] A multi-dimensional user profile efficient calculation method, the core steps of which include: 1) Population package delineation: Through the user interface (UI), users can intuitively define the required population package. The UI supports multi-level filtering, allowing users to select the target user group according to multiple dimensions (such as time period, product type, user behavior, etc.). Users can quickly define the required population through simple operations such as dragging and clicking, ensuring that the selected user group meets the marketing goals; 2) Sub-population package parsing The selected population package is split into multiple independent sub-population packages through the circumscription parsing algorithm. This process includes the following steps: Data extraction: Extract data related to the selected population from the user database, including users' basic information, historical behaviors, purchase records, etc.; Classification model: Use a machine learning-based classification model (such as decision tree, random forest or deep learning algorithm) to analyze users' behavior patterns and automatically identify and classify users with different attributes and behaviors. The model can be updated in real time to reflect the latest data changes; Generate sub-population packages: Classify users according to attributes (such as age, gender, region) and behaviors (such as purchase frequency, browsing history) to form multiple sub-population packages. These sub-population packages provide the basis for subsequent analysis and personalized recommendations; Classification algorithm formula: Assume there are feature vector X and label Y. The goal of the classification model is to minimize the loss function: where h θ (x i ) represents the predicted value of the model; 3) Population package compression and deduplication To improve the computational efficiency, this application uses the bitmap algorithm for population package compression and deduplication. The specific steps are as follows: Bitmap representation: Create a bitmap for each user. Each bit of the bitmap represents whether the user meets a specific condition. For example, the first bit represents whether the user is an active user, and the second bit represents whether the user has purchased a specific product; Compression algorithm: Use a compression algorithm (Huffman coding or run-length encoding) to further compress the bitmap to reduce storage and computational costs, which enables the system to quickly access and calculate when dealing with large-scale data.

[0025] Bitmap compression formula: Deduplication processing: Ensure that duplicate users are removed when merging population packages through operations such as NAND to obtain an accurate population package; NAND operation formula: AND(A,B)=A∧B 4) Image result generation The generation of user portrait results is based on the intersection operation of merging population packs with other tags (such as interest tags, behavior tags, etc.). The specific steps include: Tag mapping: Map the attributes and behavior tags of users to the merged population pack to ensure that the portrait information of each user is up-to-date; Personalized recommendation generation: Based on the finally generated user portrait, the system can generate personalized recommendation content for each user; these recommendations can be based on the user's historical behavior, real-time interaction information, and the similar behavior of other users; Feedback mechanism: The feedback after user interaction (such as click-through rate, purchase rate, etc.) will be recorded and used for the retraining of the model, so as to continuously optimize the accuracy of the user portrait and the recommendation effect.

[0026] The implementation of this application is particularly suitable for private domain traffic management. Through effective user tag management and segmentation management, it can achieve accurate positioning and personalized services for users. The user tag management page is as attached Figure 2 shown, and the segmentation management page is as attached Figure 3 shown. This method supports real-time updating of user tags to ensure the effectiveness and accuracy of the tags. User tags can be dynamically adjusted based on behavior changes to ensure that in a changing market environment, enterprises can respond to user needs in a timely manner.

[0027] Example 1: Scenario of constructing user portraits on an e-commerce platform Background: On a large e-commerce platform, merchants hope to achieve precise marketing through user portraits to improve conversion rates and customer satisfaction. The goal is to recommend personalized products for each user by analyzing the purchase behavior and attributes of users.

[0028] Step description: (1) Defining the population pack Merchants select the target user group through the user interface (UI). They can select "users who have purchased electronic products in the past 30 days". This interface supports multi-level screening and allows merchants to select according to multiple dimensions (such as purchase time, product category, etc.); (2) Analyzing sub-population packs The selected population pack is split into multiple sub-population packs through the circle selection analysis algorithm. Attribute-based population packs: Classify according to the basic information of users (such as age, gender, region); Behavior-based population packs: Classify according to the purchase history, browsing records, products added to the shopping cart, etc. of users; Dynamic asset-based population packs: Consider the products in the user's current shopping cart and recent activities (such as recently viewed products); The analysis algorithm uses a machine learning model to improve the ability to analyze complex user behaviors through training on historical data; (3) Compressing and de-duplicating the population pack Merchants use a bitmap algorithm to compress and deduplicate different sub-population packages. The representation of each user in the bitmap is as follows: Bitmap example: User A: 00101 User B: 01010 User C: 00111 As shown above, each bit in the bitmap represents the presence of a user under specific conditions (1 means present, 0 means absent). Through the bitmap, merchants can efficiently store and compare user data; (4) NAND operation Merchants perform a NAND operation on multiple bitmaps to obtain a combined population package, representing users who meet multiple conditions simultaneously. For example: Suppose there are two-condition bitmaps: Bit Figure 1 (Users who purchased electronic products): 00101 Bit Figure 2 (Users aged between 18 - 25): 01010 Using the NAND operation, the combined result (00100) can be obtained, indicating that only User C meets both conditions; (5) Portrait result generation Merchants perform an intersection operation on the combined population package with other tags (such as interest tags, behavior tags, etc.) to generate the final user portrait results, which are used to generate personalized recommendation lists; for example, for User C, recommend the latest electronic products, related accessories, and promotional activities.

[0029] Example 2: Content push scenario of a social media platform Background: On a social media platform, the operation team hopes to perform precise content push based on user behavior to improve user interaction and platform activity.

[0030] Steps description: (1) Population package delineation The operation team delineates the user group through the UI. For example, select "active users" and "new users". Active users refer to those who have frequently liked and commented in the past week, while new users refer to those whose registration time does not exceed one month; (2) Sub-population package parsing The delineated user group is split into multiple sub-population packages through a parsing algorithm. Attribute-based population packages: including users' interest tags (such as technology, fashion, food, etc.); Behavior-based population packages: classified according to users' like, comment, and share behaviors; (3) Population package compression and deduplication The operation team uses the bitmap algorithm to compress different user groups. The performance of each user in the bitmap is as follows: Bitmap example: User X: 11000 User Y: 10101 User Z: 11111 These bitmaps represent the status of users under multiple conditions; (4) NAND operation The operation team performs NAND operations on different bitmaps to find users who meet multiple conditions simultaneously. For example, assume there are two-condition bitmaps: Bit Figure 1 (Active users): 11000 Bit Figure 2 (Users interested in technology content): 10101 Using the NAND operation, the combined result (10000) can be obtained, indicating that only User X meets both conditions; (5) Portrait result generation Finally, the operation team performs an intersection operation on the combined population package and content tags to generate a recommended content list. For example, for User X, the latest technology articles, videos, and related activities are recommended.

[0031] Example 3: Course recommendation scenario of an online education platform Background: On an online education platform, the operation team hopes to recommend courses based on users' learning behaviors to improve learning efficiency and user retention rate.

[0032] Step description (1) Population package definition The operation team defines the user group through the UI, selecting "active learners" and "newly registered users". Active learners refer to users who have completed at least three courses in the past month, and newly registered users refer to users whose registration time does not exceed two weeks; (2) Sub-population package analysis The defined user group is split into multiple sub-population packages through an analysis algorithm. Attribute-based population packages include users' learning fields (such as programming, design, marketing); behavior-based population packages are classified according to users' course completion, evaluations, and discussion participation; (3) Population package compression and deduplication The operation team uses the bitmap algorithm to compress different user groups. The performance of each user in the bitmap is as follows: Bitmap example: User D: 11000 User E: 10101 User F: 11111; (4) NAND operation The operation team performs a NAND operation on different bitmaps to find users who meet multiple conditions simultaneously. Suppose there are two-condition bitmaps: bit Figure 1 (Active Learners): 11000 bit Figure 2 (Users interested in programming courses): 10101 Using the NAND operation, the combined result (10000) can be obtained, indicating that only user D meets both conditions; (5) Portrait result generation Finally, the operation team performs an intersection operation on the combined population package and the course tags to generate a recommended course list. For user D, the latest programming courses, relevant learning materials, and practical projects are recommended.

[0033] The embodiments of this specific implementation manner are all preferred embodiments of this application, and do not limit the protection scope of this application accordingly. The same components are denoted by the same reference numerals. Therefore, all equivalent changes made according to the structure, shape, and principle of this application shall be covered within the protection scope of this application.

Claims

1. An efficient calculation method for multi-dimensional user portraits, characterized in that: The following steps are involved: S1. The group of people to be selected is identified through the user interface, and the user interface supports multi-level screening and visual display to enhance the user interaction experience; S2. Splitting the crowd package into a number of independent sub-crowd packages using a circle selection parsing algorithm, wherein the algorithm is optimized based on a machine learning model of user behavior patterns; S3, dividing the sub-crowd packages into attribute-based crowd packages, behavior-based crowd packages, and dynamic asset-based crowd packages, and performing feature extraction on them to improve the accuracy of the portrait; S4. Use a bitmap algorithm to efficiently compress and remove duplicates from multiple crowd packages, and perform efficient AND-NOT operations to obtain a merged crowd package result; S5. Perform intersection operation on the merged population package, the attribute package, the behavior package, the model label package, etc. to obtain a final user portrait result, which is used for real-time calculation of personalized recommendations.

2. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The circle selection and parsing algorithm adopts a deep learning model based on user behavior and improves the parsing ability of complex user portraits through training.

3. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The bitmap algorithm adopts compression storage technology, including a compression method based on column storage, to reduce the consumption of computing resources and speed up the response speed.

4. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The user interface supports multiple interaction modes, including but not limited to mouse clicking, dragging, touch operation and voice recognition.

5. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The attribute class, behavior class and dynamic asset class of the crowd package can be updated according to real-time data and have self-learning capabilities to adapt to market changes.

6. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The AND-NOT operation of the merged crowd packets adopts parallel computing technology and improves computing efficiency through multi-threaded processing to support the generation of real-time user portraits.

7. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The final user portrait result includes the user's personalized recommended content, potential interest tags and behavior predictions, and the accuracy of the portrait can be continuously optimized through a feedback mechanism.

8. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The method can dynamically adjust the selected algorithm parameters according to the user's historical behavior data and real-time interaction information to improve the adaptability and real-time performance of user portraits.

9. The multi-dimensional user portrait efficient calculation method according to claim 1 is characterized in that: The user profile construction process adopts an incremental learning method, which can quickly update the user profile when new data arrives without the need for comprehensive recalculation.

10. The multi-dimensional user portrait efficient calculation method according to claim 1, characterized in that: The visualization results of the user portrait support multi-dimensional display, including charts, heat maps and distribution maps, so that users can intuitively understand the portrait information.

Images