Dynamic interest modeling method and system based on user behavior driving

Abstract

The invention discloses a dynamic interest modeling method and system based on user behavior driving, and belongs to the technical field of information recommendation. The method comprises the following steps: collecting behaviors of browsing, clicking, online chatting, exchanging contact information and the like of a user in a platform, classifying the user behaviors, and calculating corresponding behavior weights; generating a recommendation control strategy based on the behavior weight, and dynamically regulating and controlling sorting, exploration bit distribution, guaranteed exposure and recommendation frequency of recommended contents; when it is detected that the user generates a high-intention behavior, the corresponding behavior weight is updated, and dynamic optimization of a recommendation strategy is achieved in combination with weight capping and a long-term behavior-free attenuation mechanism; the updated behavior weight is used for the next round of recommendation to form closed-loop control. The method can dynamically adjust the recommendation strategy based on the real behavior of the user, improves the recommendation precision, user experience and system stability, and is suitable for recruitment information, local life service and other application scenes needing behavior-driven recommendation.

Description

Technical Field

[0001] This invention relates to the field of information recommendation technology, and in particular to a dynamic interest modeling method and system based on user behavior, which can be applied to social networking, recruitment, local information services and other scenarios that require accurate matching based on user interests and behaviors. Background Technology

[0002] Existing recommendation systems typically make recommendations based on content tags or user history, but they suffer from the following problems: 1. Inadequate behavioral response The system cannot dynamically adjust recommendation weights based on the user's actual intent (such as high-intent behavior), resulting in limited recommendation accuracy. 2. The recommended strategy is too simplistic. Most systems only focus on content quality or static tag matching, lacking behavior-driven dynamic control, and are unable to achieve real-time feedback and closed-loop optimization of user interests; 3. Lack of dynamic control mechanism User behavior typically has a static or intermittent impact on the recommendation list, making it impossible to systematically control recommendation frequency, exploration position, and minimum exposure. Therefore, existing technologies struggle to achieve dynamic adjustment and personalized recommendation optimization based on user behavior while ensuring recommendation accuracy. Summary of the Invention

[0003] Technical issues This invention aims to solve the following technical problems: 1. How to dynamically adjust recommendation weights based on actual user behavior to improve recommendation accuracy; 2. How to control the display strategy of the recommendation list (such as exploration position, minimum exposure, refresh frequency) to ensure system fairness and user experience; 3. How to achieve system modularization so that behavioral strategies and recommendation controls form a closed loop and can be continuously optimized.

[0004] Technical solution To address the above problems, the present invention provides the following technical solution: 1. User behavior collection module The system monitors various user behaviors on the platform, including but not limited to browsing, clicking, online chatting, and exchanging contact information. 2. Behavior Classification and Weight Calculation Module Classify user behavior (high intent, low intent, exploratory behavior, etc.) Behavioral weights are dynamically calculated based on behavior type and historical performance. 3. Recommended control module Based on behavioral weights, the generation of the recommendation list is controlled, including content sorting, exploration slot allocation, minimum exposure, and dynamic adjustment of recommendation frequency. 4. Closed-loop optimization module User behavior is fed back to the weight calculation module to achieve dynamic adjustment. It can be combined with existing recommendation algorithms to achieve closed-loop optimization; 5. Modular Implementation of the System Each module can be implemented independently, supporting system upgrades and strategy optimization, enhancing implementability and scalability. Attached Figure Description

[0005] Figure 1 A User Behavior-Driven Dynamic Interest Modeling Method and a System Overall Flowchart S1 User enters the recommendation system: When a user opens the platform homepage or recommendation page, the behavior collection module is triggered to start recording user behavior; S2 User Behavior Collection and Classification Module: The system collects user behavior in real time and classifies it into high-intent or low-intent behaviors, assigning them initial weights. S3 Behavior Weight Calculation Module: Dynamically calculates behavior weights based on user behavior categories and historical performance, providing a base score for recommendation ranking; S4 Recommendation Control Strategy Module: Generates a recommendation list based on behavioral weights and controls the exploration position, minimum exposure, and refresh frequency to achieve dynamic adjustment of recommendations; S5 High-Intent Behavior Judgment Node: Determines the type of the user's current behavior, and only high-intent behaviors trigger weight updates; S6 Weight Update and Control Module: Updates the weights corresponding to high-intent behaviors, and executes weight capping and long-term no-behavior decay strategies. S7 Next Round Recommendation Node: Generates the next round recommendation list using the updated weights, achieving closed-loop optimization; Figure 2 User behavior collection and classification module 1. Behavioral input: including browsing, clicking, online chatting, exchanging contact information, etc.; 2. Classification module: Divides user behavior into high-intent behavior and low-intent behavior; 3. Initial weight assignment: Assign initial weights to each type of behavior based on the behavior type to prepare for subsequent calculations; Figure 3 Recommendation Control and Weight Update Module 1. Behavioral Weight Calculation Module (S3): Calculates the overall behavioral score; 2. Recommendation list generation module (S4): Sort by calculated score, control exploration position, minimum exposure and refresh frequency; 3. High-intent behavior judgment (S5): Determine the type of user behavior. Low-intent behavior does not update the weight, while high-intent behavior triggers a weight update. 4. Weight Update and Control (S6): Execute weighting, capping, and decay strategies; 5. Next round of recommendations (S7): Generate a new recommendation list using the updated weights to achieve closed-loop optimization. Detailed Implementation

[0006] The specific embodiments of the present invention include the following steps: S1 User enters the recommendation system When a user opens the homepage or recommendation page, the system begins to monitor user behavior. S2 User Behavior Collection and Classification 1. The system collects user behavior data in real time (browsing, clicking, online chatting, exchanging contact information, etc.); 2. Divide behaviors into high-intent behaviors and low-intent behaviors, and assign them initial weights, for example: Views: Weight 0.1 Click: Weight 0.3 Online chat: Weight 1.0 Exchanging contact information: Weight 1.5 S3 Behavioral Weight Calculation 1. Dynamically calculate the impact of each action on the recommendation strategy: The behavioral score is equal to the sum of all behavioral weights multiplied by the behavioral type coefficient; 2. High-intention behaviors have a larger weighting, while low-intention behaviors have a smaller impact; S4 Recommended Control Strategy 1. Adjust the sorting of the recommendation list based on behavioral scores. 2. Control the exploration position, minimum exposure, and refresh rate; 3. Example calculation: Content score equals behavior score multiplied by freshness coefficient multiplied by quality coefficient; S5 High Intent Behavior Judgment Determine whether the user's current behavior is a high-intent behavior (such as online chatting or exchanging contact information), and only high-intent behaviors will trigger weight updates; S6 Weight Update and Control 1. When a high-intent behavior occurs, update the corresponding behavior weight; 2. Weight capping and long-term inactivity decay: Weight cap limit If there is no activity for 7 days, the weight is multiplied by 0.9. If there is no activity for 14 days, the weight is multiplied by 0.8. S7 enters the next round of recommendations 1. Generate the next round of recommendation list using the updated behavioral weights; 2. Form a closed-loop optimization to achieve dynamic recommendation.

Claims

1. A method for dynamic interest modeling based on user behavior driving, characterized in that, Includes the following steps: S1 acquires user behavior data generated on the platform's recommendation page; S2 collects and classifies the behavioral data, and the behaviors include at least browsing, clicking, online chatting, and exchanging contact information; S3 Calculates the corresponding behavior weight based on the behavior type and historical behavior performance; S4 generates a recommendation control strategy based on the aforementioned behavioral weights, sorts the recommended content, and dynamically adjusts the exploration position, minimum exposure, and recommendation frequency. S5 determines whether a user's behavior is a high-intent behavior; S6 When a high-intent behavior is detected, the corresponding behavior weight is updated, and weight capping and decay control are executed; S7 generates the next round of recommended content based on the updated behavioral weights, thus forming a closed-loop recommendation system.

2. The method according to claim 1, characterized in that, User behavior is categorized into high-intent behavior and low-intent behavior, with online chatting and exchanging contact information defined as high-intent behavior.

3. The method according to claim 1, characterized in that, Behavioral weights are dynamically calculated based on the basic weights of behavioral types and the frequency of historical behaviors.

4. The method according to claim 1, characterized in that, Recommendation control strategies include: ranking recommended content by behavioral weight; setting a predetermined proportion of exploration slots; and setting a minimum exposure guarantee mechanism for some content.

5. The method according to claim 1, characterized in that, Weight updates include: increasing the weight of high-intent behaviors; capping the weight when the upper limit is reached; and reducing the weight when there are no high-intent behaviors for a long period of time.

6. The method according to claim 1, characterized in that, Behavioral weights are used to control the maximum number of recommendations a user can receive per unit of time or the proportion of resources allocated to them, thereby controlling the frequency of recommendations.

7. A user behavior-driven dynamic interest modeling system, characterized in that, include: The user behavior collection module is used to collect user behavior data on the platform; The behavior classification and weight calculation module is used to classify user behaviors and calculate their weights. The recommendation control module is used to generate recommendation control strategies based on behavior weights; The weight update module is used to update weights when high-intent behavior is detected; The closed-loop optimization module is used to generate the next round of recommended content based on the updated weights.

8. The system according to claim 7, characterized in that, Each module can be deployed independently and supports dynamic adjustment of strategy parameters.

Images