A dynamic matching optimization method and system integrating user profiles and education service profiles

Abstract

This invention relates to the field of educational service recommendation technology, specifically disclosing a dynamic matching optimization method and system that integrates user profiles and educational service profiles. The method includes: dynamically updating two types of profiles based on user interaction behavior data and educational service group effect data to reflect the latest user learning stage and service effect type; dynamically selecting appropriate matching rules and calculating the matching degree to generate recommendations based on the current state of both profiles; and synchronously driving the further updating of both types of profiles and optimizing the parameters of the used matching rules using user feedback behavior data. The system includes a profile collaborative update module, a dynamic matching decision module, a recommendation generation module, a feedback collection module, and a closed-loop learning optimization module. This invention overcomes the shortcomings of traditional static recommendation methods by enabling dynamic evolution of profiles, adaptive adjustment of matching strategies according to state, and the introduction of a feedback-driven optimization mechanism, achieving continuous accuracy and personalization in educational service recommendations.

Description

Technical Field

[0001] This invention relates to the field of educational service recommendation technology, and more specifically, to a dynamic matching optimization method and system that integrates user profiles and educational service profiles. Background Technology

[0002] In recent years, with the increasing application of information technology in education, educational service resources and diverse user learning needs have coexisted. Achieving efficient and accurate matching between these two has become crucial for improving the quality of educational services and user experience. Personalized recommendation systems, by constructing user profiles and educational service profiles and calculating their matching degree, realize resource filtering and are an important technical means to solve this problem.

[0003] In existing technologies, common personalized recommendation methods typically rely on static or outdated user profiles for matching. User profiles are mostly built based on initial registration information or limited historical behavioral data, making it difficult to reflect dynamic changes in a user's learning process, such as shifts in interest, growth in ability, or encountering bottlenecks. Educational service profiles often only contain static metadata such as author, title, and category, lacking a quantitative description of their actual teaching effectiveness, suitable target audience, and true value within the complete learning path. In the matching process, most systems use fixed algorithms or weights to calculate similarity, failing to adaptively adjust based on the user's current learning stage (e.g., exploration, improvement, bottleneck) and the specific teaching type of the educational service (e.g., basic consolidation, skill advancement). This results in recommendations that may be effective for a period of time but cannot provide continuous and accurate guidance as the user grows, nor can they fully utilize group learning feedback to optimize the self-description of service resources.

[0004] In summary, the existing technologies have the following problems: the update mechanism for user profiles and education service profiles is rigid and fails to form a dynamic evolution linked to the user's learning process and the actual effect of the service; the matching strategy lacks context awareness and cannot be flexibly adjusted according to the real-time status of both supply and demand sides; the system as a whole lacks a learning mechanism that uses recommendation feedback results to drive the joint iterative optimization of profiles and matching models. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a dynamic matching optimization method and system that integrates user profiles and education service profiles to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a dynamic matching optimization method integrating user profiles and educational service profiles, wherein the method constructs a feedback-based collaborative evolution closed loop, enabling the matching process to adapt to the dynamic changes of users and services; the method includes the following steps: S1. Using user interaction data and educational service group effect data as inputs, respectively, perform independent dynamic updates of user profile and educational service profile, so that the user profile can represent the user's current learning stage and the educational service profile can represent its current teaching effect type. S2. Based on the user learning stage parsed from the user profile and the service effect type extracted from the education service profile, the appropriate matching rule is dynamically selected by querying the preset state-rule mapping relationship, and the matching degree between the user profile and the education service profile is calculated according to the rule, thereby realizing the context awareness of the matching strategy. S3. Sort and filter the candidate education services based on the calculated matching degree to generate the final education service recommendation; S4. Collect and record user feedback behavior data on the recommendations, which serves as the basis for evaluating the actual effectiveness of the recommendations and driving system optimization; S5. The feedback behavior data is simultaneously used as new interaction behavior data and group effect data, and input into step S1 to trigger a new round of profile update. At the same time, based on the difference between the predicted value of matching degree in this recommendation and the actual effect value calculated by feedback, the internal parameters of the matching rule selected in step S2 are calibrated, thereby completing the complete learning cycle from decision-making, execution to feedback and optimization.

[0007] Furthermore, step S1, "dynamically updating the user profile based on user interaction behavior data," is a computable process from behavior to state, specifically including: S11. Extract and calculate behavioral indicators that can quantify learning efficiency and knowledge mastery stability from the temporal behavior records generated by users' interactions with education services. S12. Compare the calculated index values ​​with the preset thresholds through historical data analysis, and classify users into different predefined learning stages based on the comparison results. S13. Based on the first update rule bound to the determined learning stage, the weights of different knowledge or skill feature dimensions in the user profile are adjusted in a differentiated manner, so that the focus of the profile matches the user's current learning needs and cognitive state.

[0008] Furthermore, step S1, "dynamically updating the education service profile based on group effect data of education services," is a process of objectively evaluating and labeling the post-learning effects of the group, specifically including: S13. From the learning paths and effectiveness data of the user group that completed the service, aggregate and calculate service effectiveness indicators that can reflect the teaching guidance and skills improvement effects. S14. Based on the comparison between the calculated effect indicators and the preset thresholds, the educational services are classified into different types of teaching effects. S15. Based on the second update rule bound to the determined effect type, revise the content of the education service profile, including strengthening or weakening the strength of descriptive tags, and adjusting the relationship between the service and other services in the learning path map, so that the profile more accurately reflects its true teaching value and positioning.

[0009] Furthermore, in step S2, the "dynamic selection of matching rules" is implemented through a predefined decision logic, which maps different combinations of "user learning stages" and "service effect types" to the optimal matching rules pre-configured for each combination. When performing matching calculations, the system activates and applies the corresponding matching rules based on the specific stage and type combination currently identified. These rules define the specific calculation criteria and weight allocation for measuring the relevance between user characteristics and service attributes in this specific scenario.

[0010] Furthermore, the "optimization of the matching rules used in step S2" in step S5 is a supervised parameter tuning process based on prediction error, which enables the system to continuously learn from practice; this process includes: S51. Record the matching rules used when generating recommendations and the predicted matching degree of their output, and calculate an actual conversion metric that characterizes the actual effectiveness of the recommendations based on feedback behavior data. S52. Calculate the numerical difference between the predicted matching degree and the actual conversion metric; S53. Based on the direction and magnitude of the difference, the rule weight parameters associated with the feature dimensions that have made significant contributions in this matching calculation are fine-tuned in a targeted manner. Through iterative adjustment, the predicted output of the matching rule gradually approaches the actual feedback.

[0011] A dynamic matching and optimization system integrating user profiles and education service profiles, wherein the system implements a closed-loop process of the above method through a modular architecture, and the various modules of the system are interconnected based on data flow and functional dependencies; the system includes: The profile collaborative update module is used to process user interaction behavior data and education service group effect data separately, so as to realize the independent and continuous updating of user profiles and education service profiles, and output their respective status representations. The dynamic matching decision module is connected to the profile collaborative update module. It is used to receive the status representation, dynamically select matching rules according to its content, and calculate the matching degree between the user profile and the education service profile. The recommendation generation module is connected to the dynamic matching decision module and is used to generate and output a list of recommended education services based on the matching degree. The feedback collection module is used to monitor and collect users' subsequent interaction behavior with the recommendation list and generate feedback behavior data. The closed-loop learning optimization module is connected to the feedback acquisition module, the profile collaborative update module, and the dynamic matching decision module, respectively. It is used to distribute the feedback behavior data to the profile collaborative update module to trigger its update. At the same time, based on the difference between the feedback data and the matching prediction value, it optimizes the parameters of the matching rules used in the dynamic matching decision module.

[0012] Furthermore, the portrait collaborative update module has parallel processing units internally to handle the two types of data separately: The user profile update unit is configured to execute steps S11, S12 and S13, which analyze individual user behavior sequences, calculate behavior indicators, determine the learning stage, and adjust the user profile feature weights according to the stage rules. The service profile update unit is configured to execute steps S13, S14 and S15, which analyze service group effect data, calculate effect indicators, determine effect type, and adjust service profile tags and associations according to type rules.

[0013] Furthermore, the internal logic of the dynamic matching decision module implements on-demand scheduling of matching strategies, including: The rule decision unit stores the state-rule mapping relationship and is used to query and output the corresponding matching rule identifier based on the input user learning stage and service effect type. The matching calculation unit is connected to the rule decision unit and is used to load and execute the matching rules specified by the rule identifier to complete the similarity calculation between the user profile and the education service profile.

[0014] Furthermore, the closed-loop learning optimization module is the core control component for realizing the system's self-evolution, including: The data distribution unit, connected to the feedback acquisition module, is used to identify and distribute the received raw feedback behavior data according to its data properties, and send it to the user profile update unit and the service profile update unit respectively, as new inputs to trigger them to perform update calculations. The rule optimization unit, which is connected to the dynamic matching decision module and the feedback acquisition module, is used to execute steps S51, S52 and S53, and to iteratively calibrate specific rule parameters in the rule base of the dynamic matching decision module by comparing the differences between historical matching predictions and actual feedback.

[0015] Furthermore, the rule optimization unit adopts an optimization strategy that focuses on key factors when performing parameter calibration; specifically, it is configured to: first analyze the user or service profile feature dimensions that contribute most significantly to the final matching degree in the previous execution of the target matching rule; Subsequently, the calculated prediction discrepancies are primarily correlated with the weight settings of these key feature dimensions, and targeted and controlled parameter adjustments are made accordingly to ensure the effectiveness of the optimization process and the overall stability of the system.

[0016] The technical effects and advantages of this invention are as follows: To address the shortcomings of static and outdated user profiles and education service profiles in existing technologies, this invention designs dynamic update mechanisms based on user interaction behavior sequences and education service group effect data, respectively. For user profiles, the system analyzes micro-behavioral indicators such as learning rate and knowledge mastery stability to automatically determine whether a user is in an improvement phase or a bottleneck phase, and dynamically adjusts the feature weights in the profile according to preset rules, ensuring the profile focuses on the knowledge domain most needed by the user. For education service profiles, the system analyzes data such as user group path continuity rate and skill gain effect to automatically identify services as different effect types such as leapfrog or consolidation, and updates their descriptive tags and relationships accordingly. This mechanism allows both types of profiles to move beyond static descriptions and become entities that continuously reflect the latest state and value, providing real-time and reliable evidence for accurate matching.

[0017] To address the problem that existing matching strategies are fixed and unable to adapt to diverse contexts, this invention proposes a dynamic matching rule selection method based on dual profile states. The system uses the user's learning stage and the teaching effectiveness type of the educational service as key contexts, and pre-defines optimal matching rule mappings for different state combinations. During recommendation, the system first parses the current state of both user profiles, and then calls the appropriate matching rule based on the mapping relationship to calculate the matching degree. This rule defines the optimal weight combination of various features in this specific context, thereby improving the adaptability and accuracy of recommendations under different user states and service scenarios.

[0018] To address the lack of continuous learning capabilities from feedback in existing systems, this invention constructs a closed-loop optimization framework. The system uses user feedback data on recommendation results as new input to drive updates to user and service profiles, ensuring that profile evolution is synchronized with real-world feedback. By comparing the predicted values ​​with the actual results based on feedback, the system fine-tunes the parameters of the matching rules used in each iteration. This process allows the matching model to learn from each recommendation practice, gradually reducing prediction bias. Through the synergistic interaction of profile updates, state-aware matching, and feedback optimization, the entire system forms a continuously self-improving loop, addressing the problem of recommendation effectiveness diminishing over time and with user growth. Attached Figure Description

[0019] Figure 1 This is the overall flowchart of the dynamic matching optimization method of the present invention.

[0020] Figure 2 This is a diagram of the dynamic matching optimization system architecture of the present invention.

[0021] Figure 3 This is a branch diagram for the dynamic matching rule selection and calculation of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1 As attached Figures 1 to 3 The dynamic matching optimization method and system that integrates user profiles and education service profiles, as shown, are implemented in the following ways: S100: Multi-source data acquisition and preprocessing The system collects raw data streams in parallel from two independent data sources during startup and operation.

[0024] The primary data source is the user interaction log system. This system uses an event-driven approach to capture real-time behavioral events generated during user interactions with educational service content within the platform.

[0025] The captured behavioral event types include: video content playback, pause, and skipping, along with corresponding precise timestamps; interactive exercise submissions, answer modification sequences, and final results; and explicit ratings and text comments for courses or exercises. This event data, along with user identifiers, content identifiers, and timestamps, is encapsulated into structured interactive behavior records.

[0026] The second data source is a learning behavior analysis database. The system extracts aggregated user group learning performance data from the database in a periodic scheduling manner.

[0027] The extracted data items include: a list of users who completed specific educational services within the statistical period; records of these users' scores on standardized skills assessments before and after service completion; and a sequence of subsequent course learning behaviors generated by these users within an observation window after service completion. This data was compiled into a group aftereffect record indexed by educational service.

[0028] Preprocessing operations include standardizing user identifiers and education service identifiers, standardizing timestamps, and filtering out obviously invalid or test-generated data records.

[0029] After preprocessing, two clear data streams are formed: a user-level interaction behavior data stream and an education service-level group aftereffect data stream. These data will serve as the sole input source for subsequent updates to the two types of dynamic profiles.

[0030] S200: Dynamic Update of User Profiles The method includes a user profile dynamic update step. This step is triggered by receiving a new user interaction behavior data stream, and its purpose is to analyze user behavior, quantify its current learning state, and adjust the internal representation of the user profile accordingly.

[0031] S201: Calculation of User Behavior Metrics The system first aggregates recent user behavior for specific knowledge units and calculates two quantitative indicators.

[0032] Furthermore, as a preferred embodiment, the calculation of the user behavior indicators includes: calculating the user's specific knowledge unit learning rate indicator and knowledge mastery fluctuation indicator.

[0033] The learning rate metric is obtained by calculating the rate of change of the average single dwell time of a user within the same knowledge unit during a continuous learning time window.

[0034] Specifically, it involves extracting the user's most recent consecutive data points on the target knowledge unit. Duration sequence of each learning session ,in The duration of the first learning session within the window. This refers to the duration of the last learning session. The learning rate metric LR can be calculated using the formula... The calculation of this indicator is performed, and the result is used to determine the learning trend.

[0035] The knowledge mastery fluctuation index is calculated by measuring a user's most recent knowledge fluctuation within the same knowledge unit. The variance of the accuracy rate of each interactive practice session was obtained.

[0036] Specifically, extract the user's most recent activity in this unit. Accuracy sequence of the exercises ,in For the first The accuracy rate of each practice session, knowledge mastery of the KMV fluctuation indicator through the formula The calculation yielded, where For sequence The arithmetic mean.

[0037] S202: Classification of User Learning Status Based on the calculated learning rate index and knowledge mastery fluctuation index, the system applies preset rules to classify the user's current state into a discrete set of states.

[0038] Furthermore, as a preferred embodiment, the preset index and learning stage mapping rule is as follows: if the learning rate index exceeds the first threshold and the knowledge mastery fluctuation index is lower than the second threshold, then the user's learning state is mapped to the improvement period. If the learning rate index is lower than the third threshold and the knowledge mastery fluctuation index is higher than the fourth threshold, then the user's learning status is mapped to a bottleneck period. The first threshold, the second threshold, the third threshold, and the fourth threshold are system parameters preset based on historical data statistical analysis or domain knowledge.

[0039] These threshold parameters are initialized during system deployment, and their specific values ​​can be set through distribution analysis of historical sample data or based on teaching theories. The system performs classification logic by comparing the calculated learning rate index (LR) and knowledge mastery fluctuation index (KMV) with these preset thresholds.

[0040] S203: User Profile Feature Weight Adjustment Based on the status label generated in S202, the system makes targeted adjustments to the user feature vector stored in the profile database.

[0041] Furthermore, as a preferred implementation, the adjustment of the user profile according to the first update rule corresponding to the learning stage specifically means: when the user's learning stage is mapped to the improvement period, the first update rule indicates that the weight of deep skill features that are strongly correlated with the knowledge points already mastered in the user profile should be increased. When a user's learning phase is mapped to a bottleneck period, the first update rule instructs to increase the weight of adjacent basic knowledge point features in the user profile that are logically related to the current weakness.

[0042] In practice, the user profile feature vector consists of multiple dimensions, each corresponding to a knowledge point or skill tag, and is accompanied by a weight value.

[0043] The system maintains an update rule mapping table. When the status is in the improvement phase, the rule execution engine identifies the set of knowledge units that the user has recently interacted with frequently and with high accuracy. It then queries the platform's knowledge graph to find the set of deeper skill nodes that these units directly lead to, and increments the weights of the dimensions corresponding to these deeper skill nodes in the user's feature vector by a fixed first preset value for each. This value could be, for example, 0.1.

[0044] When the system is in a bottleneck phase, the rule execution engine identifies the knowledge units with the highest recent error rate or those that the user has practiced repeatedly. It then queries the knowledge graph for the set of foundational and prerequisite knowledge nodes that these units depend on, and adds a second preset value to the weights of each dimension in the user's feature vector corresponding to these foundational knowledge nodes. This value could be, for example, 0.15.

[0045] The knowledge graph is a pre-constructed static database that reflects the logical relationships between knowledge points. The relationships between the aforementioned "deep skill features strongly associated with mastered knowledge points" and "adjacent basic knowledge point features logically associated with current weaknesses" are predefined in this knowledge graph.

[0046] S300: Dynamic Updates to the Education Service Profile The method also includes a dynamic update step for the education service profile. This step is triggered by the receipt of new group follow-up data streams, and its purpose is to evaluate the teaching effectiveness of the service and update the service's descriptive tags and relationships.

[0047] S301: Calculation of Education Service Effectiveness Indicators The system analyzes the post-treatment records of the target educational service group and calculates two quantitative performance indicators.

[0048] Furthermore, as a preferred embodiment, the calculation of educational service effectiveness indicators using post-effect data includes: calculating the user path continuity rate of educational services and the group skill gain indicator.

[0049] The path continuation rate metric is obtained by statistically analyzing the proportion of users who have completed the educational service and subsequently activated the designated associated services within a specified time.

[0050] Specifically, let's set up a service completion The user set is During the specified time interval Pre-set related courses have begun. The learning user subset is The path continuity rate index Where |U| represents the set The number of users in China Representing a subset The number of users in China.

[0051] The group skill gain index is obtained by calculating the difference in the average score of the user group on the preset standard skill assessment before and after completing the educational service.

[0052] Specifically, from the set Extracting a subset of samples For each of these users to obtain its learning services Before and after test scores and The group skill gain index is: ,in, This represents the summation operation. This indicates the number of users in the sample subset.

[0053] S302: Classification of Educational Service Effectiveness Types Based on the calculated path continuity rate and group skill gain indicators, the system applies preset rules to classify the teaching effectiveness of the service.

[0054] Furthermore, as a preferred embodiment, the preset index and service type mapping rule is as follows: if the path continuity rate index exceeds the fifth threshold and the group skill gain index exceeds the sixth threshold, then the service effect type is mapped to the leap type. If the group skill gain index stabilizes between the seventh and eighth thresholds, the service effect type will be mapped to consolidation.

[0055] The fifth, sixth, seventh, and eighth thresholds are preset system parameters used to evaluate the effectiveness of teaching services. These thresholds are set in a similar way to the thresholds in S202, determined through historical data analysis or domain knowledge.

[0056] The system performs classification logic by comparing the path continuity metric (PCR) and the group skill gain metric (GSG) with these thresholds.

[0057] S303: Adjustment of Education Service Profile Tags and Relationships Based on the effect type label generated by S302, the system updates the feature vector of the service and its associated graph stored in the service profile database.

[0058] Furthermore, as a preferred implementation, the adjustment of the education service profile according to the second update rule corresponding to the effect type specifically means: when the service effect type is mapped to a leap type, the second update rule indicates that the strength of the "acceleration" tag is strengthened in the description tag of the education service profile, and a strong recommendation link between it and the subsequently designated associated service is added in its relationship links. When the service effect type is mapped to consolidation, the second update rule instructs that the strength of the "basic" tag be strengthened in the description tag of the education service profile, and its direct link to high-difficulty services be weakened.

[0059] In practice, the education service profile includes a descriptive label vector and a relationship graph.

[0060] When the type is transition, the update rule indicates: increase the intensity value of the "accelerate" label in the label vector. Updated to ,in For a preset increment, for example, The possible value is 0.1.

[0061] In the relationship graph, find the preset associated courses that point to this service. The edge, and its recommendation weight Updated to ,in For example, the enhancement coefficient. The possible value is 0.5.

[0062] When the type is reinforced, the update rule indicates: strengthen the strength value of the "base" label in the label vector.

[0063] In the relationship graph, identify and reduce the links from this service to high-difficulty services. The edge weights, for example, updating the weights to ,in For example, the attenuation coefficient. The possible value is 0.3.

[0064] S400: Dynamic Matching and Recommendation When the system receives a request to generate educational service recommendations for a user, it executes a dynamic matching and recommendation step. This step utilizes the latest dynamic profile and status information generated in steps S200 and S300 for calculation.

[0065] S401: Context state acquisition and matching rule selection The system first queries the user's dynamic profile to obtain their current learning status label, and at the same time, obtains the current effect type label of all educational service profiles in the candidate pool.

[0066] Furthermore, as a preferred embodiment, the matching model dynamically selects matching rules that are compatible with the learning stage and the applicable stage based on the user's current learning stage parsed from the dynamic user profile and the service applicable stage identified by the dynamic education service profile.

[0067] The mapping relationship between the preset classification combination and the matching rule is as follows: when the user is in the improvement period and the service is a leap type, the first matching rule is selected, which gives high weight to the deep skill features in the user profile and the "accelerator" label in the service profile. When a user is in a bottleneck period and the service is in a consolidation phase, the second matching rule is selected. This rule assigns high weight to adjacent basic knowledge point features in the user profile and the "basic consolidation" tag in the service profile.

[0068] This mapping relationship is stored in the system in the form of a data table or configuration file. When the input is a user's learning stage label... Educational service effectiveness type tags At that time, by querying this table or configuration, the output consists of the specified matching rule identifier and the corresponding set of weight coefficients. .

[0069] This set of weighting coefficients defines the proportion of each feature dimension of the user profile and service profile in the similarity calculation under the corresponding rules.

[0070] S402: Rule-based matching degree calculation and recommendation generation For each candidate educational service, the system loads the matching rules and their weight coefficients determined in step S401. The matching degree is usually calculated using a weighted feature similarity measure.

[0071] Let the user profile feature vector be... The service profile feature vector is Matching rules The weight vector is defined. The calculation process can be represented as follows: ,in and It is a vector and In the The values ​​in each dimension, where sim is a function for calculating the similarity in that dimension. It is a rule In the Weight coefficients on each dimension.

[0072] After calculating the matching score for all candidate services, the system sorts them in descending order of score and selects the highest-ranked service. Each service generates a final recommendation list.

[0073] S500: Closed-Loop Feedback and Optimization After the recommendation list is generated and displayed to users, the system initiates a closed-loop feedback and optimization process, monitoring user feedback and using it to adjust internal system parameters.

[0074] S501: Feedback Behavior Data Collection and Quantification The system records the user's subsequent interactions with each item in the recommendation list. These interactions are transformed into quantifiable signals, such as whether a click was made, whether the learning was completed, the average depth of interaction during the learning process, and post-lesson assessment scores. The system then integrates these multi-dimensional signals into a comprehensive actual conversion metric.

[0075] A simple way to combine them is by weighted average: ,in For each dimension, and The actual conversion metric The value range of the feedback behavior is obtained through weighted calculation and normalized, and its matching degree with the prediction is determined. They are on the same dimension so that they can be compared directly.

[0076] S502: Feedback Data Distribution and Profile Update Triggering The system simultaneously uses the collected user feedback behavior data as input for the next round. Each user's individual behavior data is injected into their interaction behavior data stream, awaiting the triggering of step S200 for the next round of user profile updates.

[0077] Meanwhile, the user's behavior, as an example within the group, is aggregated into the post-service data of the corresponding educational services and participates in the next round of service profile updates in the S300 step.

[0078] Therefore, user feedback on the recommendation results serves as both new interactive behavior data and new group effect data, driving the continuous and collaborative evolution of user profiles and educational service profiles.

[0079] S503: Fine-tuning of matching rule weight parameters The system uses feedback data to optimize the parameters of the matching rules themselves. This sub-step aims to optimize the parameters of the matching rules based on the deviation between the actual recommended effect and the predicted effect.

[0080] Furthermore, as a preferred embodiment, the weighting coefficients of the corresponding matching rules in the modified matching rule mapping table include: Record the predicted matching degree calculated by the matching rules; Obtain a conversion metric for the actual actions taken by users based on the recommendation; Calculate the difference between the actual behavior conversion metric and the predicted matching degree; Based on the sign and magnitude of the difference value, the weighting coefficients corresponding to the main feature terms that cause the difference in the matching rule are adjusted in the same or opposite direction according to a preset adjustment step size.

[0081] In practice, the system will record the matching rules used to generate this recommendation in step S402. and the calculated predicted matching degree In step S501, the actual conversion metric for this recommendation is calculated. The system calculates the prediction error: ,in, For actual conversion measurement, To predict the matching degree.

[0082] One implementation method for adjusting weight coefficients is as follows: First, identify the feature dimensions that significantly affect the current matching result, and then adjust them based on the difference values. The direction and magnitude of these dimensions are determined, and the corresponding weight coefficients are updated according to preset rules. For example, the following strategy can be used: The method for identifying the "main feature that causes the difference" is as follows: calculate the matching degree in this case. In the middle, the contribution of each feature dimension Select the top contributors Each feature dimension is used as the primary feature term.

[0083] Subsequently, the system adjusts the rules according to the direction of the error. The weighting parameters were slightly adjusted.

[0084] like Then the weight coefficients corresponding to these main feature terms Increase by a tiny step ; like Then reduce the step size of these weight coefficients. .

[0085] in, The first in the matching rule Weight coefficients for each feature dimension Representing the user profile 3D features With service portrait 3D features The similarity.

[0086] Preset adjustment step size It is a small positive number to ensure the stability of parameter updates and avoid drastic fluctuations in matching rules due to the randomness of a single feedback.

[0087] S600: Continuously execute steps The steps S100 to S500 above describe a complete data-driven processing cycle. In actual operation, data collection is continuous, and user profile and service profile updates are triggered asynchronously or periodically based on the arrival of new data. Matching recommendations occur in real time upon user request, while closed-loop optimization is performed after each recommendation generates feedback.

[0088] These steps operate continuously in a loop, forming a closed-loop dynamic optimization system. In this system, user profiles evolve based on individual behavioral data and state judgment rules, education service profiles evolve based on group effect data and effect classification rules, and matching rules are dynamically selected based on the states evolved from the former two, and parameters are fine-tuned based on the difference between the matching results and the actual feedback.

[0089] Then, by coupling the data stream and feedback signal, a system capable of coordinated and dynamic evolution is formed, realizing an internal parameter adjustment mechanism based on feedback.

[0090] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic matching optimization method integrating user profiles and education service profiles, characterized in that, The method includes the following steps: S1. Dynamically update user profiles based on user interaction behavior data, and dynamically update education service profiles based on the group effect data of education services; S2. Based on the user learning stage represented by the user profile and the service effect type represented by the education service profile, dynamically select matching rules and calculate the matching degree between the user profile and the education service profile. S3. Generate educational service recommendations based on the matching degree; S4. Collect user feedback data on the recommendations; S5. Based on the feedback behavior data, update the user profile and the education service profile, and optimize the matching rules used in step S2.

2. The dynamic matching optimization method for integrating user profiles and education service profiles according to claim 1, characterized in that, Step S1, "dynamically updating user profiles based on user interaction behavior data," includes: S11. Calculate user behavior indicators based on the interaction behavior data, and determine the user's current learning stage based on the preset indicator-stage mapping rules. S12. Adjust the feature weights of the user profile according to the first update rule corresponding to the learning stage.

3. The dynamic matching optimization method for integrating user profiles and education service profiles according to claim 1, characterized in that, Step S1, "Dynamically updating the education service profile based on the group effect data of education services," includes: S13. Calculate the service effect index based on the group effect data, and determine the current effect type of the education service based on the preset index-type mapping rule; S14. Adjust the tag strength and association of the education service profile according to the second update rule corresponding to the effect type.

4. The dynamic matching optimization method for integrating user profiles and education service profiles according to claim 1, characterized in that, In step S2, "dynamically selecting matching rules" means: based on the combination of the user's learning stage and the service effect type, querying a preset matching rule mapping table and selecting the corresponding matching rule.

5. The dynamic matching optimization method for integrating user profiles and education service profiles according to claim 1, characterized in that, The "optimization of the matching rules used in step S2" in step S5 includes: S51. Obtain the predicted matching degree calculated by the matching rule, and the actual conversion metric determined based on the feedback behavior data; S52. Calculate the difference between the predicted matching degree and the actual conversion metric; S53. Adjust the weight coefficients in the matching rules based on the difference value.

6. A dynamic matching and optimization system that integrates user profiles and educational service profiles, characterized in that, The system includes: The profile collaborative update module is used to dynamically update user profiles based on user interaction behavior data, and to dynamically update education service profiles based on the group effect data of education services. The dynamic matching decision module, connected to the profile collaborative update module, is used to dynamically select matching rules based on the user learning stage parsed from the user profile and the service effect type extracted from the education service profile, and to calculate the matching degree between the user profile and the education service profile. The recommendation generation module is connected to the dynamic matching decision module and is used to generate educational service recommendations based on the matching degree. The feedback collection module is used to collect user feedback behavior data on the recommendations; The closed-loop learning optimization module is connected to the feedback acquisition module, the profile collaborative update module, and the dynamic matching decision module, respectively. It is used to trigger the profile collaborative update module to perform updates based on the feedback behavior data and to optimize the parameters of the matching rules used in the dynamic matching decision module.

7. The dynamic matching and optimization system for integrating user profiles and education service profiles according to claim 6, characterized in that, The portrait collaborative update module includes: The user profile update unit is used to calculate user behavior indicators based on user interaction behavior data, determine the user learning stage based on a preset indicator-stage mapping rule, and adjust the feature weights of the user profile according to the first update rule corresponding to the learning stage. The service profile update unit is used to calculate service performance indicators based on the group performance data of education services, determine the service performance type based on the preset indicator-type mapping rules, and adjust the label strength and correlation of the education service profile according to the second update rule corresponding to the performance type.

8. The dynamic matching and optimization system for integrating user profiles and education service profiles according to claim 6, characterized in that, The dynamic matching decision module includes: The rule decision unit is used to query a preset matching rule mapping table based on the combination of the user's learning stage and the type of educational service effect, so as to select the corresponding matching rule; A matching calculation unit, connected to the rule decision unit, is used to calculate the matching degree between the user profile and the education service profile using the selected matching rules.

9. The dynamic matching and optimization system for integrating user profiles and educational service profiles according to claim 6, characterized in that, The closed-loop learning optimization module includes: The data distribution unit, connected to the feedback collection module, is used to send the feedback behavior data to the user profile update unit and service profile update unit in the profile collaborative update module, respectively, to trigger their update operations; The rule optimization unit is used to obtain the predicted matching degree output by the dynamic matching decision module and the actual conversion metric determined based on feedback behavior data, calculate the difference between the two, and adjust the weight coefficients of relevant matching rules in the dynamic matching decision module according to the difference.

10. The dynamic matching and optimization system for integrating user profiles and educational service profiles according to claim 9, characterized in that, The rule optimization unit is specifically configured to: identify feature dimensions that contribute more than a preset threshold to the predicted matching degree, and adjust the weight coefficients of the identified feature dimensions in the matching rules according to the direction of the difference value.

Images