Service information recommendation method and system fusing user tag data

Abstract

The invention discloses a service information recommendation method and system fusing user tag data, and the method comprises the steps: calling user historical access data, and generating user tag data; performing preliminary analysis on the user label data by introducing an NLP model of an XGBoost attention mechanism, generating an initial weight ratio and a service information link, and displaying the service information link on a user interface at the top; obtaining operation feedback information of a user on the service information link, analyzing and processing the operation feedback information based on a Q-learning framework, and adjusting an initial weight proportion; and performing optimization calculation by combining the optimization weight proportion and the operation feedback information with the user tag data, and displaying an optimized service information link. By dynamically adjusting the recommendation weight and combining real-time feedback of the user, the problem of insufficient accuracy caused by static weight of a traditional recommendation system is solved, and the method has the advantages of dynamically optimizing the recommendation weight and improving personalized recommendation accuracy.

Description

Technical Field

[0001] The present invention relates to the field of information service technology, and in particular to a service information recommendation method and system integrating user tag data. Background Art

[0002] In today's Internet and mobile application ecosystem, conventional information service push is based on fixed information content and is pushed and displayed according to set trigger conditions. User portraits or tag systems are built in combination with user historical data to push personalized information services in a fixed pattern. Although this method achieves a certain degree of personalization, the lack of dynamic scheduling leads to a serious lag in information service updates. Recommended content is difficult to effectively optimize based on real-time user feedback, resulting in the system being unable to keenly capture short-term changes in user interests or immediate intentions, resulting in poor information service recommendation results and a decline in user experience. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology. The present invention provides a service information recommendation method and system that integrates user tag data. By obtaining user feedback data on information service links in real time and optimizing the keyword weights of user tag data information, the information service recommendation effect of dynamic adjustment and optimization is achieved, thereby improving the accuracy of information service recommendations and the user experience.

[0004] The present invention provides a service information recommendation method integrating user tag data, the recommendation method comprising:

[0005] S11: Detect the user account login status, retrieve the user's historical access data, and generate user tag data;

[0006] S12: Performing a preliminary analysis on the user tag data by introducing an NLP model with an XGBoost attention mechanism, generating an initial weight ratio, calculating and generating a service information link based on the initial weight ratio, and displaying the service information link at the top of the user interface;

[0007] S13: Obtaining user operation feedback information on the service information link, analyzing and processing the operation feedback information based on a Q-learning framework, adjusting the initial weight ratio, and generating an optimized weight ratio;

[0008] S14: Optimizing the weight ratio and the operation feedback information in combination with the user tag data to perform optimization calculations, and displaying the optimized service information link on the user interface.

[0009] Furthermore, the step S11 includes:

[0010] Obtain a login request signal from a user account, retrieve historical access data of the corresponding user account from a database based on the user account information, and set the historical access data as user tag data;

[0011] During the use of the user account, the access information content of the user account is detected in real time, and the user tag data is updated in real time according to the access information content.

[0012] Furthermore, the step S12 includes:

[0013] S121: Extracting several keywords from user tag data based on the NLP semantic recognition model;

[0014] S122: Perform word frequency analysis on several keywords using the XGBoost attention mechanism algorithm, and set initial weight ratios of several keywords based on the word frequency analysis data;

[0015] S123: extracting one or more keywords from the keywords based on the initial weight ratio, and generating a service information link according to the extracted keywords.

[0016] Furthermore, the step S121 includes:

[0017] Perform semantic analysis on user tag data through semantic models and classify the user tag data into several semantic categories based on the data content;

[0018] The data content in the user tag data is classified one by one into several semantic categories.

[0019] Furthermore, the initial weight calculation formula for each keyword is:

[0020] ;

[0021] in, is the weight of the ith keyword, is the original score of the i-th keyword, is the original score of the jth keyword, It is a positive operation.

[0022] Furthermore, the step S123 includes:

[0023] According to the semantic categories of the extracted keywords, the keywords are arranged and combined according to the semantic categories to generate a number of service information links;

[0024] The priority of the service information link is calculated according to the weight ratio of the keywords in each service information link, and the plurality of service information links are sorted according to the priority.

[0025] Furthermore, the step S13 includes:

[0026] Detecting user operations on service information links and generating feedback information data;

[0027] The feedback information data is weighted and optimized through the Q-learning framework, and the optimized weight ratio is output.

[0028] Furthermore, the calculation formula for weight optimization calculation is:

[0029] ;

[0030] in, is the weight value of the t+1th iteration, is the weight value at the tth iteration, is the learning rate, is the feedback data at time t, is the feedback fluctuation at time t, is the minimum value, is the feedback influence coefficient, is the number of clicks at time t, is the number of clicks at the initial moment.

[0031] Furthermore, the step S14 includes:

[0032] The real-time access page layout of the user account is obtained, and the display position of the service information link is adjusted according to the access page layout.

[0033] The present invention also provides a service information recommendation system integrating user tag data, the recommendation system comprising:

[0034] Detection component: used to detect user account login status, retrieve user historical access data, and generate user tag data;

[0035] Service link component: Used to perform preliminary analysis of user tag data using an NLP model that introduces the XGBoost attention mechanism, generate an initial weight ratio, calculate and generate a service information link based on the initial weight ratio, and display the service information link at the top of the user interface;

[0036] Weight optimization component: used to obtain user operation feedback information on the service information link, analyze and process the operation feedback information based on the Q-learning framework, adjust the initial weight ratio, and generate an optimized weight ratio;

[0037] Link optimization component: used to combine the optimization weight ratio and operation feedback information with user tag data for optimization calculation, and display the optimized service information link on the user interface.

[0038] The present invention provides a service information recommendation method and system that integrates user tag data. By dynamically adjusting the recommendation weights and combining them with real-time user feedback, it solves the problem of insufficient accuracy caused by static weights in traditional recommendation systems. It has the advantages of dynamically optimizing recommendation weights and improving the accuracy of personalized recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 This is a flow chart of a service information recommendation method integrating user tag data in an embodiment of the present invention;

[0041] Figure 2 This is a specific operation flow chart of step S12 in an embodiment of the present invention;

[0042] Figure 3 2 is a schematic diagram of a service information recommendation system integrating user tag data in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Example 1:

[0045] Figure 1 A flow chart of a service information recommendation method integrating user tag data in an embodiment of the present invention is shown. The service information recommendation method includes:

[0046] S11: Detect the user account login status, retrieve the user's historical access data, and generate user tag data.

[0047] Detecting the login status of a user account refers to real-time monitoring of the login behavior of a user account, including: obtaining the login request signal of the user account based on real-time detection of the system, retrieving the historical access data of the corresponding user account in the database based on the user account information, and setting the historical access data as user tag data, scheduling the historical access data of the user account in the system, including the browsing history, search history, data reading history, etc. of the user account, and using the historical access data as the user's tag data to provide data support for the user's personalized service information recommendation.

[0048] After the login request signal of the user account is obtained, the user account information is matched with the historical access data in the database through a unique identifier. The historical access data is retrieved and used as the initial benchmark for the user tag data. During the use of the user account, the access information content is detected in real time, and the detection frequency can be set to collect page dwell time, click events or search keywords every 5 seconds. After the access information content is parsed into structured data, the user tag data is dynamically adjusted through an incremental update algorithm. For example, a sliding window mechanism is used to retain behavioral data within the last 30 minutes, and overwrite old data in the initial benchmark that exceeds the set time threshold. Real-time updated user tag data is stored in an independent cache area, forming a hierarchical storage structure with historical access data. The cache area data update cycle can be set to 10 seconds.

[0049] During the use of the user account, the access information content of the user account is detected in real time, and the user tag data is updated in real time according to the access information content.

[0050] When the user account login request signal is triggered, the unique identifier in the user account information is used to query the database and retrieve the historical access records of the account in the past 6 months as the initial tag data. During the user's subsequent use, the real-time detection module continues to monitor the user's operation behavior. For example, the current page browsing content is semantically analyzed and keywords are extracted. When the frequency of keyword occurrence exceeds the preset threshold, the weight of the corresponding interest tag in the user tag data is increased. When a user visits the same category more than 3 times in a row, the priority of the category tag is increased to 1.5 times the corresponding tag weight in the initial benchmark data. The updated user tag data is synchronized to the input of the recommendation model, so that the model can perform weight calculations based on a data set containing real-time behavioral features. By real-time monitoring and updating of user tag data, the user's short-term behavioral changes can be immediately fed back to the tag system for dynamic service information recommendation updates, so that relevant service information links are displayed first in subsequent recommendation results.

[0051] S12: The NLP model with the XGBoost attention mechanism is introduced to preliminarily analyze the user label data, generate an initial weight ratio, calculate and generate a service information link according to the initial weight ratio, and display the service information link on the top of the user interface.

[0052] Specifically, Figure 2 A specific operation flowchart of step S12 in the embodiment of the application is shown, and the step S12 includes:

[0053] S121: Extracting a plurality of keywords from the user label data based on an NLP semantic recognition model;

[0054] Specifically, the NLP semantic recognition model can realize semantic parsing by using a pre-training model based on a Transformer architecture, encode the user label data by using a BERT model, and extract entity words and behavior feature words with semantic correlation as candidate keywords.

[0055] Further, the step S121 includes:

[0056] The semantic model is used to perform semantic analysis on the user label data, and a plurality of semantic categories are divided according to the data content of the user label data;

[0057] The data content in the user label data is correspondingly induced in the plurality of semantic categories.

[0058] The division of the semantic categories is realized according to a context similarity threshold value, specifically by calculating a cosine similarity value, and when the vector similarity of two label data exceeds 0.85, they are classified into the same semantic category. The corresponding relationship between the data content and the semantic categories is maintained by a dynamic mapping table, which is updated every hour according to the user behavior data, and the historical mapping records in the last 30 days are reserved as training samples during the update. The number of semantic categories is controlled within the interval of 5-8, and when a new semantic category is detected to exceed the threshold value, a category merging operation based on a density clustering algorithm is triggered, and the merging radius parameter is set to 0.3.

[0059] Further, the keywords of the user label data are divided into semantic categories after semantic recognition, so as to classify the keywords of the user label data and realize the combination and arrangement of a plurality of keywords, thereby constructing a service information link conforming to the user label data.

[0060] Specifically, when the user tag data contains discrete keywords such as "sofa" and "dining table", the semantic model first extracts the context vector of each keyword through a bidirectional encoder, and then uses a hierarchical clustering algorithm to classify keywords with a vector distance less than 0.2 into the "furniture selection" semantic category. In this process, the original position information of the keywords is retained in the metadata field of the mapping table for subsequent time sequence correlation analysis in the weight calculation stage. In the service information linking stage, keywords within the same semantic category are forced to be combined into associated word groups, for example, "decoration style" and "furniture selection" form a combined feature under the "home design" semantic category, and the combined weight is calculated through the interaction term in the XGBoost attention mechanism, and the weight adjustment coefficient is set to 0.2-0.5. The semantic units generated in this way participate in reward value distribution as independent decision nodes in the Q-learning optimization stage, so that the click rate feedback can be accurately traced back to the feature optimization of a specific semantic dimension. This scheme improves the semantic coverage of the recommendation link to more than 92%, while reducing the mismatch rate to less than 8%.

[0061] S122: Perform word frequency analysis on a plurality of keywords through an XGBoost attention mechanism algorithm, and set initial weight proportions of the plurality of keywords according to the word frequency analysis data;

[0062] Specifically, the original score of each keyword is generated by the XGBoost attention mechanism algorithm, and by inputting the original score into the exponential function, all negative scores are converted into positive numbers approaching zero, while positive scores are exponentially increased. For example, when the original scores of two keywords are -0.5 and 1.2 respectively, after exponential operation, they are converted to about 0.61 and 3.32 respectively, and the difference multiple expands from 1.7 times to 5.4 times. Then, normalization processing is performed, and the exponential score of each keyword is divided by the sum of the exponential scores of all keywords, so that the weight value is distributed in the range of 0 to 1 and the sum is 1. The weight value is used for priority calculation of service information linking, and through weight mapping and taking positive operation by the XGBoost attention mechanism, the calculation of the two can be simplified.

[0063] The XGBoost attention mechanism algorithm can build a word frequency weight mapping relationship through an integrated decision tree model, where each decision tree corresponds to a word frequency distribution pattern, and the weight distribution strategy is iteratively optimized through gradient boosting, for example, the word frequency statistics and user historical click behavior data are jointly input into the model to generate dynamic weight values reflecting user preference intensity.

[0064] The initial weight calculation formula of each keyword is:

[0065] ;

[0066] wherein, is the weight of the ith keyword, is the original score of the i-th keyword, is the original score of the jth keyword, It is a positive operation.

[0067] During the keyword weight calculation process, the original score of each keyword is first obtained. For example, for the five keywords A, B, C, D, and E, the original scores are 2.5, -1.2, 3.8, 0.7, and 1.9, respectively. Next, an exponential operation is performed on each original score, resulting in exp(2.5) = 12.18, exp(-1.2) = 0.30, exp(3.8) = 44.70, exp(0.7) = 2.01, and exp(1.9) = 6.69. The sum of all the exponential operation results is then calculated: 12.18 + 0.30 + 44.70 + 2.01 + 6.69 = 65.88.

[0068] Finally, the exponential calculation result for each keyword is divided by the sum to obtain the final weight: A's weight is 12.18 / 65.88 = 0.185, B's weight is 0.30 / 65.88 = 0.005, C's weight is 44.70 / 65.88 = 0.679, D's weight is 2.01 / 65.88 = 0.031, and E's weight is 6.69 / 65.88 = 0.102. An exponential function is used to perform a nonlinear transformation on the raw scores, converting scores within any real range into positive numbers and preventing negative scores from interfering with the weight calculation. Furthermore, the exponential calculation amplifies differences in scores, significantly increasing the weights of high-scoring keywords. Normalization constrains the weights to the range of 0 to 1, ensuring that the sum of the weights is 1, resulting in a stable and comparable weight ratio. This calculation method enhances the rationality of weight distribution, provides a mathematical basis for prioritizing service information links, and ensures that recommendation results are more tailored to users' actual needs.

[0069] S123: extracting one or more keywords from the keywords based on the initial weight ratio, and generating a service information link according to the extracted keywords.

[0070] Specifically, step S123 includes:

[0071] According to the semantic categories of the extracted keywords, the keywords are arranged and combined according to the semantic categories to generate a number of service information links;

[0072] The priority of the service information link is calculated according to the weight ratio of the keywords in each service information link, and the plurality of service information links are sorted according to the priority.

[0073] When generating service information links, the extracted keywords are first classified according to their semantic categories. For example, keywords may be distributed across the semantic categories of "housing," "decoration," and "home." The "housing" category might include keywords such as "school district housing" and "subway housing"; the "decoration" category might include keywords such as "fine decoration" and "simple decoration"; and the "home" category might include keywords such as "smart appliances" and "Nordic style."

[0074] Next, keywords within the same semantic category are permuted and combined to generate multiple service information links. For example, links such as "school district housing + fine decoration," "subway housing + simple decoration," and "school district housing + smart appliances" can be generated. This combination ensures that the keywords within the links have clear semantic relevance.

[0075] The link priority is then calculated based on the weight ratio of the keywords within each service information link. For example, if the weight of "school district housing" is 0.4 and the weight of "fine decoration" is 0.3, the priority of the link for "school district housing + fine decoration" might be 0.7. In contrast, if the weight of "subway housing" is 0.2 and the weight of "simple decoration" is 0.1, the priority of the link for "subway housing + simple decoration" would be 0.3.

[0076] Finally, all generated service information links are sorted according to the calculated priority. Links with higher priorities are placed first and presented to users first. This ensures semantic coherence of links while reflecting user preferences.

[0077] S13: Obtaining user operation feedback information on the service information link, analyzing and processing the operation feedback information based on a Q-learning framework, adjusting the initial weight ratio, and generating an optimized weight ratio.

[0078] Specifically, step S13 includes:

[0079] Detect the user's operation behavior on the service information link and generate feedback information data, which includes: click data, dwell time, etc., wherein the click data includes valid click action data of the service information link being clicked to view, invalid click data of the service information link being clicked to close, and other click data. The dwell time is the length of time the service information link stays on the user interface, that is, the length of time from the generation of the service information link to the time it is clicked to view or clicked to close.

[0080] The feedback information data is weighted and optimized through the Q-learning framework, and the optimized weight ratio is output.

[0081] The calculation formula for weight optimization calculation is:

[0082] ;

[0083] in, is the weight value of the t+1th iteration, is the weight value at the tth iteration, is the learning rate, is the feedback data at time t, is the feedback fluctuation at time t, is the minimum value, is the feedback influence coefficient, is the number of clicks at time t, is the number of clicks at the initial moment.

[0084] Furthermore, in this embodiment, taking click-through rate as an example, by introducing the Q-learning framework, the system can establish a dynamic mapping relationship between feedback data and weight adjustment, effectively responding to real-time changes in user interests. This adaptive learning mechanism enables the recommendation system to more accurately match the user's dynamic intentions, improving the relevance of recommended content and user experience. At the same time, because Q-learning's strategy optimization takes into account long-term cumulative rewards, the system not only pays attention to immediate feedback when adjusting weights, but also predicts and optimizes long-term effects, thereby gradually improving the quality of recommendations in the continuous interaction process. This dynamic and forward-looking optimization method significantly improves the recommendation system's response speed and adaptability to changes in user preferences.

[0085] S14: Optimizing the weight ratio and the operation feedback information in combination with the user tag data to perform optimization calculations, and displaying the optimized service information link on the user interface.

[0086] The real-time access page layout of the user account is obtained, and the display position of the service information link is adjusted according to the access page layout.

[0087] Specifically, taking a shopping website as an example, when a user visits an e-commerce product details page, the document object model tree structure of the current page is parsed in real time to identify the layout distribution of the main content area, product parameter module and user comment area of ​​the page. By calculating the visual density values ​​of each area of ​​the page, it is detected that there is a blank block with a width of 280 pixels in the right sidebar area and the area remains fixed when scrolling. Based on the layout analysis results, the service information link is dynamically inserted into the second-level container of the blank area of ​​the sidebar, and the link display form is set to a floating card component that matches the page style. The card height adapts to the link content and moves synchronously with the scrolling of the page. When the user switches to mobile browsing, it is detected that there is a fixed navigation bar with a height of 80 pixels at the bottom of the page. The service information link is adjusted to be displayed 10 pixels above the navigation bar, and a semi-transparent background is used to avoid blocking the main content of the page.

[0088] Through the above technical solution, this application can dynamically select the optimal display area based on the actual rendering structure of the page, using the fixed positioning characteristics of the sidebar on the desktop to achieve continuous exposure of recommended displays, and adapting to the bottom navigation interaction habits on the mobile terminal for precise reach. This solution effectively solves the problem of insufficient adaptability of traditional recommendation positions to the dynamic layout of the page, maintaining the integrity of the page content through an embedding method that minimizes visual interference, while utilizing the inherent interactive hotspots of the page to improve the visibility and click-through conversion efficiency of the recommended content.

[0089] The embodiments of the present invention provide a service information recommendation method and system that integrates user tag data. By dynamically adjusting the recommendation weights and combining real-time user feedback, it solves the problem of insufficient accuracy caused by static weights in traditional recommendation systems. It has the advantages of dynamically optimizing recommendation weights and improving the accuracy of personalized recommendations.

[0090] Example 2:

[0091] Figure 3 A schematic diagram of a service information recommendation system integrating user tag data according to an embodiment of the present invention is shown. The recommendation system includes:

[0092] Detection component 10: used to detect the user account login status, retrieve the user's historical access data, and generate user tag data; the user's historical access data includes: the user account's browsing history, search history, data reading history, etc., and the historical access data is used as the user's tag data to provide data support for the user's personalized service information recommendation.

[0093] Service link component 20: used to perform preliminary analysis on user tag data by introducing an NLP model with XGBoost attention mechanism, generate an initial weight ratio, calculate and generate a service information link based on the initial weight ratio, and display the service information link at the top of the user interface.

[0094] The weight optimization component 30 is used to obtain the user's operational feedback information on the service information link, analyze and process the operational feedback information based on the Q-learning framework, adjust the initial weight ratio, and generate an optimized weight ratio;

[0095] Link optimization component 40: used to combine the optimization weight ratio and operation feedback information with user tag data to perform optimization calculation and display the optimized service information link on the user interface.

[0096] The recommendation system includes a detection component 10, which retrieves historical access data in real time when the user logs in and generates dynamically updated user tag data. The service link component 20 uses the XGBoost attention mechanism and natural language processing model to perform semantic analysis on the tag data, extract keywords and calculate the initial weight ratio, generate the service information link and display it at the top area of ​​the user interface. The weight optimization component 30 collects user click and stay time operation feedback through the Q-learning framework, constructs a state-action matrix to dynamically correct the initial weight. Based on the real-time page layout characteristics, the link optimization component 40 adaptively adjusts the optimized service information link to the blank area of ​​the interface or the high-frequency interaction area for display.

[0097] Through the above technical solutions, this application solves the problems of data update lag and recommendation bias caused by insufficient coordination between modules in traditional recommendation systems. The collaborative operation of the detection component and the service link component ensures the timeliness of user tag data and the accuracy of initial recommendations; the weight optimization component uses a reinforcement learning mechanism to dynamically adjust weight parameters, effectively capturing user interest migration; the link optimization component optimizes display positions based on interface layout features, improving user operation efficiency. These components form a data closed loop, enabling adaptive iteration of the entire process of recommendation results, from generation to optimization.

[0098] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0099] In addition, the above is a detailed introduction to a service information recommendation method and system that integrates user tag data provided by an embodiment of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A service information recommendation method integrating user tag data, characterized in that: The recommended methods include: S11: Detect the user account login status, retrieve the user's historical access data, and generate user tag data; S12: Performing a preliminary analysis on the user tag data by introducing an NLP model with an XGBoost attention mechanism, generating an initial weight ratio, calculating and generating a service information link based on the initial weight ratio, and displaying the service information link at the top of the user interface; S13: Obtaining user operation feedback information on the service information link, analyzing and processing the operation feedback information based on a Q-learning framework, adjusting the initial weight ratio, and generating an optimized weight ratio; S14: Optimizing the weight ratio and the operation feedback information in combination with the user tag data to perform optimization calculations, and displaying the optimized service information link on the user interface.

2. The service information recommendation method integrating user tag data according to claim 1 is characterized in that: The step S11 includes: Obtain a login request signal from a user account, retrieve historical access data of the corresponding user account from a database based on the user account information, and set the historical access data as user tag data; During the use of the user account, the access information content of the user account is detected in real time, and the user tag data is updated in real time according to the access information content.

3. The service information recommendation method integrating user tag data according to claim 1, characterized in that: The step S12 includes: S121: Extracting several keywords from user tag data based on the NLP semantic recognition model; S122: Perform word frequency analysis on several keywords using the XGBoost attention mechanism algorithm, and set initial weight ratios of several keywords based on the word frequency analysis data; S123: extracting one or more keywords from the keywords based on the initial weight ratio, and generating a service information link according to the extracted keywords.

4. The service information recommendation method integrating user tag data according to claim 3 is characterized in that: The step S121 includes: Perform semantic analysis on user tag data through semantic models and classify the user tag data into several semantic categories based on the data content; The data content in the user tag data is classified one by one into several semantic categories.

5. The service information recommendation method integrating user tag data according to claim 3 is characterized in that: The initial weight calculation formula for each keyword is: ； in, is the weight of the i-th keyword, is the original score of the i-th keyword, is the original score of the jth keyword, It is a positive operation.

6. The service information recommendation method integrating user tag data according to claim 4 is characterized in that: The step S123 includes: According to the semantic categories of the extracted keywords, the keywords are arranged and combined according to the semantic categories to generate a number of service information links; The priority of the service information link is calculated according to the weight ratio of the keywords in each service information link, and the plurality of service information links are sorted according to the priority.

7. The service information recommendation method integrating user tag data according to claim 1, characterized in that: The step S13 includes: Detecting user operations on service information links and generating feedback information data; The feedback information data is weighted and optimized through the Q-learning framework, and the optimized weight ratio is output.

8. The service information recommendation method integrating user tag data according to claim 7 is characterized in that: The calculation formula for weight optimization calculation is: ； in, is the weight value of the t+1th iteration, is the weight value at the tth iteration, is the learning rate, is the feedback data at time t, is the feedback fluctuation at time t, is the minimum value, is the feedback influence coefficient, is the number of clicks at time t, is the number of clicks at the initial moment.

9. The service information recommendation method integrating user tag data according to claim 1, characterized in that: The step S14 includes: The real-time access page layout of the user account is obtained, and the display position of the service information link is adjusted according to the access page layout.

10. A service information recommendation system integrating user tag data, characterized in that: The recommendation system includes: Detection component: used to detect user account login status, retrieve user historical access data, and generate user tag data; Service link component: Used to perform preliminary analysis of user tag data using an NLP model that introduces the XGBoost attention mechanism, generate an initial weight ratio, calculate and generate a service information link based on the initial weight ratio, and display the service information link at the top of the user interface; Weight optimization component: used to obtain user operation feedback information on the service information link, analyze and process the operation feedback information based on the Q-learning framework, adjust the initial weight ratio, and generate an optimized weight ratio; Link optimization component: used to combine the optimization weight ratio and operation feedback information with user tag data for optimization calculation, and display the optimized service information link on the user interface.

Images