Internet-based advertisement monitoring method and system
By clustering and calculating interest levels in advertising text data, combined with dynamic delivery strategies, the problem of user fatigue caused by homogeneous ads has been solved, achieving freshness and efficient delivery of advertising content, and improving user experience and operational efficiency of the advertising platform.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-03-10
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Users experience visual and psychological fatigue from prolonged exposure to homogenized advertisements, leading to decreased click-through rates, loss of attention, and even active ad blocking. This damages users' trust in the platform or brand and can trigger negative emotions or complaints.
By clustering advertising text data and calculating interest levels, and intelligently switching to new clusters when interest values decrease, combined with random rotation and periodic removal mechanisms, the advertising delivery strategy is dynamically adjusted to ensure the relevance and freshness of the advertising theme and avoid fatigue caused by repeated exposure.
This approach achieves the goals of maintaining user interest, keeping ad content fresh, extending interaction time, improving display effects, reducing active blocking and negative feedback, increasing user acceptance, optimizing traffic allocation on the advertising platform, and reducing manual monitoring costs.
Smart Images
Figure CN121639284A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of advertising and promotion technology, and specifically to an internet-based advertising monitoring method and system. Background Technology
[0002] Advertising, as the name suggests, is the act of widely disseminating information to the general public. Economic advertising refers to advertising aimed at profit, typically commercial advertising. It is a paid means of disseminating information about goods or services to consumers or users through advertising media to promote those goods or services.
[0003] By deeply analyzing user behavior data, advertising systems can accurately identify user interests and preferences. However, this also leads to users constantly receiving highly similar ads, resulting in a lack of novelty. Prolonged exposure to the same type of content causes visual and psychological fatigue, gradually making users immune to advertising information and significantly reducing their willingness to click. As the rate of repeated exposure increases, user attention quickly slips away, and some users may even actively block or skip ads. Furthermore, homogenized advertising can damage users' overall impression of the platform or brand, reduce trust, and trigger negative emotions or complaints. Summary of the Invention
[0004] The purpose of this invention is to provide an internet-based advertising monitoring method and system, addressing the following technical problems: By deeply analyzing user behavior data, advertising systems can accurately identify user interests and preferences. However, this also leads to users constantly receiving highly similar ads, resulting in a lack of novelty. Prolonged exposure to the same type of content causes visual and psychological fatigue, gradually making users immune to advertising information and significantly reducing their willingness to click. As the rate of repeated exposure increases, user attention quickly slips away, and some users may even actively block or skip ads. Furthermore, homogenized advertising can damage users' overall impression of the platform or brand, reduce trust, and trigger negative emotions or complaints.
[0005] The objective of this invention can be achieved through the following technical solutions: An internet-based advertising monitoring method includes the following steps: Obtain text data from the advertisement, and cluster the advertisement based on the text data to obtain clusters; Obtain user behavior data for browsing individual advertisements, calculate user interest in the advertisements based on the behavior data, and calculate the interest value of the cluster. X j The interest level of the j-th advertisement in the cluster is represented by N, and the total number of advertisements in the cluster is represented by N. The clusters are sorted in descending order according to the size of their corresponding interest values to obtain a cluster ranking. The first cluster L1 in the cluster ranking is obtained. The interest values corresponding to the cluster L1 are periodically obtained. The interest values are sorted in time order to obtain a reference ranking. When the reference sorting meets the preset constraints, the advertising delivery strategy is adjusted.
[0006] As a further aspect of the present invention: the process of obtaining clusters includes: Collect the text data within the advertisement, preprocess the text data to obtain processed text, segment the processed text to obtain processed words, and generate a first set DYJ={A1, A2, ..., An}, where An represents the nth processed word and n represents the number of the processed word; Count the number of times the processing word corresponding to a single advertisement appears, and record it as the target number. Generate the second set DRJ of advertisement i. i ={B1 i B2 i , ..., Bm i}, Bm i This represents the target number of times the m-th processed word in advertisement i; Set the number of topics K, and use the first set and the second set as input to train the LDA model until the topic-word distribution and the advertisement-topic distribution converge. Determine the probability of a single topic for each of the processed words, and take the processed words with a probability greater than a preset value as subtopics of that topic. Sort the topics in ascending order according to the number of their corresponding subtopics, which is called quantity sorting.
[0007] As a further aspect of the present invention, the process of obtaining clusters further includes: Obtain the first-ranked topic C1 in the quantity ranking, obtain the sub-topics of topic C1, denoted as grouped sub-topics, and group the topics whose corresponding sub-topics contain all the grouped sub-topics and topic C1 into a cluster; Remove topic C1 from the quantity sort to obtain a new quantity sort. Repeat the above steps to obtain a new cluster. Repeat the above steps to obtain all clusters. Where no corresponding subtopic contains the topic of all the grouped subtopics, topic C1 is treated as a cluster. As a further aspect of the present invention: calculating a user's interest in an advertisement based on the behavioral data includes: The behavioral data includes the average duration of browsing ads, the total number of times ads are clicked, and the number of times ads are forwarded. The interest level is calculated based on the behavioral data and the superior-inferiority distance method.
[0008] As a further aspect of the present invention: the process of adjusting the advertising delivery strategy includes: Set constraints: F a <F1; F b <F1, b∈(a, a+M]; F a This represents the interest value at the a-th position in the reference sorting, where M is a preset value; The clusters other than cluster L1 are marked as undetermined clusters. The topic C1' that is first in the quantity ranking corresponding to the undetermined cluster is obtained. The sub-topic corresponding to topic C1' is obtained and recorded as undetermined sub-topic. Obtain the topic C1'' that is first in the quantity ranking corresponding to the cluster L1, and obtain the subtopic corresponding to the topic C1'', which is denoted as the comparison subtopic; Calculate the selection value XZ=X3 / |X1-X2| for a single undetermined cluster, where X3 represents the number of the undetermined sub-topics that are the comparison sub-topics, and X1 and X2 represent the number of the undetermined sub-topics and the comparison sub-topics, respectively. Take the undetermined cluster corresponding to the maximum selection value as the target cluster and deliver the advertisement belonging to the target cluster to the user.
[0009] As a further aspect of the present invention: the process of delivering advertisements belonging to the target cluster to users includes: The advertisements within the target cluster are randomly delivered to the user.
[0010] As a further aspect of the present invention, the process of adjusting the advertising delivery strategy also includes the following steps: When F1 is less than the preset threshold, remove F1 and obtain a new reference sort, and then perform the subsequent steps.
[0011] An internet-based advertising monitoring system includes: Clustering module: Obtains text data from advertisements, and clusters the advertisements based on the text data to obtain clusters; Analysis module: Acquires user behavior data for browsing individual advertisements, calculates user interest in the advertisements based on the behavior data, and calculates the interest value of the cluster. X j The interest level of the j-th advertisement in the cluster is represented by N, and the total number of advertisements in the cluster is represented by N. Adjustment module: Sort the clusters in descending order according to the size of their corresponding interest values to obtain a cluster ranking, obtain the first cluster L1 in the cluster ranking, periodically obtain the interest value corresponding to the cluster L1, and sort the interest values according to the time axis to obtain a reference ranking; When the reference sorting meets the preset constraints, the advertising delivery strategy is adjusted.
[0012] The beneficial effects of this invention compared to the prior art are as follows: 1) This invention monitors the interest values of each cluster in real time and sets a preset decay constraint. When the interest of the current first cluster shows a downward trend, it will intelligently trigger the switching between clusters and recommend the user to the next optimal cluster. This enables the background to manage the priority of each advertisement within the cluster and intelligently rotate them. This ensures that the advertisement theme is dynamically balanced between relevance and novelty, avoiding visual fatigue and psychological rejection caused by continuous exposure to the same type of content. Through this relevant but not identical display strategy, users can maintain their interest stickiness while keeping the advertisement content fresh, thereby extending the interaction time with the platform and improving the overall advertising display effect. 2) When replacing clusters, a selection value formula is constructed based on the sub-topic difference and the number of common sub-topics. This allows the system to prioritize target clusters that are both deeply matched to the user's current interests and have a moderate degree of difference. Furthermore, multi-dimensional feature fusion is used to further refine the candidate range. While ensuring user familiarity, new advertising elements with information gain are injected, balancing stable recommendations with the need for novel exploration. Through adaptive calculation and dynamic sorting, the system can flexibly adjust recommendation strategies according to different users' interest curves, effectively improving users' cognitive efficiency and curiosity regarding new content, stimulating sustained click intentions and interaction motivations, and ensuring a consistent experience even when users browse across different scenarios, thus constructing a more diverse and attractive advertising delivery model. 3) For multiple ad creatives within the same cluster, this solution employs a random rotation and periodic removal mechanism. By randomly shuffling the ad presentation order, the system can balance the exposure opportunities of each creative under the same theme, preventing visual fatigue caused by excessive frequency of a particular material. Simultaneously, periodically removing low-interest historical samples ensures the timeliness of interest estimation, enabling rapid response to the latest user feedback and forming a natural "cold start—heat boost—rebalancing" cycle. This allows users to encounter more diverse visual styles and perspectives in a short period, enhancing their overall understanding and immersion in the brand story, effectively reducing active blocking and negative feedback, and increasing user acceptance of the ads. 4) For advertising platforms and advertisers, this invention constructs a dynamic traffic allocation system based on interest and selection values. During the campaign, the system can continuously evaluate the marginal return potential of each cluster, accurately allocating limited traffic to the optimal cluster and avoiding budget waste due to audience saturation. The threshold monitoring mechanism automatically eliminates inefficient strategies in the background, reducing the cost of manual monitoring and testing, and making campaign adjustments more intelligent and efficient. Attached Figure Description
[0013] The invention will now be further described with reference to the accompanying drawings.
[0014] Figure 1 This is a flowchart illustrating an internet-based advertising monitoring method according to the present invention. Detailed Implementation
[0015] 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.
[0016] Please see Figure 1 As shown, this invention is an internet-based advertising monitoring method, comprising the following steps: Obtain text data from the advertisement, and cluster the advertisement based on the text data to obtain clusters; In a preferred embodiment of the present invention, the text data within the advertisement is collected, the text data is preprocessed to obtain processed text, the processed text is segmented to obtain processed words, and a first set DYJ={A1, A2, ..., An} is generated, where An represents the nth processed word and n represents the number of the processed word. Count the number of times the processing word corresponding to a single advertisement appears, and record it as the target number. Generate the second set DRJ of advertisement i. i ={B1 i B2 i , ..., Bm i}, Bm i This represents the target number of times the m-th processed word in advertisement i; Set the number of topics K, and use the first set and the second set as input to train the LDA model until the topic-word distribution and the advertisement-topic distribution converge. Determine the probability of a single topic for each of the processed words, and take the processed words with a probability greater than a preset value as sub-topics of that topic. Sort the topics in ascending order according to the number of their corresponding sub-topics, which is called quantity sorting. In a preferred embodiment of this invention, the process of obtaining clusters further includes: Obtain the first-ranked topic C1 in the quantity ranking, obtain the sub-topics of topic C1, denoted as grouped sub-topics, and group the topics whose corresponding sub-topics contain all the grouped sub-topics and topic C1 into a cluster; Remove topic C1 from the quantity sort to obtain a new quantity sort. Repeat the above steps to obtain a new cluster. Repeat the above steps to obtain all clusters. Where no corresponding subtopic contains the topic of all the grouped subtopics, the topic C1 is treated as a cluster; For example, the system first indiscriminately extracts all visible text and hidden fields from the entire platform—including titles, descriptions, landing page keywords, ALT text, tags, etc.—and aggregates them into a raw text pool. Then, it enters the cleaning phase: the program batch-removes HTML tags, punctuation marks, emoticons, URLs, and number strings; unifies full-width and half-width characters; performs dictionary merging on synonymous phrases (e.g., treating "discount," "offer," and "special offer" as related); and filters out non-informative functional words using a stop word list, making the remaining content more reflective of the advertisement's theme. Each cleaned advertisement is then fed into a word segmentation engine, where it is broken down into independent terms based on semantic boundaries and its word form is restored. The engine output is aggregated to form a platform-level thesaurus, i.e., the first set. Next, the system returns to each individual advertisement, counting the frequency of each word in the corresponding text, and saving this "word-frequency" list as the second set for that advertisement. Then, with a preset number of themes K, the system feeds the "global thesaurus + all advertisement word frequency lists" into a Latent Dirichlet Allocation (LDA) model. The model iteratively updates the "topic-word probability" and "advertisement-topic probability" until the perplexity no longer decreases, indicating that the topic structure has stabilized. At this point, the system traverses each topic, calculating the probability of occurrence for all its words; words with probabilities exceeding a threshold are marked as subtopics of that topic. Topics are then sorted from fewest to most subtopics for easier clustering later. During actual clustering, the program first takes the top-ranked topic from the list, treating all its subtopics as "grouped subtopics," and then scans the remaining topics. If a topic's set of subtopics completely covers these grouped subtopics, both are grouped into the same cluster; if no topic can achieve full coverage, the current topic becomes a separate cluster. After assignment, the processed topic is removed from the list, and the process is repeated for the new top-ranked topic until all topics have found a cluster. In this way, whether it is financial management advertisements that repeatedly mention "yield" and "annualized return", game promotions that frequently mention "dungeons" and "leveling up", or travel ads that frequently use "flight tickets" and "hotels", the system will aggregate them into semantically consistent clusters based on the overlapping relationship of sub-topics, providing a structured basis for subsequent interest assessment and dynamic delivery, without the need for manual labeling of massive amounts of ads one by one. Understandably, by cleaning, segmenting, and counting the word frequency of all ad text, and introducing LDA topic modeling, combined with subtopic overlap to achieve automatic clustering, the system can accurately classify massive amounts of ads semantically without manual intervention, significantly reducing the cost of manual annotation. High-frequency co-occurring words are extracted into subtopics, which can intuitively present the topic tags of each cluster, facilitating subsequent interpretation and monitoring; the sorting-coverage clustering strategy avoids misclassification or detachment of "long-tail" ads, improving consistency within clusters. The resulting structured ad pool lays the data foundation for subsequent interest calculation, dynamic rotation, and precise targeting, while also supporting rapid incremental updates when new ads are added, maintaining the model's real-time performance, and enabling the platform to efficiently and stably output differentiated targeting strategies when facing constantly changing ad creatives. Obtain user behavior data for browsing individual advertisements, calculate user interest in the advertisements based on the behavior data, and calculate the interest value of the cluster. X j The interest level of the j-th advertisement in the cluster is represented by N, and the total number of advertisements in the cluster is represented by N. In another preferred embodiment of the present invention, calculating the user's interest in the advertisement based on the behavioral data includes: The behavioral data includes the average duration of browsing ads, the total number of times ads are clicked, and the number of times ads are forwarded. The interest level is calculated based on the behavioral data and the superior-inferiority distance method. Understandably, the system first records three key behavioral metrics simultaneously while users browse ads: first, dwell time, which is the total time from when the ad appears until the user leaves; second, the cumulative number of clicks, such as clicking the play button, clicking a link, or expanding details; and third, the cumulative number of active forwarding behaviors, including sharing to social media or copying a link to send to friends. Once these three raw data points are collected, the system places the three-dimensional behavioral vector corresponding to each ad into the same comparison matrix and performs range normalization on the three dimensions of dwell time, clicks, and forwarding, ensuring that the relationship of higher values being better and lower values being worse remains consistent across all dimensions. Next, the algorithm finds two sets of reference points for each ad in the matrix: one set represents the ideal best-case scenario, where all three metrics reach the maximum value among ads in the same group; the other set represents the ideal worst-case scenario, where all three metrics fall at the minimum value. Subsequently, for any given ad, the Euclidean distance to the best and worst reference points is calculated, and the relative proximity is determined accordingly. The closer the ad is to the best point and the farther it is from the worst point, the more favored the current user is, and the higher its interest score. After scoring all ads in the matrix, the system returns to each previously defined cluster and averages the interest scores of all ads within the cluster to obtain a freshness attraction value representing the entire cluster. The higher this value, the more active the ads under that theme are, and the more priority they will receive in subsequent ad placement scheduling. If this value is observed to decrease continuously over time, the system will trigger an inter-cluster switch, directing traffic to the next high-potential cluster to prevent user fatigue due to repeated exposure. The clusters are sorted in descending order according to the size of their corresponding interest values to obtain a cluster ranking. The first cluster L1 in the cluster ranking is obtained. The interest values corresponding to the cluster L1 are periodically obtained. The interest values are sorted in time order to obtain a reference ranking. When the reference sorting meets the preset constraints, the advertising delivery strategy is adjusted; It is worth noting that the process of adjusting advertising placement strategies includes: Set constraints: F a <F1; F b <F1, b∈(a, a+M]; F a This represents the interest value at the a-th position in the reference sorting, where M is a preset value; The clusters other than cluster L1 are marked as undetermined clusters. The topic C1' that is first in the quantity ranking corresponding to the undetermined cluster is obtained. The sub-topic corresponding to topic C1' is obtained and recorded as undetermined sub-topic. Obtain the topic C1'' that is first in the quantity ranking corresponding to the cluster L1, and obtain the subtopic corresponding to the topic C1'', which is denoted as the comparison subtopic; Calculate the selection value XZ = X3 / |X1-X2| for a single undetermined cluster, where X3 represents the number of the undetermined sub-topics that are the comparison sub-topics, and X1 and X2 represent the number of the undetermined sub-topics and the comparison sub-topics, respectively. Take the undetermined cluster corresponding to the maximum selection value as the target cluster and deliver the advertisement belonging to the target cluster to the user. First, all clusters are arranged into a sequence based on their interest values from highest to lowest, with cluster L1 at the front. The platform samples at fixed time intervals, writing the current L1 interest value into the reference sequence, forming a progressive curve over time. When any Fa in the sequence is detected to be smaller than the earliest F1, and Fb remains lower than F1 at every point within a preset window, the system determines that the attractiveness of L1 is continuously declining and needs to be switched. At this point, all clusters except L1 are marked as undetermined clusters; for each undetermined cluster, the top topic C1' is selected based on its quantity, and all sub-topics of that topic are extracted, called undetermined sub-topics; then, the top topic C1'' of L1 is selected, and the corresponding sub-topics are extracted, called comparison sub-topics. Next, the system calculates three data points: the number of undetermined sub-topics that completely overlap with comparison sub-topics X3, the total number of undetermined sub-topics X1, and the total number of comparison sub-topics X2. Then, X3 is divided by the absolute value of the difference between the number of sub-topics in the two groups to obtain the selection value XZ. The cluster with the highest XZ among the undetermined clusters is identified as the target cluster. Novelty is measured by the difference in scale between the two sets of subtopics, and then the two are combined into a ratio. A larger intersection indicates that the undetermined cluster shares the same core focus with the current cluster, ensuring the content still aligns with user interests; a larger scale difference indicates that the undetermined cluster introduces additional subtopics while maintaining commonalities, providing sufficient new information. Placing the intersection in the numerator and the difference in the denominator is equivalent to compressing both similarity and difference dimensions with a single formula, finding the optimal balance of "familiarity and novelty" within the same measurement system. This ensures that the selected target cluster is neither identical to existing clusters, leading to information redundancy, nor deviates from the user's current interest trajectory, causing thematic jumps. The platform pushes advertising materials belonging to this target cluster to the current user group in a random order, achieving a smooth thematic transition while injecting a moderate sense of novelty without deviating from user interests, suppressing the fatigue effect caused by repeated exposure. It is understood that the process of delivering advertisements belonging to the target cluster to users includes: Randomly deliver advertisements within the target cluster to users; It is worth noting that the process of delivering ads belonging to the target cluster to users includes: The advertisements within the target cluster are randomly delivered to the user.
[0017] An internet-based advertising monitoring system includes: Clustering module: Obtains text data from advertisements, and clusters the advertisements based on the text data to obtain clusters; Analysis module: Acquires user behavior data for browsing individual advertisements, calculates user interest in the advertisements based on the behavior data, and calculates the interest value of the cluster. X j The interest level of the j-th advertisement in the cluster is represented by N, and the total number of advertisements in the cluster is represented by N. Adjustment module: Sort the clusters in descending order according to the size of their corresponding interest values to obtain a cluster ranking, obtain the first cluster L1 in the cluster ranking, periodically obtain the interest value corresponding to the cluster L1, and sort the interest values according to the time axis to obtain a reference ranking; When the reference sorting meets the preset constraints, the advertising delivery strategy is adjusted.
[0018] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.
Claims
1. An Internet-based advertisement monitoring method, characterized by, The method comprises the following steps: Obtaining text data in an advertisement, clustering the advertisement based on the text data to obtain a cluster; acquiring behavior data of a user browsing a single advertisement, calculating an interest degree of the user to the advertisement based on the behavior data, calculating an interest value of the clustering cluster , X j represents an interest degree of a jth advertisement in the clustering cluster, and N represents a total number of advertisements in the clustering cluster Sorting the cluster in descending order according to the size of the corresponding interest value to obtain a cluster order, obtaining the first cluster L1 in the cluster order, periodically obtaining the interest value corresponding to the cluster L1, and sorting the interest value in time sequence to obtain a reference order; When the reference order meets a preset constraint condition, adjusting the delivery strategy of the advertisement.
2. The Internet-based advertisement monitoring method of claim 1, wherein, The process of obtaining the cluster comprises: Collecting the text data in the advertisement, pre-processing the text data to obtain processed text, and performing word segmentation on the processed text to obtain processed words to generate a first set DYJ={A1, A2, …, An}, An represents the nth processed word, and n represents the number of the processed word; Counting the number of times of appearance of the processing word corresponding to each of the advertisements, denoted as a target number, to generate a second set DRJ of the advertisement i i ={B1 i , B2 i , …, Bm i} and Bm i denotes the target number corresponding to the mth processing word in the advertisement i; Setting the number of topics K, taking the first set and the second set as input to train an LDA model until the topic-word distribution and the advertisement-topic distribution converge; Determining the probability of a single topic in each processed word, taking the processed word with a probability greater than a preset value as a subtopic of the topic, and sorting the topics in ascending order according to the size of the number of corresponding subtopics, denoted as a quantity order.
3. The Internet-based advertisement monitoring method of claim 2, wherein, The process of obtaining the cluster further comprises: Obtaining the first topic C1 in the quantity order, obtaining the subtopics of the topic C1, denoted as grouped subtopics, dividing the topics corresponding to the grouped subtopics containing all the grouped subtopics from the topic C1 into a cluster; Removing the topic C1 from the quantity order to obtain a new quantity order, repeating the above steps to obtain a new cluster, and repeating the above steps to obtain all clusters; When there is no topic corresponding to the grouped subtopics containing all the grouped subtopics, the topic C1 is taken as a cluster.
4. The Internet-based advertisement monitoring method of claim 1, wherein, The interest degree of the user to the advertisement is calculated based on the behavior data, which comprises: The behavior data comprises the average duration of browsing the advertisement, the total number of times of clicking the advertisement, and the number of forwarding times, and the interest degree is calculated based on the behavior data and the distance between the superior and inferior solutions.
5. The Internet-based advertisement monitoring method of claim 3, wherein, The process of adjusting the delivery strategy of the advertisement comprises: Setting a constraint condition: F a <F1; F b <F1, b e (a, a + M] F a interest value of the a-th position in the reference ranking, and M is a preset value. Marking the clusters other than the cluster L1 as pending clusters, obtaining the first topic C1' in the quantity order corresponding to the pending clusters, obtaining the subtopics corresponding to the topic C1', denoted as pending subtopics; Obtaining the first topic C1'' in the quantity order corresponding to the cluster L1, obtaining the subtopics corresponding to the topic C1'', denoted as comparison subtopics; Calculating the selection value XZ of a single pending cluster XZ=X3 / |X1-X2|, X3 represents the number of comparison subtopics in the pending subtopics, X1 and X2 respectively represent the number of the pending subtopics and the comparison subtopics, taking the pending cluster corresponding to the maximum selection value as a target cluster, and delivering the advertisement belonging to the target cluster to the user.
6. The Internet-based advertisement monitoring method of claim 5, wherein, The process of delivering the advertisement belonging to the target cluster to the user comprises: Randomly delivering the advertisement in the target cluster to the user.
7. The Internet-based advertisement monitoring method of claim 5, wherein, In the process of adjusting the delivery strategy of the advertisement, the following steps are further included: When F1 is less than a preset threshold, F1 is removed, a new reference ranking is obtained, and subsequent steps are performed.
8. An Internet-based advertisement monitoring system, characterized by comprising: The method comprises the following steps: a clustering module: obtaining text data in an advertisement, clustering the advertisement based on the text data, and obtaining a clustering cluster; an analysis module: obtaining behavior data of a user browsing a single advertisement, calculating an interest degree of the user to the advertisement based on the behavior data, calculating an interest value of the clustering cluster , X j represents an interest degree of a jth advertisement in the clustering cluster, and N represents a total number of advertisements in the clustering cluster an adjustment module: sorting the clustering cluster in descending order according to the size of the corresponding interest value to obtain a clustering ranking, obtaining a first clustering cluster L1 in the clustering ranking, periodically obtaining the interest value corresponding to the clustering cluster L1, sorting the interest value in a time axis order to obtain a reference ranking; and when the reference ranking meets a preset constraint condition, adjusting the delivery strategy of the advertisement.