Intelligent advertisement recommendation management system based on internet keyword search

Abstract

The invention discloses an intelligent advertisement recommendation management system based on internet keyword search, and relates to the technical field of advertisement putting, the system comprises an information extraction module, a keyword analysis module, a keyword combination module, an advertisement putting module and a real-time adjustment module, the keyword combination module comprises a user intention unit and an overall trend unit, the advertisement putting module puts the first recommended advertisement and the second recommended advertisement to the user according to preset resources, the real-time adjustment module comprises a value analysis unit and an advertisement adjustment unit, and the system has the advantages that short-term and long-term behavior data of the user are combined through the keyword analysis module, outdated content and short-term fluctuation interference are avoided, user requirements are accurately mastered, and user experience is improved. The keyword combination module generates two types of recommended advertisements through double units of user intention and overall trend, personalization and market trend are considered, attraction is enhanced, and the advertisement adjustment module dynamically distributes budget and optimizes resource configuration according to advertisement value scores, so that conversion efficiency is comprehensively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of advertisement recommendation, in particular to an advertisement intelligent recommendation management system based on internet keyword search. BACKGROUND

[0002] Advertisement, as its name implies, is to inform the public of something. Nowadays, with the continuous development of society and economy, people's socialization has gradually developed towards the Internet, and has occupied a large part of people's daily life, leading to the gradual decline of old socialization and entertainment methods. Under this background, a large number of businesses implant advertisements into the Internet, and advertising in the Internet has become the main promotion method of all businesses today. Advertising placement has become an important means for enterprises to promote products and services and attract potential customers.

[0003] At present, advertisement recommendation only relies on basic information and historical browsing behavior for recommendation, which is easy to ignore the specific needs of users in different periods, and the browsing behavior of users may be affected by various factors, such as accidental clicks, recommendations of others, etc., so as to fail to deeply understand the users, leading to the disconnection between recommended content and user needs, low accuracy of advertisement recommendation. At present, advertisement recommendation is usually based on the keywords searched by users, lacking deep analysis of keyword related words and keyword market trends, resulting in one-sided advertisement recommendation content and weak attraction to users, affecting the conversion efficiency. SUMMARY

[0004] The present application aims to provide an advertisement intelligent recommendation management system based on internet keyword search, which solves the problems raised in the background.

[0005] To achieve the above purpose, the present application provides the following technical scheme: an advertisement intelligent recommendation management system based on internet keyword search, comprising an information extraction module, a keyword analysis module, a keyword combination module, an advertisement placement module and a real-time adjustment module.

[0006] The information extraction module is used to collect user behavior data and advertisement promotion data.

[0007] The keyword analysis module analyzes the score of each keyword based on short-term user behavior data and long-term user behavior data, obtains short-term keyword score and long-term keyword score, and adds the weighted scores to obtain the total keyword score. The total keyword score of each keyword is analyzed, and the total keyword score with the highest score is determined as the optimal keyword.

[0008] The keyword combination module includes a user intent unit and an overall trend unit. The user intent unit analyzes the proportion of modifiers that co-occur with the preferred keywords to obtain a user intent score for the modifiers. The overall trend unit analyzes the proportion of modifiers that co-occur with all users under the preferred keywords to obtain a market intent score for the modifiers. Based on the highest values ​​of the user intent score and the market intent score for the modifiers, recommended ad one and recommended ad two are obtained.

[0009] The advertising delivery module delivers recommended ad one and recommended ad two to users based on preset resources;

[0010] The real-time adjustment module includes a value analysis unit and an advertising adjustment unit. The value analysis unit analyzes advertising promotion data to obtain the value scores of recommended ad one and recommended ad two. The advertising adjustment unit dynamically adjusts the advertising budget based on the value scores.

[0011] Optionally, the keyword analysis module first sets short-term and long-term time periods, where the short-term is one week and the long-term is three months, and then obtains user behavior data within one week and three months.

[0012] The short-term keyword i-score (S) is obtained by weighting and aggregating users' keyword search frequency, browsing time, and click behavior over a week. rec,i ;

[0013] Then, the long-term keyword i-score S is obtained by weighting and integrating the user's keyword search frequency, browsing time, and click behavior over the past three months. glo,i ;

[0014] Score the short-term keyword i. rec,i and long-term keyword i score S glo,i Weighted fusion yields the total score S for keyword i. all,i Finally, the total scores of all keywords are compared, and the keyword with the highest score is selected as the preferred keyword Yi.

[0015] Optionally, the user intent unit first obtains the number of times the user simultaneously searches for the preferred keyword Yi and the number of modifiers, and then calculates the ratio of the number of times each modifier appears to the total number of times all modifiers appear, to obtain the user's modifier ratio;

[0016] Further analysis of user purchases reveals the proportion of products containing modifiers and preferred keywords (Yi) to products containing only preferred keywords (Yi), thus revealing the historical purchase impact factor.

[0017] Based on the proportion of modifiers used by users and the influence of historical purchases, a user's intention score for modifiers is obtained. The ad recommended is the one with the highest user intention score for modifiers.

[0018] Optionally, the overall trend unit obtains the number of times all users simultaneously search for the preferred keyword Yi and the number of modifiers, and then calculates the ratio of the number of times each modifier appears to the total number of times all modifiers appear, to obtain the market's modifier proportion;

[0019] Based on the advertising promotion data, the relevant product promotion information including modifiers is analyzed to obtain the product popularity score including modifiers;

[0020] The market's intention score for modifiers is obtained based on the proportion of modifiers in the market and the popularity score of products including modifiers. The second recommended advertisement is based on the highest market intention score for modifiers.

[0021] Optionally, the process of scoring product popularity including modifiers is as follows:

[0022] First, obtain the percentage of sales of products including modifiers from the advertising data to the total sales volume, thus obtaining the sales percentage.

[0023] The percentage of product views including modifiers is then used to calculate the product popularity score, which is obtained by weighting and combining the sales percentage and the page view percentage.

[0024] Optionally, the value analysis unit analyzes the value scores of recommended advertisement one and recommended advertisement two, as follows:

[0025]

[0026] In the above formula, E k The value score of the recommended advertisement is k, where k ranges from 1 to 2. When k is 1, E1 is the value score of the first recommended advertisement, and when k is 2, E2 is the value score of the second recommended advertisement.

[0027] ROI k The return on investment for recommended ad k;

[0028] ROI ref,k The target return on investment for recommended ad k;

[0029] θ1 is the investment return rate impact coefficient, with a value of 0.5;

[0030] HR avg,k The average return on investment for historical recommended ads k;

[0031] HR avg,ref,k The average target return on investment for historical recommended ads k;

[0032] θ2 is the historical investment return rate influence coefficient, with a value of 0.2.

[0033] CF kThe interaction impact score reflects the attractiveness of the advertisement to users.

[0034] θ3 is the interaction influence coefficient, with a value of 0.3;

[0035] Interactive Influence Score CF k The calculation process is as follows:

[0036]

[0037] In the above formula, AT k The average dwell time for recommended ad k;

[0038] AT avg,k The average dwell time of all users for the recommended ad;

[0039] This represents the percentage of time spent in the room.

[0040] SC k The number of times ad k is shared;

[0041] C k The number of clicks for recommended ad k;

[0042] Substituting the placement data of Recommended Ad 1 and Recommended Ad 2 into the above formula, we finally obtain the value score E1 of Recommended Ad 1 and the value score E2 of Recommended Ad 2.

[0043] Optionally, after obtaining the value score E1 of recommended ad one and the value score E2 of recommended ad two, the advertising adjustment unit dynamically adjusts the advertising budget according to the value scores. The adjustment process is as follows:

[0044]

[0045] P new,k The adjusted recommended ad budget;

[0046] P old,k The initial budget for recommended ad k;

[0047] ΔP k To adjust the budget for recommended ad k by a percentage, positive values ​​increase the budget, while negative values ​​decrease it. To avoid frequent minor adjustments, a threshold ΔP is set for the budget adjustment percentage of recommended ad k. k Adjustment operation is initiated when the value is ≥0.1;

[0048] Recommended ad budget adjustment ratio ΔP k The process of obtaining the result is as follows:

[0049]

[0050] λ is the adjustment coefficient, with a value ranging from 0.2 to 0.5;

[0051] E1 represents the value score of the first recommended advertisement, and E2 represents the value score of the second recommended advertisement.

[0052] Optionally, when collecting data, the information extraction module identifies users whose average value for any user behavior indicator is more than 10 times the average value of that indicator for all users as abnormal users, and determines abnormal user data as invalid data, thereby reducing interference from extreme users and malicious programs on the system.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] I. This invention analyzes the score of each keyword based on short-term and long-term user behavior data through a keyword analysis module, obtaining short-term keyword scores and long-term keyword scores. By combining short-term and long-term user behavior data, including browsing time and click counts, it can conduct in-depth analysis of users' actual needs, avoid recommending outdated content, and prevent short-term fluctuations such as accidental clicks and recommendations from others from interfering with advertising content. It fully considers users' needs at different times, avoids the disconnect between recommended content and user needs, and improves the accuracy of advertising recommendations.

[0055] Second, the keyword combination module of this invention determines two types of recommended advertisements from both the user and market perspectives through user intention units and overall trend units. It fully considers keyword-related words and keyword market trends, recommending personalized content to users while providing the current overall market trend as a reference, making the recommended advertisement content more comprehensive, enhancing its appeal to users, and improving conversion efficiency.

[0056] Third, this invention, through its advertising adjustment unit, can assess the user appeal of personalized advertising content and overall market trends based on the value scores of recommended ad one and recommended ad two, and dynamically adjust the budget according to the actual promotion effect. This avoids investing too many resources in low-value ads or insufficient resources in high-value ads, thereby achieving reasonable resource allocation and improving advertising conversion efficiency. Attached Figure Description

[0057] Figure 1 This is a block diagram of the system modules of the present invention;

[0058] Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0059] 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.

[0060] For examples, please refer to Figure 1 and Figure 2 This implementation provides an intelligent advertising recommendation management system based on Internet keyword search, including an information extraction module, a keyword analysis module, a keyword combination module, an advertising delivery module, and a real-time adjustment module;

[0061] The information extraction module is used to collect user behavior data and advertising promotion data;

[0062] When collecting data, the information extraction module identifies users whose average behavior data is more than 10 times greater than the average behavior data of all users as abnormal users. This reduces interference from extreme users and malicious programs on the system. For example, if the average browsing time of all users is 10 seconds, that is, the average time that all users spend browsing each product, when the system detects that a user has an average browsing time of 100 seconds per product, the user is identified as an abnormal user, and the abnormal user data is determined to be invalid data, thus reducing interference from extreme users and malicious programs on the system.

[0063] The keyword analysis module analyzes the score of each keyword based on short-term and long-term user behavior data to obtain short-term keyword scores and long-term keyword scores. These scores are then weighted and summed to obtain the total keyword score. The total score of each keyword is analyzed, and the keyword with the highest total score is determined as the preferred keyword.

[0064] By combining short-term and long-term user behavior data, including browsing time and click counts, we can conduct in-depth analysis of users' actual needs, avoid recommending outdated content, and prevent short-term fluctuations such as accidental clicks and recommendations from others from interfering with ad delivery. This fully considers users' needs at different times, avoids recommending content that is out of touch with user needs, and improves the accuracy of ad recommendations.

[0065] The keyword combination module includes a user intent unit and an overall trend unit. The user intent unit analyzes the proportion of modifiers that co-occur with the preferred keywords to obtain the user's intent score for the modifiers. The overall trend unit analyzes the proportion of modifiers that co-occur with all users under the preferred keywords to obtain the market's intent score for the modifiers. Based on the highest values ​​of the user's intent score for the modifiers and the market's intent score for the modifiers, recommended ad one and recommended ad two are obtained.

[0066] The advertising delivery module delivers recommended ad one and recommended ad two to users based on preset resources, with recommended ad one and recommended ad two having the same initial budget;

[0067] The keyword combination module analyzes the modifiers that co-occur with the preferred keywords and determines recommended ads from both the user and market perspectives. It fully considers keyword-related terms and keyword market trends, recommending personalized content to users while providing a reference for the overall current market trends. This makes the recommended ad content more comprehensive, enhances its appeal to users, and improves conversion efficiency.

[0068] The real-time adjustment module includes a value analysis unit and an advertising adjustment unit. The value analysis unit analyzes advertising promotion data to obtain the value scores of recommended ad one and recommended ad two. The advertising adjustment unit dynamically adjusts the advertising budget based on the value scores.

[0069] The real-time adjustment module evaluates the attractiveness of personalized ad content and overall market trends to users based on the actual promotion effect of the two ads, and adjusts resources accordingly. This allocates resources to high-value ads, reduces the waste of resources on low-value ads, and improves the overall conversion efficiency of the ads.

[0070] Furthermore, the keyword analysis module analyzes the score of each keyword based on short-term and long-term user behavior data, obtaining short-term keyword scores and long-term keyword scores. These scores are then weighted and summed to obtain the total keyword score. The total score of each keyword is analyzed, and the keyword with the highest total score is determined as the preferred keyword. The calculation process is as follows:

[0071]

[0072] In the above formula, S all,i The total score for keyword i is given, and the closer the result is to 1, the stronger the correlation between user behavior data and that keyword.

[0073] β is the short-term score impact coefficient, with a value of 0.7, where the short-term period is set to one week;

[0074] (1-β) Long-term score impact coefficient, with a long-term duration of at least three months;

[0075] S rec,i The i-score for short-term keywords reflects the user's recent interests;

[0076] S glo,iThe long-term keyword score reflects the user's long-term tendencies. By adding the user's short-term and long-term keyword scores, the timeliness changes of user interests and the overall historical trend can be fully considered, avoiding interference from short-term accidental searches on the accuracy of ad recommendations. At the same time, in order to ensure the real-time nature of the ads, the influence coefficient of the recent score β is set to be greater than the influence coefficient of the long-term score (1-β), avoiding the recommendation of outdated content.

[0077] Among them, the short-term keyword i score S rec,i The calculation process is as follows:

[0078]

[0079] In the above formula, A rec,i The number of searches for keyword i within a short period of time;

[0080] A rec,max The maximum number of searches among all keywords in the short term;

[0081] W1 is the influence coefficient of short-term search frequency, with a value of 0.4. This is a short-term search frequency item. The higher the short-term search frequency, the more urgent the demand, directly reflecting the user's immediate purchase intention. By assigning a higher weight to this item, we can ensure that ad recommendations can quickly respond to users' immediate needs, improving the timeliness of recommendations and conversion rates.

[0082] T rec,i The browsing time of products related to keyword i in a short period of time is recorded. Each product contains at least one keyword as a product tag. The browsing time of products related to each keyword is recorded when the user enters the keyword to search.

[0083] T rec,max The longest browsing time for related products among all keywords in the short term;

[0084] W2 is the short-term browsing time impact coefficient, with a value of 0.3. This is a short-term browsing time item. The higher the value, the higher the user's attention to the product. By introducing this item, we can more accurately measure the intensity of user interest, avoid recommending products that have only been searched but not deeply viewed, and improve the accuracy of recommended ads.

[0085] C rec,i The number of clicks on products related to keyword i within a short period of time;

[0086] C rec,max This represents the maximum number of clicks for related products among all keywords in the short term.

[0087] W3 is the short-term click impact factor. This is a short-term click count item. Clicks are a deeper behavior than searches, indicating that users are willing to learn more about the product content. This item can take into account users' actual operational preferences, reflect users' potential purchase needs, and improve advertising conversion efficiency.

[0088] By weighting and summing short-term search counts, short-term browsing time, and short-term click counts, a comprehensive analysis of users' short-term interests can be achieved, and the short-term keyword i-score S rec,i The higher weighting ensures that ad recommendations can quickly respond to users' recent needs and improve conversion rates.

[0089] Long-term keyword i score S glo,i The calculation process is as follows:

[0090]

[0091] A glo,i The number of searches for keyword i over a long period of time;

[0092] A glo,max This represents the maximum number of searches for all keywords over a long period.

[0093] W4 is the influence coefficient of long-term search frequency, with a value of 0.4;

[0094] T glo,i This refers to the browsing time of products related to keyword i over a long period of time.

[0095] T glo,max The longest browsing time for related products among all keywords over a long period;

[0096] W5 is the long-term browsing time impact factor, with a value of 0.3;

[0097] C glo,i The number of clicks on keyword i over a long period of time;

[0098] C glo,max This represents the maximum number of clicks for related products across all keywords over a long period.

[0099] W6 is the long-term click impact coefficient, with a value of 0.3;

[0100] By incorporating long-term keyword i-score S glo,i The system takes into full account users' long-term historical behavior in order to identify users' stable interests. For example, if a user has a long-term preference for running shoes but recently searched for casual shoes, but the browsing time and number of clicks are low, the short-term keyword i-score will be low. rec,i The overall score after combination is low, with sports shoes still dominating. Even if short-term demand fluctuates, the system will still retain relevant content in the recommendations to avoid recommendation deviations caused by accidental search behavior and ensure the accuracy of ad placement.

[0101] Specifically, the total score of keyword i is S all,i By combining short-term and long-term user behavior data, including browsing time and click counts, we can conduct in-depth analysis of users' actual needs. This avoids recommending outdated content and prevents short-term, accidental keyword searches from interfering with ad delivery. It fully considers users' needs at different times, improves the accuracy of ad recommendations, and finally selects the keyword with the highest score as the preferred keyword Yi by comparing the total scores of all keywords.

[0102] For example, if a user frequently searches for down jackets in a short period of time, but frequently searched for dresses several months ago, the recent score influence coefficient β will have a high weight, causing the score of the keyword "down jacket" to far exceed that of "dress". The system will prioritize recommending down jacket ads, rather than dresses that the user has long since stopped paying attention to, thus ensuring the timeliness of ad recommendations.

[0103] For example, if a user frequently searches for athletic shoes but recently searches for casual shoes a few times based on a friend's recommendation, and if the user is interested in casual shoes, then the browsing time and clicks on related products for that keyword will increase, improving the short-term keyword score (S). rec,i The higher the score, the greater the probability that the system will recommend casual shoe ads; conversely, if the user is not interested, the short-term keyword score will be lower. rec,i The system's recommendations remain primarily focused on athletic shoe ads to avoid interference from short-term, accidental searches and improve the accuracy of ad recommendations.

[0104] Furthermore, after obtaining the preferred keyword Yi, the proportion of modifiers co-occurring with the preferred keyword Yi is analyzed through the user intent unit in the keyword combination module, which serves as user intent. Then, the overall trend unit combines the proportion of modifiers co-occurring with all users under this preferred keyword to serve as the overall trend, reflecting the current trend of the product. Finally, based on the highest score of user intent and the highest score of the overall trend, recommended ad one and recommended ad two are obtained respectively. The user intent analysis process is as follows:

[0105]

[0106] In the above formula, U(Mj) is the user's intention score for modifier j, which ranges from 0 to 1. The higher the value, the more interested the user is in the modifier. Modifiers are the search terms that accompany keywords. For example, if the keyword is shoes, the modifiers could be anti-slip, breathable, sports, etc.

[0107] M represents the set of candidate modifiers, M = {M1, M2, M3...Mn}, which represents all modifiers related to the preferred keyword Yi;

[0108] U co(Mj,Yi) represents the co-occurrence frequency of modifier j and preferred keyword Yi in user searches. For example, if modifier j is "sports" and preferred keyword Yi is "men's shoes", then U co (Mj,Yi) represents the number of times a user simultaneously searches for "men's shoes, sports";

[0109] γ1 is the historical purchase influence coefficient, with a value of 0.2, which reflects the degree of influence of historical purchase behavior on user intention score;

[0110] P(Mj,Yi) represents the proportion of products containing modifier j and preferred keyword Yi among the products purchased by the user, out of the total number of products containing preferred keyword Yi.

[0111] [1+γ1×P(Mj,Yi)] represents the historical purchase impact factor. By combining users' historical purchase data, we can avoid the one-sidedness of relying solely on search co-occurrence. For example, when users search for "men's shoes," it frequently co-occurs with "sports," but most of their historical purchases of men's shoes are "business" styles. In this case, the historical purchase impact factor [1+γ1×(P(Mj,Yi)] will increase significantly, ultimately affecting the user's intention score for the modifier "business," making the user's intention closer to their actual consumption preferences and improving advertising conversion efficiency.

[0112] By analyzing users' intention scores for all modifiers, if a user's intention score U(Mj) for modifier j is the highest among all modifiers, then the user's preferred combination tag is UYi(j). Subsequently, relevant advertisements can be recommended to users based on the user's preferred combination tag UYi(j). Each advertisement product should include the preferred keyword Yi and modifier j. In practice, multiple modifiers can be selected from high to low scores and then combined with the preferred keyword Yi to further improve the accuracy of advertisement recommendations and increase conversion rates.

[0113] After obtaining the user's intention score U(Mj) for modifier j, the overall user interest trend is analyzed, as follows:

[0114]

[0115] In the above formula, G(Mj) is the market intention score for modifier j, which ranges from 0 to 1. The higher the value, the more interested the overall market is in the modifier, reflecting the current market trend of related products.

[0116] G co (Mj,Yi) represents the co-occurrence frequency of modifier j and preferred keyword Yi in all user searches. For example, if modifier j is "sports" and preferred keyword Yi is "men's shoes", then G co (Mj,Yi) represents the number of times a user simultaneously searches for "men's shoes, sports";

[0117] G totThe number of modifiers that co-occur with the preferred keyword Yi in all user searches;

[0118] γ2 is the influence coefficient of product popularity, with a value of 0.2, reflecting the degree of influence of product popularity on the overall trend;

[0119] H(Mj) is the product popularity score including modifier j, and the calculation process is as follows;

[0120]

[0121] In the formula, D j For products containing the modifier 'j', short-term sales volume;

[0122] D tot (Yi) represents the short-term sales volume of all products related to the preferred keyword Yi;

[0123] γ3 is the coefficient affecting product sales, with a value of 0.5. As a percentage of sales, it reflects users' actual purchasing behavior; the higher the percentage, the stronger the user recognition of the product.

[0124] V j For products containing the modifier "j", the number of short-term pageviews is calculated.

[0125] V tot (Yi) represents the short-term page views of all products related to the preferred keyword Yi;

[0126] γ4 is the influence coefficient of product pageviews, with a value of 0.5. View count percentage is a key factor in the pre-purchase process. Higher view counts indicate a stronger attraction of the product to users, increasing the probability of a purchase. This is measured by sales percentage. Percentage of page views The weighted combination reflects both the actual acceptance of the product and the level of attention paid to the product before purchase.

[0127] [1 + γ2 × H(Mj)] represents the product popularity item, allowing the overall trend to consider not only search co-occurrence but also the actual market popularity of the product, recommending products that are more in line with current trends and popularity to users. For example, among the modifiers for "men's shoes", the search co-occurrence of "breathable" is not the highest, but recently the sales and page views of men's shoes with "breathable" are significantly higher than other modifiers, and the total score will also increase accordingly, making the trend analysis closer to the real consumption dynamics;

[0128] After obtaining the market's intention score G(Mj) for modifier j, compare the intention scores of all users for different modifiers, and use the highest score as the market's preferred combination label.

[0129] The ad corresponding to the user's preferred combination tag is recommended ad one, and the ad corresponding to the market preferred combination tag is recommended ad two. Recommended ad one and recommended ad two are delivered to users through the ad delivery module, and the initial budget for recommended ad one and recommended ad two is equal.

[0130] Furthermore, after Recommended Ad 1 and Recommended Ad 2 are launched, the conversion performance of the two ads is analyzed periodically through the real-time adjustment module to obtain the value scores of Recommended Ad 1 and Recommended Ad 2. Then, the budgets for the two ads are adjusted based on the value scores. The specific process is as follows:

[0131]

[0132] In the above formula, E k The value score of the recommended advertisement is k, where k ranges from 1 to 2. When k is 1, E1 is the value score of the first recommended advertisement, and when k is 2, E2 is the value score of the second recommended advertisement.

[0133] ROI k The return on investment for recommended ad k;

[0134] ROI ref,k The target return on investment for recommended ad k;

[0135] θ1 is the investment return rate impact coefficient, with a value of 0.5. As the return on investment item and the main source of revenue from advertising, it is given the highest weight.

[0136] HR avg,k The average return on investment for historical recommended ads k;

[0137] HR avg,ref,k The average target return on investment for historical recommended ads k;

[0138] θ2 is the historical investment return rate influence coefficient, with a value of 0.2. This is a historical return item, reflecting the user's historical ad conversion efficiency. For example, if the user's ad conversion efficiency is low recently, but high in the long term, then the historical return item... A larger value increases the k-value score (E) of the recommended ad. k This helps avoid reducing promotion due to short-term fluctuations in user conversion rates, which could lead to decreased long-term revenue and improve the stability of ad conversion rates.

[0139] CF k The interaction impact score reflects the attractiveness of the advertisement to users.

[0140] θ3 is the interaction influence coefficient, with a value of 0.2;

[0141] Interactive Influence Score CF k The calculation process is as follows:

[0142]

[0143] In the above formula, AT k The average dwell time for recommended ad k;

[0144] AT avg,k The average dwell time of all users for the recommended ad;

[0145] The dwell time percentage reflects the time users actively invest. The longer the dwell time, the more attractive the ad content is to the user. To avoid interference from low-quality content, it is important to note that clicks alone cannot determine whether users are truly paying attention to the content. For example, if a user clicks on an ad and immediately exits, the average dwell time of all users on recommended ads is used as a benchmark to avoid natural differences in dwell time caused by ad type and industry characteristics, and to ensure comparability between different ads.

[0146] SC k The number of times ad k is shared;

[0147] C k The number of clicks for recommended ad k;

[0148] The average share rate is used to measure the average share rate. Sharing is the act of users actively recommending ads to others, which expands the reach of the ads. Furthermore, recommendations from friends are more likely to convert into purchases and are more valuable than clicks. Percentage of stay Add them together to get the interaction impact score (CF). k If only dwell time is used, creative ideas that are "short, quick, and effective" but have strong dissemination power, such as viral short videos, may be overlooked; if only share rate is used, high-quality creative ideas with sufficient content depth but weak dissemination power, such as professional product reviews, may be missed. Combining the two can cover different types of high-quality creative ideas and more comprehensively reflect the overall value of advertising.

[0149] The real-time adjustment module first evaluates the value of recommended ads by combining return on investment, historical return on investment, and interaction impact data, which represent short-term benefit, long-term benefit, and user interaction dimensions, respectively. By weighting and integrating the multi-dimensional data, the value assessment of ads can be more comprehensive and accurate, so as to make precise adjustments to recommended ad one or recommended ad two in the future.

[0150] After substituting the promotion data of Recommended Ad 1 and Recommended Ad 2 into the above formula to obtain the value score E1 of Recommended Ad 1 and the value score E2 of Recommended Ad 2, the budget is adjusted through the ad adjustment unit. The specific process is as follows:

[0151]

[0152] P new,k The adjusted recommended ad budget;

[0153] P old,k The initial budget for recommended ad k;

[0154] ΔP k The recommended ad budget adjustment ratio k is set as follows: a positive number indicates an increase in the budget, and a negative number indicates a decrease in the budget. To avoid frequent minor adjustments, a value ΔP is set for the recommended ad budget adjustment ratio k. k The adjustment operation is only initiated when the value is ≥0.1;

[0155] Recommended ad budget adjustment ratio ΔP k The process of obtaining the result is as follows:

[0156]

[0157] λ is the adjustment coefficient, with a value ranging from 0.2 to 0.5;

[0158] E1 represents the value score of the first recommended ad, and E2 represents the value score of the second recommended ad.

[0159] The ad adjustment unit allows for dynamic budget adjustments based on the relative merits of Recommended Ad 1 (value score E1) and Recommended Ad 2 (value score E2), preventing excessive resource allocation for low-value ads or insufficient resources for high-value ads, thus achieving rational resource allocation and improving ad conversion efficiency.

[0160] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent advertising recommendation management system based on internet keyword search, characterized in that: It includes an information extraction module, a keyword analysis module, a keyword combination module, an ad placement module, and a real-time adjustment module; The information extraction module is used to collect user behavior data and advertising promotion data; The keyword analysis module analyzes the score of each keyword based on short-term and long-term user behavior data to obtain short-term keyword scores and long-term keyword scores. The short-term keyword scores and long-term keyword scores are then weighted and summed to obtain the total keyword score. The total score of each keyword is analyzed, and the keyword with the highest total score is determined as the preferred keyword. The keyword combination module includes a user intent unit and an overall trend unit. The user intent unit analyzes the proportion of modifiers that co-occur with the preferred keywords to obtain a user intent score for the modifiers. The overall trend unit analyzes the proportion of modifiers that co-occur with all users under the preferred keywords to obtain a market intent score for the modifiers. Based on the highest values ​​of the user intent score and the market intent score for the modifiers, recommended ad one and recommended ad two are obtained. The advertising delivery module delivers recommended ad one and recommended ad two to users based on preset resources; The real-time adjustment module includes a value analysis unit and an advertising adjustment unit. The value analysis unit analyzes advertising promotion data to obtain the value scores of recommended ad one and recommended ad two. The advertising adjustment unit dynamically adjusts the advertising budget based on the value scores.

2. The intelligent advertising recommendation management system based on Internet keyword search according to claim 1, characterized in that: The keyword analysis module first sets short-term and long-term time periods, with the short-term being one week and the long-term being three months, and then obtains user behavior data within one week and three months. The short-term keyword i-score (S) is obtained by weighting and aggregating users' keyword search frequency, browsing time, and click behavior over a week. rec,i ; Then, the long-term keyword i-score S is obtained by weighting and integrating the user's keyword search frequency, browsing time, and click behavior over the past three months. glo,i ; Score the short-term keyword i. rec,i and long-term keyword i score S glo,i Weighted fusion yields the total score S for keyword i. all,i Finally, the total scores of all keywords are compared, and the keyword with the highest score is selected as the preferred keyword Yi.

3. The intelligent advertising recommendation management system based on Internet keyword search according to claim 2, characterized in that: The user intent unit first obtains the number of times the user simultaneously searches for the preferred keyword Yi and the number of modifiers, and then calculates the ratio of the number of times each modifier appears to the total number of times all modifiers appear, thus obtaining the user's modifier percentage. Further analysis of user purchases reveals the proportion of products containing modifiers and preferred keywords (Yi) to products containing only preferred keywords (Yi), thus revealing the historical purchase impact factor. Based on the proportion of modifiers used by users and the influence of historical purchases, a user's intention score for modifiers is obtained. The ad recommended is the one with the highest user intention score for modifiers.

4. The intelligent advertising recommendation management system based on Internet keyword search according to claim 3, characterized in that: The overall trend unit obtains the number of times all users simultaneously search for the preferred keyword Yi and the number of modifiers, and then calculates the ratio of the number of times each modifier appears to the total number of times all modifiers appear, to obtain the market's modifier proportion; Based on the advertising promotion data, the relevant product promotion information including modifiers is analyzed to obtain the product popularity score including modifiers; The market's intention score for modifiers is obtained based on the proportion of modifiers in the market and the popularity score of products including modifiers. The second recommended advertisement is based on the highest market intention score for modifiers.

5. The intelligent advertising recommendation management system based on Internet keyword search according to claim 4, characterized in that: The product popularity scoring process, which includes modifiers, is as follows: First, obtain the percentage of sales of products including modifiers from the advertising data to the total sales volume, thus obtaining the sales percentage. The percentage of product views including modifiers is then used to calculate the product popularity score, which is obtained by weighting and combining the sales percentage and the page view percentage.

6. The intelligent advertising recommendation management system based on Internet keyword search according to claim 5, characterized in that: The value analysis unit analyzes the value scores of recommended advertisement one and recommended advertisement two, as follows: In the above formula, E k The value score of the recommended advertisement is k, where k ranges from 1 to 2. When k is 1, E1 is the value score of the first recommended advertisement, and when k is 2, E2 is the value score of the second recommended advertisement. ROI k The return on investment for recommended ad k; ROI ref,k The target return on investment for recommended ad k; θ1 is the investment return rate impact coefficient, with a value of 0.5; HR avg,k The average return on investment for historical recommended ads k; HR avg,ref,k The average target return on investment for historical recommended ads k; θ2 is the historical investment return rate influence coefficient, with a value of 0.

2. CF k The interaction impact score reflects the attractiveness of the advertisement to users. θ3 is the interaction influence coefficient, with a value of 0.3; Interactive Influence Score CF k The calculation process is as follows: In the above formula, AT k The average dwell time for recommended ad k; AT avg,k The average dwell time of all users for the recommended ad; This represents the percentage of time spent in the room. SC k The number of times ad k is shared; C k The number of clicks for recommended ad k; Substituting the placement data of Recommended Ad 1 and Recommended Ad 2 into the above formula, we finally obtain the value score E1 of Recommended Ad 1 and the value score E2 of Recommended Ad 2.

7. The intelligent advertising recommendation management system based on Internet keyword search according to claim 6, characterized in that: After obtaining the value score E1 of recommended ad one and the value score E2 of recommended ad two, the advertising budget is dynamically adjusted by the advertising adjustment unit according to the value scores. The adjustment process is as follows: P new,k The adjusted recommended ad budget; P old,k The initial budget for recommended ad k; ΔP k To adjust the budget for recommended ad k by a percentage, positive values ​​increase the budget, while negative values ​​decrease it. To avoid frequent minor adjustments, a threshold ΔP is set for the budget adjustment percentage of recommended ad k. k Adjustment operation is initiated when the value is ≥0.1; Recommended ad budget adjustment ratio ΔP k The process of obtaining the result is as follows: λ is the adjustment coefficient, with a value ranging from 0.2 to 0.5; E1 represents the value score of the first recommended advertisement, and E2 represents the value score of the second recommended advertisement.

8. The intelligent advertising recommendation management system based on Internet keyword search according to claim 1, characterized in that: When collecting data, the information extraction module identifies users whose average value for any behavioral indicator is more than 10 times the average value of that indicator for all users as abnormal users. Abnormal user data is then deemed invalid, thereby reducing interference from extreme users and malicious programs on the system.

Images