Advertisement sorting method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of advertisement sorting methods, in particular to an advertisement sorting method and device, electronic equipment and a storage medium, and the method specifically comprises the following steps: 1, collecting and preprocessing advertisement basic information: collecting the basic information of to-be-sorted advertisements, including advertisement householder information, advertisement content information and advertisement putting information, then, the collected information is preprocessed; 2, user portrait and interest dimension calculation: constructing user portraits based on historical behavior data of platform users, including user basic attributes, short-term interest tags, long-term interest tags and consumption ability levels, and then calculating the matching degree of advertisements and user interests; according to the invention, the problem of dimension simplification in the prior art is solved, and three core dimensions of user interest, advertisement value and user experience are constructed.

Description

Technical Field

[0001] This invention relates to the technical field of advertising sorting methods, and in particular to an advertising sorting method, apparatus, electronic device, and storage medium. Background Technology

[0002] In the current digital advertising landscape, ad ranking is a core element in balancing advertisers' marketing goals with user experience. Existing ad ranking technologies mainly fall into two categories: one is the "Cost Per Click (CPC) × Quality Score" ranking model, exemplified by Google AdWords. This model focuses solely on the ad's immediate conversion potential and basic quality, neglecting the match between the ad and the user's long-term interests, as well as the long-term impact of ad content on the platform ecosystem. The other is collaborative filtering ranking based on users' historical click data. This method relies excessively on historical data and is extremely unsuitable for newly launched ads (cold start ads) or scenarios with sudden changes in user interests, easily leading to a "Matthew effect," where popular ads continue to receive high exposure, while high-quality new ads struggle to get a chance to be displayed.

[0003] Meanwhile, existing technologies generally suffer from the problem of fixed weights. For example, during major promotional periods and daily operations on e-commerce platforms, the core indicators for ad ranking (such as conversion priority and user experience weight) should differ significantly, but most existing methods use fixed weight calculations, which cannot dynamically adapt to the needs of different scenarios. Summary of the Invention

[0004] In response to the technical problems mentioned in the background art, the present invention provides a method, apparatus, electronic device, and storage medium for sorting advertisements.

[0005] The technical solution adopted in this invention is: a method for sorting advertisements, specifically including the following steps:

[0006] Step 1: Collection and preprocessing of basic advertising information. Collect basic information of the advertisements to be sorted, including advertiser information, advertisement content information, and advertisement placement information. Then, preprocess the collected information.

[0007] Step 2: User profile and interest dimension calculation. Based on the platform users' historical behavior data, a user profile is constructed, including basic user attributes, short-term interest tags, long-term interest tags, and spending power level. Then, the matching degree between advertisements and user interests is calculated.

[0008] Step 3: Calculate the advertising value dimensions. The advertising value dimensions include immediate conversion value and long-term ecosystem value, which together constitute the comprehensive advertising value score.

[0009] Step 4: User experience dimension calculation. The user experience dimension comprehensively considers the impact of advertisements on the user browsing experience, including the frequency of ad display, the suitability of ad format and the suitability of user privacy protection, and finally obtains the comprehensive user experience score.

[0010] Step 5: Based on the above steps, calculate the multi-dimensional weighted comprehensive score;

[0011] Step 6: Conduct compliance review and anomaly filtering of advertisements;

[0012] Step 7: Sort the ads that have passed the filtering in Step 6 from highest to lowest to obtain the initial ad sorting list.

[0013] In one embodiment, in step 1, the advertiser information includes advertiser qualification level A. q Historical number of violations (V) n The advertising content information includes the advertising type T. a Advertising content compliance score: C s Advertising placement information includes the target audience G. t Preset bid B p ;

[0014] Advertiser qualification level A q Quantification:

[0015] The qualification levels are divided into four levels: “Excellent”, “Good”, “Average”, and “Poor”, which correspond to quantitative values ​​of 4, 3, 2, and 1, respectively.

[0016] V number of violations in historical ad placements n Normalization is performed using the formula: Where V max V represents the maximum number of violations committed by a single advertiser in the platform's historical records. norm V is the normalized violation coefficient, ranging from 0 to 1. The more violations, the higher the V value. norm The larger the value, the stronger the negative impact on ad ranking.

[0017] Advertising content compliance score: C s Standardization process: using formulas Where C min =0,C max =10,C std This is the standardized compliance coefficient, ranging from 0 to 1. A higher compliance score results in a higher C. std The larger.

[0018] In one embodiment, in step 2, the user's basic attributes include age A. g Gender G d , region R gShort-term interest tags include product / service categories frequently viewed in the past 7 days (L7); long-term interest tags include product / service categories consistently viewed in the past 90 days (L). 90 Consumption capacity level P c ;

[0019] The degree of match between advertisements and user interests is calculated, and it is divided into short-term interest match degree M. s Match with long-term interests M l The specific formula is as follows: Short-term interest matching degree: Where N s-match N represents the number of matches between the ad category and the user's high-frequency interest tags over the past 7 days. s-total M represents the total number of high-frequency interest tags used by users in the past 7 days. s The value ranges from 0 to 1; the more matches, the higher the value of M. s The larger;

[0020] Long-term interest matching degree: Where N l-match N represents the number of matches between the ad category and the user's long-term interest tags over the past 90 days. l-total The total number of long-term interest tags for users over the past 90 days; P c-match M is the matching coefficient between the target consumer group of the advertisement and the user's spending power level. A match is assigned a value of 1, and a mismatch is assigned a value of 0. l The value range is 0-1;

[0021] Overall score for interest dimension;

[0022] User Interest Dimension Overall Score I u The calculation formula is: I u =α×M s +(1-α)×M l , where α is the short-term interest weighting coefficient, with a value ranging from 0 to 1.

[0023] In one embodiment, in step 3, the overall advertising value score is V. a ;

[0024] include:

[0025] Instant conversion value calculation;

[0026] Instant conversion value V min The formula for calculating V, based on the advertiser's preset bid and historical conversion probability, is: min =B p ×P cv B p Preset click bids for advertisers, P cv This represents the historical click-through rate of the ad, with a value ranging from 0 to 1.

[0027] For newly launched cold start ads, P cv Using the average conversion rate P of similar advertisements cvavg Substitution, i.e., P cv =P cvavg ;

[0028] Long-term ecological value calculation;

[0029] Long-term ecological value V lo To assess the long-term impact of advertising on the platform ecosystem, including advertiser qualifications, historical violations, and content compliance, the formula is: V lo =A q ×(1-V norm )×C std , where A q V is a quantitative value for the qualification level of advertisers. norm C is the normalized violation coefficient. std The standardized compliance coefficient;

[0030] Overall advertising value score;

[0031] Overall Advertising Value Score V a The calculation formula is: V a =β×V im +(1-β)×V lo , where β is the weighting coefficient for instant conversion value, ranging from 0 to 1, and is dynamically adjusted according to the platform's operational scenarios.

[0032] In one embodiment, in step 4, the overall user experience score is E. u ;

[0033] include:

[0034] Score for ad display frequency control;

[0035] Ad display frequency control score F c The formula is: Where F current F represents the number of times the ad has been shown to the current user today. max The maximum number of times a single advertisement can be displayed per day as set for the platform;

[0036] When F current ≤F max Time F c Values ​​range from 0 to 1; the more times the image is displayed, the higher the F value. c The smaller; when F c urrent>F max At that time, F c =0, the advertisement will no longer be shown to this user on that day to avoid excessively disturbing the user;

[0037] Ad format suitability score;

[0038] Ad format suitability F a The formula for considering the suitability of the ad format to the user's current browsing context is: Where n is the number of adaptation dimensions, W i M represents the weights for each dimension. i These are the matching coefficients for each dimension;

[0039] User privacy protection adaptation score;

[0040] User privacy protection compatibility P p The formula for determining whether ad placement aligns with users' privacy preferences is:

[0041]

[0042] Overall user experience score;

[0043] Overall user experience score E u The calculation formula is: E u =γ×F c +δ×F a +η×P p , where γ, δ, and η are the weights of each dimension, and γ+δ+η=1.

[0044] In one embodiment, in step 5, the multi-dimensional weighted comprehensive score is the core basis for ad ranking, and its calculation formula is: S total =ω1×I u +ω2×V a +ω3×E u ,in:

[0045] S total The overall score for the advertisement has a range of values:

[0046] O-(ω1×1+ω2×(B pmax ×1+4×1×1)+ω3×1);

[0047] ω1, ω2, and ω3 are the weights of the user interest dimension, advertising value dimension, and user experience dimension, respectively, and ω1+ω2+ω3=1.

[0048] In one embodiment, step 6: after calculating S total Afterwards, compliance reviews and anomaly filtering are required for advertisements to prevent illegal or abnormal advertisements from appearing in the ranking results. The specific steps are as follows:

[0049] Compliance Review: If the advertisement's content receives a compliance score of C... s<6, or the number of historical violations by the advertiser (V) n >3, that is Assume V max =10, then V norm If the value is greater than 0.27, it is judged as non-compliant and will be directly filtered out and will not participate in the subsequent sorting.

[0050] Abnormal Ad Filtering: If the ad's preset bid B p B's bid was more than three times the average bid for ads in the same category. p >3×B avg Or the historical click-through rate (P / C) of the ad. cv P's conversion rate is 0.3 times lower than the average conversion rate of similar ads. cv <0.3×P cvavg If the value is ) = , it is considered an abnormal advertisement and will be filtered out.

[0051] Step 7: Ad sorting and dynamic updates;

[0052] Initial sorting;

[0053] The advertisements that have passed the screening in step 6 will be processed according to the S calculated in step 5. total Sort the ads from highest to lowest to obtain the initial sorted list L. init ;

[0054] Dynamic update mechanism;

[0055] Time-triggered update: Recalculate the S of all ads every 30 minutes. total P cv Real-time changing metrics, and reordering;

[0056] Event-triggered update: When a user performs a key action, the user interest dimension score I is immediately recalculated. u And update the relevant ads' S total And the sorting results;

[0057] Ad status change triggers update: When the advertiser adjusts bid B p Advertising placement time S t When an ad ends or a new ad goes live, immediately include the ad in the calculation and update the sort list;

[0058] Final sorted output;

[0059] After initial sorting and dynamic update, the final advertisement sorting list L is output. final It will then be pushed to the display area of ​​the page the user is currently browsing.

[0060] In one embodiment, an advertisement sorting device includes:

[0061] The basic information collection and preprocessing module for advertisements collects basic information about the advertisements to be sorted and then preprocesses the collected information.

[0062] The user profile and interest dimension calculation module constructs user profiles based on the platform users' historical behavior data, and then calculates the matching degree between advertisements and user interests.

[0063] The advertising value calculation module includes immediate conversion value and long-term ecosystem value, which together constitute the comprehensive advertising value score.

[0064] The User Experience Dimension Calculation Module comprehensively considers the impact of advertisements on the user's browsing experience and ultimately derives a comprehensive user experience score.

[0065] The multi-dimensional calculation module performs multi-dimensional weighted comprehensive score calculation.

[0066] The review and filtering module performs compliance reviews and filters out anomalies in advertisements.

[0067] The sorting module sorts the filtered ads from highest to lowest to obtain an initial ad sorting list.

[0068] The beneficial effects of this invention are as follows: Compared with the prior art, this invention solves the problem of the single dimension in the prior art by constructing three core dimensions: user interest, advertising value, and user experience. The user interest dimension is divided into short-term and long-term interests, and the weights are dynamically adjusted based on user activity to accurately capture changes in user interests. The advertising value dimension integrates immediate conversion and long-term ecological value, ensuring advertisers achieve their short-term marketing goals while maintaining the long-term health of the platform ecosystem through considerations of advertiser qualifications, violations, and compliance. The user experience dimension addresses multiple aspects, including display frequency, format adaptation, and privacy protection, to avoid excessive user disturbance. Furthermore, it improves the poor adaptability to cold starts and sudden changes in interest. For cold start ads, the average conversion rate of ads in the same category is used to replace missing historical conversion data, allowing them to participate in the ranking normally and avoiding the "Matthew effect," giving high-quality new ads a reasonable exposure opportunity. For sudden changes in user interest, the short-term interest tag calculation and event-triggered update mechanism can quickly respond to key user behaviors and adjust the ranking in a timely manner, ensuring that ads always match the user's current interests and improving ranking accuracy. Attached Figure Description

[0069] Figure 1 This is a three-dimensional structural schematic diagram of the present invention; Detailed Implementation

[0070] In the description of this invention, it should be noted that the terms "front", "up", "down", "left", "right", "vertical", "horizontal", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0071] To address the problems existing in the background art, this application proposes the following technical solution:

[0072] Step 1: Collection and preprocessing of basic advertising information;

[0073] Content collected;

[0074] Collect basic information about the ads to be sorted, including advertiser information (advertiser qualification level A). q Historical number of violations (V) n ), Ad content information (Ad type T) a Such as product advertisements and service advertisements; advertising content compliance score C. s Generated by the platform's compliance review system, with values ​​ranging from 0 to 10; advertising placement information (target audience G). t Preset bid B p Unit: Yuan / Click Targeting Time Period (S) t ).

[0075] Preprocessing operations;

[0076] Advertiser qualification level A q Quantification:

[0077] The qualification levels are divided into four levels: “Excellent”, “Good”, “Average”, and “Poor”, which correspond to quantitative values ​​of 4, 3, 2, and 1, respectively.

[0078] V number of violations in historical ad placements n Normalization is performed using the formula: Where V max V represents the maximum number of violations committed by a single advertiser in the platform's historical records. norm V is the normalized violation coefficient, ranging from 0 to 1. The more violations, the higher the V value. norm The larger the value, the stronger the negative impact on ad ranking.

[0079] Advertising content compliance score: C s Standardization process: using formulas Where C min =0,C max =10,C stdThis is the standardized compliance coefficient, ranging from 0 to 1. A higher compliance score results in a higher C. std The larger the value, the stronger the positive impact on ad ranking.

[0080] The preprocessing formula in this step aims to transform non-quantitative qualitative indicators (such as qualification level) and quantitative indicators of different magnitudes (such as number of violations and compliance score) into standardized indicators with a unified value range (0-1 or 1-4), laying the foundation for subsequent multi-dimensional weighted calculations and avoiding deviations in ranking results due to differences in indicator magnitudes.

[0081] Step 2: User profile and interest dimension calculation;

[0082] User profile building;

[0083] Interest matching degree calculation;

[0084] User profiles are built based on the platform users' historical behavioral data (browsing history, click history, purchase history, collection history), including basic user attributes (age A). g Gender G d , region R g Short-term interest tags (product / service categories frequently viewed in the past 7 days, L7) and long-term interest tags (product / service categories consistently viewed in the past 90 days, L7) 90 Consumption capacity level P c (Divided into four levels: "high", "medium-high", "medium", and "low", corresponding to quantification values ​​of 4, 3, 2, and 1, respectively).

[0085] The degree of match between advertisements and user interests is calculated, and it is divided into short-term interest match degree M. s Match with long-term interests M l The specific formula is as follows: Short-term interest matching degree: Where N s-match N represents the number of matches between the ad category and the user's high-frequency interest tags over the past 7 days. s-total M represents the total number of high-frequency interest tags used by users in the past 7 days. s The value ranges from 0 to 1; the more matches, the higher the value of M. s The larger.

[0086] Long-term interest matching degree: Where N l-match N represents the number of matches between the ad category and the user's long-term interest tags over the past 90 days. l-total The total number of long-term interest tags for users over the past 90 days; P c-match This is the matching coefficient between the target consumer group of the advertisement and the user's spending power level; a match is assigned a value of 1, and a mismatch is assigned a value of 0. M lWith a value range of 0-1, this formula takes into account both category matching and spending power matching, making it more in line with users' long-term and stable consumption needs.

[0087] Overall score for interest dimension;

[0088] User Interest Dimension Overall Score I u The calculation formula is: I u =α×M s +(1-α)×M l α is the short-term interest weighting coefficient, with a value ranging from 0 to 1. It is dynamically adjusted by the platform based on user activity. If the user's activity level in the past 7 days (such as the average number of daily views) is higher than the platform average, then α is 0.6 (with more emphasis on short-term interests); if it is lower than the average, then α is 0.4 (with more emphasis on long-term interests).

[0089] By separating short-term and long-term interest matching, the problem of existing technologies relying solely on historical click data and failing to adapt to sudden changes in user interests is solved. At the same time, the introduction of dynamic adjustment of α based on user activity further improves the accuracy of interest matching, providing a core basis for subsequent ad ranking based on user dimensions.

[0090] Step 3: Calculate the advertising value dimensions;

[0091] The advertising value dimensions include immediate conversion value and long-term ecosystem value, which together constitute the overall advertising value score V. a .

[0092] Instant conversion value calculation;

[0093] Instant conversion value V im The formula for calculating V, based on the advertiser's preset bid and historical conversion probability, is: im =B p ×P cv B p Preset click bid (RMB / click) for advertisers, P cv This is the historical click-through rate of the ad (number of clicks converted / total number of clicks), with a value range of 0-1.

[0094] For newly launched cold start ads, P cv Using the average conversion rate P of similar advertisements cvavg Substitution, i.e., P cv =P cvavg This solves the problem of missing conversion data for cold-start advertising.

[0095] Long-term ecological value calculation;

[0096] Long-term ecological value V loTo assess the long-term impact of advertising on the platform ecosystem, including advertiser qualifications, historical violations, and content compliance, the formula is: V lo =A q ×(1-V norm )×C std , where A q V is a quantitative value for the advertiser's qualification level (1-4). norm C is the normalized violation coefficient (0-1). std The standardized compliance coefficient is (0-1).

[0097] This embodies the concept that advertisements with "high-quality advertisers + low violations + high compliance" have greater long-term ecological value. For example, an advertisement with a qualification level of 4 (excellent), a violation coefficient of 0 (no violations), and a compliance coefficient of 1 (perfect compliance) has a V lo =4×1×1=4, the long-term ecological value is the highest. 3.3 Overall score of advertising value

[0098] Overall Advertising Value Score V a The calculation formula is: V a =β×V im +(1-β)×V lo β is the weighting coefficient for instant conversion value, ranging from 0 to 1, and is dynamically adjusted according to the platform's operational scenarios. During major promotional periods (such as "618" and "Double 11"), the platform focuses on instant conversion, and β is set to 0.7. During daily operations, the focus is on long-term ecosystem value, and β is set to 0.5. During the new product promotion period, in order to support high-quality new advertisements, β is set to 0.4 (reducing the weighting of instant conversion and increasing the weighting of long-term ecosystem value).

[0099] Unlike existing technologies that focus solely on the single dimension of immediate conversion value, this step separates immediate conversion from long-term ecosystem value and introduces a scenario-based dynamic weight β. This ensures that advertisers' short-term marketing needs are met while also taking into account the long-term healthy development of the platform ecosystem. At the same time, by using the average conversion rate of the same category to replace the conversion data of cold start ads, the cold start problem is effectively solved.

[0100] Step 4: Calculate user experience dimensions;

[0101] The user experience dimension comprehensively considers the impact of advertising on the user browsing experience, including ad display frequency, ad format suitability, and user privacy protection suitability, ultimately resulting in a comprehensive user experience score E. u .

[0102] Score for ad display frequency control;

[0103] To avoid user aversion to the same advertisement, an ad display frequency control score F is introduced. c The formula is: Where F current F represents the number of times the ad has been shown to the current user today. max The maximum number of times a single ad can be displayed per day for the platform (the default value is 5, which can be adjusted according to the ad type).

[0104] When F current ≤F max Time F c Values ​​range from 0 to 1; the more times the image is displayed, the higher the F value. c The smaller; when F c urrent>F max At that time, F c =0, the advertisement will no longer be shown to the user on that day to avoid excessively disturbing the user.

[0105] Ad format suitability score

[0106] Ad format suitability F a The formula for assessing the suitability of ad formats (such as image ads, video ads, and text ads) with the user's current browsing context is: Where n is the number of adaptation dimensions (default 3: device type, network environment, browsing time period), W i Let M be the weight of each dimension (W1 = 0.4, W2 = 0.3, W3 = 0.3). i These are the matching coefficients for each dimension (1 for a match, 0 for a non-match).

[0107] For example, when a user browses on a mobile device (device type matching M1=1), a 4G network (network environment matching video ad M2=1), or during evening hours (browsing time matching video ad M3=1), the F of the video ad... a =0.4×1+0.3×1+0.3×1=1, the highest compatibility; if the user browses on a PC (device type mismatch video ad M1=0), 2G network (network environment mismatch video ad M2=0), or during daytime hours (M3=1), the video ad's F a =0.4×0+0.3×0+0.3×1=0.3, which indicates a low degree of fit.

[0108] User privacy protection adaptation score;

[0109] User privacy protection compatibility P p The formula for determining whether ad placement aligns with users' privacy preferences is:

[0110]

[0111] For example, if a user sets "Do not allow location-based advertising", then only location-based ads will be displayed. p=0 (completely exceeded), ad P for mixed geographic and interest-based advertising p =0.5 (partially exceeded), interest-based advertising P p =1 (Completely matches).

[0112] Overall user experience score;

[0113] Overall user experience score E u The calculation formula is: E u =γ×F c +δ×F a +η×P p , where γ, δ, and η are the weights of each dimension, and γ+δ+η=1 (default γ=0.3, δ=0.4, η=0.3).

[0114] By integrating display frequency, format adaptation, and privacy protection, we can comprehensively measure the impact of advertising on user experience and avoid user churn caused by existing technologies neglecting user experience.

[0115] This step fills a gap in existing technologies regarding user experience considerations through the design of multi-dimensional user experience metrics. In particular, the introduction of user privacy protection adaptability complies with current data security regulations. At the same time, by controlling the display frequency, it effectively improves user acceptance of advertisements and provides key user experience dimensions for ad ranking.

[0116] Step 5: Calculate the multi-dimensional weighted comprehensive score;

[0117] The multi-dimensional weighted comprehensive score is the core basis for ad ranking, and its calculation formula is: S total =ω1×I u +ω2×V a +ω3×E u ,in:

[0118] S total The overall score for the advertisement, with a value range of O-(ω1×1+ω2×(B)). pmax ×1+4×1×1)+ω3×1)(Because there is an upper limit to the score of each dimension, there is also a reasonable upper limit to the overall score);

[0119] ω1, ω2, and ω3 are the weights of the user interest dimension, advertising value dimension, and user experience dimension, respectively, and ω1+ω2+ω3=1;

[0120] The weight values ​​are dynamically adjusted based on the platform's operational goals: when the platform goal is "to increase user stickiness", ω1 = 0.4, ω2 = 0.3, and ω3 = 0.3; when the goal is "to increase advertising revenue", ω1 = 0.2, ω2 = 0.5, and ω3 = 0.3; and when the goal is "to optimize the platform ecosystem", ω1 = 0.3, ω2 = 0.4, and ω3 = 0.3.

[0121] By weighting and integrating three core dimensions—user interest, advertising value, and user experience—the problem of single-dimensionality in existing technologies is solved. At the same time, by dynamically adjusting the weights to adapt to different operational goals, the ranking model achieves flexibility and targeting, ensuring that the ranking results are consistent with the platform's strategic objectives.

[0122] Step 6: Compliance review and abnormal advertising filtering;

[0123] After calculating S total Afterwards, compliance reviews and anomaly filtering are required for advertisements to prevent illegal or abnormal advertisements from appearing in the ranking results. The specific steps are as follows:

[0124] Compliance Review: If the advertisement's content receives a compliance score of C... s <6 (i.e., C) std <0.6), or the number of historical violations by the advertiser (V). n >3 (i.e.) Assume V max =10, then V norm If the value is greater than 0.27, it is considered non-compliant and will be filtered out directly, not participating in subsequent sorting.

[0125] Abnormal Ad Filtering: If the ad's preset bid B p More than 3 times the average bid of similar ads (B p >3×B avg (), or the historical click-through rate (P) of the advertisement. cv The conversion rate is 0.3 times lower than the average conversion rate of similar advertisements (P). cv <0.3×P cvavg If an ad is found to be an abnormal ad (possibly due to malicious bidding or low quality issues), it will be filtered out.

[0126] This step brings compliance review and anomaly filtering forward, completing the screening after calculating the comprehensive score and before the formal ranking. This avoids the problem of non-compliant ads occupying ranking resources and wasting platform traffic, while ensuring that all ads entering the ranking are compliant and high-quality ads, thus improving the effectiveness of the ranking results.

[0127] Step 7: Ad sorting and dynamic updates;

[0128] Initial sorting;

[0129] The advertisements that have passed the screening in step 6 will be processed according to the S calculated in step 5. total Sort the ads from highest to lowest to obtain the initial sorted list L. init .

[0130] Dynamic update mechanism;

[0131] To adapt to changes in user interests and ad delivery, a dynamic ranking update mechanism is set up:

[0132] Time-triggered update: Recalculate the S of all ads every 30 minutes. total (Focus on updating F) c (Display frequency), P cv (Real-time changing indicators such as conversion probability), and reorder them;

[0133] Event-triggered update: When a user performs a key action (such as clicking on a certain type of advertisement or purchasing a certain type of product), the user's interest dimension score I is immediately recalculated. u And update the relevant ads' S total And the sorting results;

[0134] Ad status change triggers update: When the advertiser adjusts bid B p Advertising placement time S t When an ad ends or a new ad goes live, the ad is immediately included in the calculation, and the sort list is updated.

[0135] Final sorted output

[0136] After initial sorting and dynamic update, the final advertisement sorting list L is output. final This data is then pushed to the display area of ​​the page the user is currently viewing. During the display process, the system records data such as the number of ad impressions, clicks, and conversions in real time, providing information for subsequent metrics (such as PageRank). cv F c The updates provide data support, forming a closed loop of "data collection - calculation and sorting - display and feedback - data update".

[0137] The initial ranking is based on the overall score S total This ensures the scientific and rational nature of the ranking, while the dynamic update mechanism solves the problem of static ranking results in existing technologies that cannot adapt to real-time changes. Time-triggered updates guarantee the timeliness of the indicators, event-triggered updates ensure a rapid response to changes in user interests, and ad status change-triggered updates enable real-time adaptation to ad delivery dynamics. The combination of these three features ensures that ad ranking remains accurate and efficient.

[0138] In summary, firstly, this embodiment addresses the issue of limited dimensions in existing technologies by constructing three core dimensions: user interest, advertising value, and user experience. The user interest dimension is broken down into short-term and long-term interests, with weights dynamically adjusted based on user activity levels. This accurately captures changes in user interests and avoids ranking biases caused by sudden shifts in interest. The advertising value dimension integrates immediate conversion and long-term ecosystem value, ensuring advertisers achieve their short-term marketing goals while maintaining the long-term health of the platform ecosystem through considerations of advertiser qualifications, violations, and compliance. The user experience dimension addresses display frequency, format adaptation, and privacy protection to avoid excessive user disruption, comply with data security regulations, significantly improve user acceptance of advertisements, and make the ranking results more comprehensive and scientific. Secondly, it breaks the limitations of fixed weighting by introducing multiple dynamic weighting coefficients. These coefficients can be flexibly adjusted based on user activity, platform operation scenarios (major promotions, daily activities, new product launches), and platform strategic goals (improving user stickiness, advertising revenue, and platform ecosystem). This allows the ranking to accurately adapt to different needs. Furthermore, it improves the poor adaptability to cold starts and sudden changes in user interest. For cold start ads, it uses the average conversion rate of ads in the same category to replace missing historical conversion data, ensuring they can participate in the ranking normally and avoiding the "Matthew effect," giving high-quality new ads a reasonable exposure opportunity. For sudden changes in user interest, through short-term interest tag calculation and event-triggered update mechanisms, it can quickly respond to key user behaviors and adjust the ranking in a timely manner, ensuring that ads always match the user's current interests and improving ranking accuracy. In addition, it achieves deep integration of compliance and privacy protection. The solution prioritizes compliance review and anomaly filtering, filtering out illegal, low-quality, and abnormal ads before ranking, preventing illegal ads from wasting ranking resources and traffic. It also introduces user privacy protection adaptability indicators to ensure that ad placement complies with user privacy settings, reducing user resistance and increasing platform trust. Finally, a data closed loop and a win-win situation for all parties are formed. The system records ad display, click and conversion data in real time, providing support for the updating of metrics.

[0139] Although embodiments of the invention have been shown and described, the scope of the invention will be defined by the appended claims and their equivalents by those skilled in the art.

Claims

1. A method for sorting advertisements, characterized in that, Specifically, the following steps are included: Step 1: Collection and preprocessing of basic advertising information. Collect basic information of the advertisements to be sorted, including advertiser information, advertisement content information, and advertisement placement information. Then, preprocess the collected information. Step 2: User profile and interest dimension calculation. Based on the historical behavior data of platform users, user profiles are constructed, including basic user attributes, short-term interest tags, long-term interest tags, and spending power levels. Then, the matching degree between advertisements and user interests is calculated. Step 3: Calculate the advertising value dimensions. The advertising value dimensions include immediate conversion value and long-term ecosystem value, which together constitute the comprehensive advertising value score. Step 4: User experience dimension calculation. The user experience dimension comprehensively considers the impact of advertisements on the user browsing experience, including the frequency of ad display, the suitability of ad format and the suitability of user privacy protection, and finally obtains the comprehensive user experience score. Step 5: Based on the above steps, calculate the multi-dimensional weighted comprehensive score; Step 6: Conduct compliance review and anomaly filtering of advertisements; Step 7: Sort the ads that have passed the filtering in Step 6 from highest to lowest to obtain the initial ad sorting list.

2. The method for sorting advertisements according to claim 1, characterized in that, In step 1, the advertiser information includes the advertiser's qualification level A. q Historical number of violations (V) n The advertising content information includes the advertising type T. a Advertising content compliance score: C s Advertising placement information includes the target audience G. t Preset bid B p ; Advertiser qualification level A q Quantification: The qualification levels are divided into four levels: "Excellent", "Good", "Average", and "Poor", which correspond to quantitative values ​​of 4, 3, 2, and 1, respectively. V number of violations in historical ad placements n Normalization is performed using the formula: Where V m ax represents the maximum number of violations committed by a single advertiser in the platform's historical records, and V represents... norm V is the normalized violation coefficient, ranging from 0 to 1. The more violations, the higher the V value. norm The larger the value, the stronger the negative impact on ad ranking. Advertising content compliance score: C s Standardization process: using formulas Where C min =0,C max =10,C std This is the standardized compliance coefficient, ranging from 0 to 1. A higher compliance score results in a higher C. std The larger.

3. The method for sorting advertisements according to claim 2, characterized in that, In step 2, the user's basic attributes include age A. g Gender G d , region R g Short-term interest tags include product / service categories frequently viewed in the past 7 days (L7); long-term interest tags include product / service categories consistently viewed in the past 90 days (L). 90 Consumption capacity level P c ; The degree of match between advertisements and user interests is calculated, and it is divided into short-term interest match degree M. s Match with long-term interests M l The specific formula is as follows: Short-term interest matching degree: Where N s-match N represents the number of matches between the ad category and the user's high-frequency interest tags over the past 7 days. s-total M represents the total number of high-frequency interest tags used by users in the past 7 days. s The value ranges from 0 to 1; the more matches, the higher the value of M. s The larger; Long-term interest matching degree: Where N l-match N represents the number of matches between the ad category and the user's long-term interest tags over the past 90 days. l-total The total number of long-term interest tags for users over the past 90 days; P c-match M is the matching coefficient between the target consumer group of the advertisement and the user's spending power level. A match is assigned a value of 1, and a mismatch is assigned a value of 0. l The value range is 0-1; Overall score for interest dimension; User Interest Dimension Overall Score I u The calculation formula is: I u =α×M s +(1-α)×M l , where α is the short-term interest weighting coefficient, with a value ranging from 0 to 1.

4. The method for sorting advertisements according to claim 3, characterized in that, In step 3, the overall advertising value score is V. a ; include: Instant conversion value calculation; Instant conversion value V im The formula for calculating V, based on the advertiser's preset bid and historical conversion probability, is: im =B p ×P cv B p Preset click bids for advertisers, P cv This represents the historical click-through rate of the ad, with a value ranging from 0 to 1. For newly launched cold start ads, P cv Using the average conversion rate P of similar advertisements cvavg Substitution, i.e., P cv =P cvavg ; Long-term ecological value calculation; Long-term ecological value V lo To assess the long-term impact of advertising on the platform ecosystem, including advertiser qualifications, historical violations, and content compliance, the formula is: V lo =A q ×(1-V norm )×C std , where A q V is a quantitative value for the qualification level of advertisers. norm C is the normalized violation coefficient. std The standardized compliance coefficient; Overall advertising value score; Overall Advertising Value Score V a The calculation formula is: V a =β×V im +(1-β)×V lo , where β is the weighting coefficient for instant conversion value, ranging from 0 to 1, and is dynamically adjusted according to the platform's operational scenarios.

5. The method for sorting advertisements according to claim 4, characterized in that, In step 4, the overall user experience score is E. u ; include: Score for ad display frequency control; Ad display frequency control score F c The formula is: Where F current F represents the number of times the ad has been shown to the current user today. max The maximum number of times a single advertisement can be displayed per day as set for the platform; When F current ≤F max Time F c Values ​​range from 0 to 1; the more times the image is displayed, the higher the F value. c The smaller; when F c urrent>F max At that time, F c =0, the advertisement will no longer be shown to this user on that day to avoid excessively disturbing the user; Ad format suitability score; Ad format suitability F a The formula for considering the suitability of the ad format to the user's current browsing context is: Where n is the number of adaptation dimensions, W i M represents the weights for each dimension. i These are the matching coefficients for each dimension; User privacy protection adaptation score; User privacy protection compatibility P p The formula for determining whether ad placement aligns with users' privacy preferences is: Overall user experience score; Overall user experience score E u The calculation formula is: E u =γ×F c +δ×F a +η×P p , where γ, δ, and η are the weights of each dimension, and γ+δ+η=1.

6. The method for sorting advertisements according to claim 5, characterized in that, In step 5, the multi-dimensional weighted comprehensive score is the core basis for ad ranking, and its calculation formula is: S total =ω1×I u +ω2×V a +ω3×E u ,in: S total The overall score for the advertisement has a range of values: O-(ω1×1+ω2×(B pmax ×1+4×1×1)+ω3×1); ω1, ω2, and ω3 are the weights of the user interest dimension, advertising value dimension, and user experience dimension, respectively, and ω1+ω2+ω3=1.

7. The method for sorting advertisements according to claim 6, characterized in that, Step 6: After calculating S total Afterwards, compliance reviews and anomaly filtering are required for advertisements to prevent illegal or abnormal advertisements from appearing in the ranking results. The specific steps are as follows: Compliance Review: If the advertisement's content receives a compliance score of C... s <6, or the number of historical violations by the advertiser (V) n >3, that is Assume V max =10, then V norm If the value is greater than 0.27, it is judged as non-compliant and will be directly filtered out and will not participate in the subsequent sorting. Abnormal Ad Filtering: If the ad's preset bid B p B's bid was more than three times the average bid for ads in the same category. p >3×B avg Or the historical click-through rate (P / C) of the ad. cv P's conversion rate is 0.3 times lower than the average conversion rate of similar ads. cv <0.3×P cvavg If the value is equal to 0, it is considered an abnormal advertisement and will be filtered out. Step 7: Ad sorting and dynamic updates; Initial sorting; The advertisements that have passed the screening in step 6 will be processed according to the S calculated in step 5. total Sort the ads from highest to lowest to obtain the initial sorted list L. init ; Dynamic update mechanism; Time-triggered update: Recalculate the S of all ads every 30 minutes. total P cv Real-time changing metrics, and reordering; Event-triggered update: When a user performs a key action, the user interest dimension score I is immediately recalculated. u And update the relevant ads' S total And the sorting results; Ad status change triggers update: When the advertiser adjusts bid B p Advertising placement time S t When an ad ends or a new ad goes live, immediately include the ad in the calculation and update the sort list; Final sorted output; After initial sorting and dynamic update, the final advertisement sorting list L is output. final It will then be pushed to the display area of ​​the page the user is currently browsing.

8. An advertising sorting device, characterized in that, include: The basic information collection and preprocessing module for advertisements collects basic information about the advertisements to be sorted and then preprocesses the collected information. The user profile and interest dimension calculation module constructs user profiles based on the platform users' historical behavior data, and then calculates the matching degree between advertisements and user interests. The advertising value calculation module includes immediate conversion value and long-term ecosystem value, which together constitute the comprehensive advertising value score. The User Experience Dimension Calculation Module comprehensively considers the impact of advertisements on the user's browsing experience and ultimately derives a comprehensive user experience score. The multi-dimensional calculation module performs multi-dimensional weighted comprehensive score calculation. The review and filtering module performs compliance reviews and filters out anomalies in advertisements. The sorting module sorts the filtered ads from highest to lowest to obtain an initial ad sorting list.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory is used to store computer programs; and the processor is used to execute the programs stored in the memory to implement the advertising sorting method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the advertising sorting method according to any one of claims 1 to 6.

Images