Advertisement material intelligent scoring method based on data integrity adaptive weight

Abstract

The present application relates to the technical field of data analysis, and more particularly to an advertisement material intelligent scoring method based on data completeness adaptive weight, which comprises the following steps: first, extracting advertisement delivery materials and collecting corresponding delivery effect data; second, calculating the data completeness of the delivery effect data; third, according to the data completeness, calling corresponding content quality weight and data performance weight; fourth, inputting the advertisement delivery materials after feature extraction into a preset multi-modal large language model to output initial scores of the advertisement materials containing content quality dimensions and data performance dimensions; and finally, weighting and summing the initial scores of the advertisement materials by using the content quality weight and the data performance weight to obtain advertisement delivery decision scores for supporting delivery decision ranking. Compared with the prior art, the present application solves the cold start dead loop problem and provides reliable advertisement delivery scores and suggestions for advertisers.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology. More specifically, this invention relates to an intelligent scoring method for advertising creatives based on adaptive weights of data completeness. Background Technology

[0002] Ad performance data scoring is a reference score used to quantify the effectiveness of advertisers' campaigns and support ad placement decisions. Traditionally, this involves analysts calculating a score based on historical ad delivery data, such as click-through rate (CTR), conversion rate (CVR), and cost per mille (CPM). The calculation typically uses static weights: CTR is normalized and multiplied by a weight of 0.4, CVR by a weight of 0.3, and cost per mille (CPM) by a weight of 0.3, then summed to obtain the final score. This approach lacks automation and adapts poorly to static weighting scenarios, proving inefficient and inaccurate in AI and big data applications.

[0003] To improve the efficiency of advertising placement data scoring, Chinese Patent No. CN119295148B discloses a placement analysis method and system based on content characteristics and dissemination effects. It introduces a feature analysis engine and PageRank model to score advertising placement content, which to a certain extent improves the scoring efficiency of advertising placement content and has high accuracy.

[0004] However, for newly created ad creatives, especially those with no or only a small amount of historical data, the aforementioned data-driven approach cannot provide meaningful scores. This results in new creatives being consistently at a disadvantage in ad placement decisions, creating a "cold start vicious cycle"—a vicious cycle where without an initial score, there is no opportunity to place ads, and without an opportunity to place ads, there is no way to accumulate data and obtain a score.

[0005] In summary, existing advertising data scoring technologies suffer from a cold start loop problem. Summary of the Invention

[0006] To address the cold start deadlock problem in existing advertising data scoring technologies, this invention discloses an intelligent scoring method for advertising creatives based on adaptive weights for data completeness.

[0007] In a first aspect, this invention discloses an intelligent scoring method for advertising creatives based on adaptive weights of data completeness, comprising:

[0008] Extract advertising creatives and collect corresponding campaign performance data; Calculate the data completeness of the campaign performance data; Based on the data completeness, retrieve the corresponding content quality weight and data performance weight; After feature extraction of the advertising creative, it is input into a preset multimodal large language model, and the output is an initial score of the advertising creative that includes content quality dimension and data performance dimension; The initial scores of ad creatives are weighted and summed using content quality weights and data performance weights to obtain an ad placement decision score used to support the ranking of placement decisions.

[0009] Beneficial Effects: This invention introduces data integrity verification in the initial stage of ad creative analysis to dynamically adjust the weights of content quality and data performance. These two dynamic weights are then weighted together with the initial ad creative score output by artificial intelligence to obtain the ad placement decision score. Unlike existing technologies, this invention overcomes the "cold start loop" defect caused by relying entirely on "historical data performance data," enabling reliable ad placement decision scores even in scenarios with no or only a small amount of historical data.

[0010] Preferably, the data completeness for calculating the campaign performance data includes: Iterate through all dimensions of the metrics data in the campaign performance data and count the number of metrics that are not empty and are greater than 0. Determine the data completeness based on the number of indicators.

[0011] Preferably, based on data completeness, the corresponding content quality weight and data performance weight are retrieved, including: Based on the type tags of the advertising creative, retrieve the corresponding dynamic weight database; The content quality weight and data performance weight corresponding to data completeness are obtained by matching from the dynamic weight database; Record the weights for content quality and data performance.

[0012] Preferably, after feature extraction of the advertising creative, it is input into a pre-defined multimodal large language model, which outputs an initial score for the advertising creative that includes content quality and data performance dimensions, including: Extracting video content input features and data text input features from advertising materials; The video content input features and data text input features are input into the multimodal large language model, and a preset rating prompt word template is invoked; the rating prompt word template guides the model to output structured rating results; The multimodal large language model is driven to output the initial score of the advertising creative.

[0013] Preferably, the video content input features include data source locators, video features, and keyframe features; the data text input features include available indicator features and missing indicator features.

[0014] Preferably, after obtaining the advertising placement decision score, the method of the present invention further includes: The advertising placement decision score is clamped to a percentage range for output, resulting in a percentage output result. Match advertising placement decision suggestions corresponding to the percentage-based output results from the pre-set decision knowledge base; The advertising placement decision suggestions are input into a multimodal large language model, and the placement decision analysis text is output.

[0015] Preferably, after outputting the delivery decision analysis text, the method of the present invention further includes: Identify whether the delivery decision analysis text contains formatting errors; If so, correct any formatting errors.

[0016] Preferably, the campaign performance data includes conversion metrics, playback behavior metrics, interaction behavior metrics, and basic campaign metrics used to measure costs.

[0017] Preferably, the sum of the content quality weight and the data performance weight is 1.

[0018] Secondly, the present invention discloses an intelligent scoring system for advertising creatives based on adaptive weights of data integrity, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent scoring method for advertising creatives based on adaptive weights of data integrity described in the first aspect is implemented.

[0019] The advantages of this invention, which differ from existing technologies, are as follows: (1) The method of the present invention overcomes the defect of the "cold start dead loop" in the prior art due to its complete reliance on "historical data performance data", so that the advertising materials can still obtain a reliable advertising decision score even in the case of no advertising historical data or only a small amount of advertising historical data.

[0020] (2) The method of the present invention has an adaptive weight smooth transition function, which can automatically adjust the weight distribution of content quality score and data performance score according to the data completeness of the material, so as to adapt to complex advertising scenarios.

[0021] (3) The method of the present invention supports comprehensive evaluation from dimensions such as conversion, playback behavior, interaction behavior and cost estimation, and can provide advertisers with reliable and clear placement analysis suggestions, thereby improving the experience. Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart of the intelligent scoring method for advertising materials based on adaptive weighting of data integrity in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the intelligent scoring system for advertising materials based on adaptive weighting of data integrity in Embodiment 2 of the present invention. Detailed Implementation

[0023] 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, not all, of the embodiments of the present invention. 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.

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] Example 1 like Figure 1 As shown, this embodiment discloses an intelligent scoring method for advertising creatives based on adaptive weights of data completeness, including: S10: Extract advertising creative materials and collect corresponding advertising performance data.

[0026] In this embodiment, advertising creative materials are primarily extracted through the data interface or data warehouse of the advertising platform. The advertising platform is an online advertising interaction platform, which can be a video interaction or e-commerce platform such as TikTok, Taobao, or YouTube. The data warehouse is deployed on local computer equipment or a server.

[0027] It should be noted that, in order to improve the reliability of the method in this embodiment in data scoring and analysis, the above-mentioned campaign performance data includes conversion index data, playback behavior index data, interaction behavior index data, and basic campaign index data.

[0028] More specifically, the basic metrics for ad placement include the ad spend, number of impressions, and number of clicks for a single ad placement; the conversion metrics include the number of conversions (whether interactive users clicked to trigger a link jump) and the cost per conversion; the playback behavior metrics include the number of clicks to play, the number of times the play was 25% complete, the number of times it was 50% complete, the number of times it was 75% complete, and the number of times it was 100% complete; and the interaction behavior metrics include the number of likes and shares.

[0029] In summary, by introducing the above four dimensions and the corresponding 12 key indicators for integrated analysis, a reliable data foundation can be provided for subsequent data analysis, making the obtained scoring results more accurate and more versatile.

[0030] S20: Calculate the data completeness of the campaign performance data.

[0031] Specifically, step S20 above includes: S21: Iterate through all dimensions of the metrics data in the campaign performance data and count the number of metrics that are not empty and are greater than 0.

[0032] In this embodiment, the number of non-empty indicators that are greater than 0 can be counted by iterating through the above 12 indicators. As for data items that are empty or less than 0, they can be marked as missing and encoded to form a missing item feature matrix, which can be used as context data for subsequent model analysis.

[0033] S22: Determine the data completeness based on the number of indicators.

[0034] More specifically, the formula for calculating data integrity in step S22 above is:

[0035] In the formula, This indicates the data integrity level, and its value ranges from 0 to 1. Indicates the number of valid indicators; This indicates the total number of indicators.

[0036] Through the above steps S21-S22, the data integrity can be calculated efficiently and conveniently, providing reliable data evidence for dynamic weight allocation.

[0037] S30: Based on the data completeness, retrieve the corresponding content quality weight and data performance weight.

[0038] In this embodiment, the sum of the two weights is 1.

[0039] Furthermore, step S30 above includes: S31: Retrieve the corresponding dynamic weight database based on the type tag of the advertising creative.

[0040] In one embodiment, different dynamic weight databases can be configured for different ad creatives. For example, if an ad creative belongs to the category of fast-moving consumer goods (FMCG) cosmetics, a type label can be assigned to the table header when configuring the dynamic weight database to achieve classification. Then, when analyzing a specific ad creative, its type label is extracted first, and then the corresponding dynamic weight database is matched from the local storage medium.

[0041] S32: Match the content quality weight and data performance weight corresponding to the data completeness from the dynamic weight database.

[0042] In this embodiment, to achieve adaptive dynamic decision-making, the method mainly employs a three-level classification decision-making mechanism, and its decision classification table is shown in Table 1: Table 1

[0043] In other embodiments, to improve the level of detail of the method in this embodiment, more classification levels and their corresponding weight configurations can be set.

[0044] S33: Record the weights for content quality and data performance.

[0045] Finally, record the current weighting and data completeness to explain the scoring criteria to users.

[0046] Through steps S31-S33, the method automatically adjusts the weighting of content quality and data performance scores based on the data completeness of the materials. When the data completeness is 0%, AI is used directly for content scoring; when the data completeness is between 0-50%, content accounts for 70% and data accounts for 30%; when the data completeness exceeds 50%, data accounts for 60% and content accounts for 40%. This three-level adaptive mechanism enables the scoring scheme to smoothly cover the entire lifecycle from "cold start of new materials" to "performance evaluation of mature materials," exhibiting superior versatility.

[0047] S40: After extracting features from the advertising creative, input it into the preset multimodal large language model, and output the initial score of the advertising creative, which includes the content quality dimension and the data performance dimension.

[0048] Specifically, step S40 includes: S41: Extract video content input features and data text input features of advertising materials.

[0049] The video content input features include data source locators, video features, and keyframe features. The text data input features include available indicator features and missing indicator features.

[0050] More specifically, the aforementioned data source locators can be obtained by encoding and converting the video source URL or image source URL. Video features and keyframe features are the standard features extracted from source material, and will not be elaborated upon here.

[0051] Available metrics include a metric name and a specific value, for example, "impressions (views): 50,000, clicks (link clicks): 1,500, CTR (click-through rate): 3.0%". Missing metrics, i.e., missing items among the above 12 metrics, retain their names, while assigning empty values ​​or 0 to their specific values.

[0052] S42: Input the video content input features and data text input features into the multimodal large language model, and call the preset scoring prompt word template.

[0053] It should be explained that the aforementioned scoring prompt word template is used to guide the model to output structured scoring results. It can be pre-loaded into a template knowledge base to support the use of multimodal large language models. The aforementioned multimodal large language model can employ existing mainstream models.

[0054] S43: Drives the multimodal large language model to output the initial score of advertising creatives.

[0055] It needs further explanation that, in this embodiment, the aforementioned multimodal large language model primarily scores based on content quality and data performance dimensions. Regarding the content dimension, the lower the similarity of the analyzed material's features to other materials, the higher its clarity, the more compact the information rhythm, and the higher the copywriting quality, the higher the corresponding score. Conversely, for the data performance dimension, the higher its effective value, the higher the corresponding score. In addition to generating the above two sets of scores, it can also indicate the evaluation and optimization content corresponding to the multimodal large language model data. The generation guidance of the above scoring and content is mainly achieved using the aforementioned scoring prompt word template. When the generated content is incomplete, prompt words can also be constructed online to guide the model's output.

[0056] S50: The initial score of the ad creative is weighted and summed using content quality weight and data performance weight to obtain the ad placement decision score used to support the ranking of placement decisions.

[0057] Specifically, the formula for calculating the above-mentioned advertising placement decision score is as follows:

[0058] In the formula, This indicates a score for ad placement decisions; Indicates content quality weight; This indicates the content quality score; Indicates a score for data performance; This indicates the weight of the data performance.

[0059] Furthermore, in order to eliminate misunderstandings by advertisers and optimize the intuitive display effect, after step S50 above, the method of this embodiment further includes: First, the ad placement decision score is clamped to a percentage range for output, resulting in a percentage-based output. Then, ad placement decision suggestions corresponding to the percentage-based output are matched from a pre-defined decision knowledge base. Finally, the ad placement decision suggestions are input into a multimodal large language model, outputting the ad placement decision analysis text.

[0060] It should be further noted that the knowledge base template for providing prompts for ad placement decision-making can be configured as follows: Table 2

[0061] Furthermore, after outputting the delivery decision analysis text, the method in this embodiment also includes: Identify whether there are formatting errors in the delivery decision analysis text.

[0062] If so, correct any formatting errors.

[0063] As a preferred embodiment, considering that the original text returned by the above model may contain non-standard formatting, the following parsing error-tolerant processing can be performed: First, remove any possible errors in the response. <think> ...< / think> <think> ...< / think> Tags (the built-in thought process of some models).

[0064] Next, extract the JSON content from the Markdown code block (`json ... `).

[0065] Then, automatic repairs are performed on the JSON, fixing issues such as extra quotes, orphaned objects in arrays, and response truncation. In addition, callbacks can be used to automatically map historical versions to maintain backward compatibility.

[0066] After the repair is complete, an integrity check must be performed to ensure that necessary fields such as rating data and recommendations have values ​​and are within a reasonable range. If the check fails, a clear error message will be returned.

[0067] Based on the above technical discussion, and differing from existing technologies, the method of this embodiment has at least the following advantages: Firstly, the method in this embodiment can automatically adjust the weighting of content quality score and data performance score based on the data completeness of the materials: For example, when data completeness is 0%, the scoring relies entirely on AI. When data completeness is between 0% and 50%, content accounts for 70% and data accounts for 30%. When data completeness exceeds 50%, data accounts for 60% and content accounts for 40%. This adaptive mechanism allows the scoring scheme to smoothly cover the entire lifecycle from "cold start of new content" to "effective evaluation of mature content".

[0068] Secondly, the method in this embodiment uses 12 indicators across four dimensions—basic delivery indicators, conversion indicators, playback behavior indicators, and interaction indicators—to calculate data completeness, providing a systematic data quality quantification standard that allows users to clearly understand the reliability of the current rating.

[0069] Third, the AI ​​model simultaneously receives visual content from videos / images and text summaries of 12 indicators (including descriptions of available and missing indicators). When conducting cross-modal comprehensive evaluation, it outputs content quality scores, data performance scores, and optimization suggestions, achieving in-depth analysis from two dimensions: "content + data".

[0070] Fourth, the scoring results are mapped to three decision levels: For example, the categories of Excellent (80-100 points, it is recommended to increase investment), High Potential (60-79 points, it is recommended to continue after optimization), and Need for Iteration (1-59 points, it is recommended to modify or stop investment) effectively lower the threshold for manual decision-making.

[0071] In summary, the method described in this embodiment can be widely applied to scenarios such as creative material selection for advertising platforms, creative material performance evaluation for e-commerce operations, and creative quality control for content creation teams. Through a dynamic weighting mechanism that adapts to data completeness, it effectively solves the industry pain point of cold-start scoring for new creative materials, achieving continuous and reasonable scoring throughout the entire lifecycle of the materials. Furthermore, the three-level classification lowers the decision-making threshold for non-professionals and possesses good industrial applicability.

[0072] Example 2 like Figure 2 As shown, this embodiment discloses an intelligent scoring system for advertising creatives based on adaptive weights of data integrity, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent scoring method for advertising creatives based on adaptive weights of data integrity described in Embodiment 1 is implemented.

[0073] The system in this embodiment also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art, and therefore will not be described in detail here.

[0074] In this embodiment, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions that can be stored or otherwise maintained by such a computer-readable medium.

[0075] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0076] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for intelligent scoring of advertising creatives based on adaptive weighting of data completeness, characterized in that, include: Extract advertising creatives and collect corresponding campaign performance data; Calculate the data completeness of the campaign performance data; Based on the data completeness, retrieve the corresponding content quality weight and data performance weight; After feature extraction of the advertising materials, the data is input into a preset multimodal large language model, and the initial score of the advertising materials, including content quality and data performance dimensions, is output. The initial score of the advertising creative is weighted and summed using the content quality weight and the data performance weight to obtain an advertising placement decision score used to support the ranking of placement decisions.

2. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 1, characterized in that, Calculating the data completeness of the campaign performance data includes: Iterate through all dimensions of the metrics data in the campaign performance data and count the number of metrics that are non-empty and greater than 0. The data completeness is determined based on the number of indicators.

3. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 1, characterized in that, Based on the data completeness, retrieve the corresponding content quality weight and data performance weight, including: Based on the type tags of the advertising materials, retrieve the corresponding dynamic weight database; The content quality weight and data performance weight corresponding to the data completeness are obtained by matching from the dynamic weight database. The content quality weight and the data performance weight are recorded.

4. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 1, characterized in that, After feature extraction of the advertising creatives, the data is input into a preset multimodal large language model, which outputs an initial score for the advertising creatives that includes content quality and data performance dimensions, including: Extract the video content input features and data text input features of the advertising materials; The video content input features and the data text input features are input into the multimodal large language model, and a preset rating prompt word template is invoked; wherein, the rating prompt word template guides the model to output a structured rating result; The multimodal large language model is driven to output the initial score of the advertising material.

5. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 4, characterized in that, The video content input features include data source locators, video features, and keyframe features; the data text input features include available indicator features and missing indicator features.

6. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 1, characterized in that, After obtaining the advertising placement decision score, the method further includes: The advertising placement decision score is clamped to a percentage range for output, resulting in a percentage output result. Match the advertising placement decision suggestions corresponding to the percentage output results from the preset decision knowledge base; The advertising placement decision suggestions are input into the multimodal large language model, and the placement decision analysis text is output.

7. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 6, characterized in that, After outputting the delivery decision analysis text, the method further includes: Identify whether the delivery decision analysis text contains formatting errors; If so, correct any formatting errors.

8. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 1, characterized in that, The campaign performance data includes conversion metrics, playback behavior metrics, interaction behavior metrics, and basic campaign metrics used to measure costs.

9. The intelligent scoring method for advertising creatives based on adaptive weighting of data completeness according to claim 1, characterized in that, The sum of the content quality weight and the data performance weight is 1.

10. An intelligent scoring system for advertising creatives based on adaptive weighting of data completeness, characterized in that, It includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the intelligent scoring method for advertising creatives based on adaptive weights of data integrity as described in any one of claims 1-9.

Images