Advertisement quotation method and related device

Abstract

The invention provides an advertisement quotation method and a related device. The method comprises the steps of obtaining an advertisement quotation request of an advertiser; performing advertisement field matching on an advertisement type corresponding to the advertisement quotation request, and determining an advertisement field to which the advertisement type belongs; obtaining user data and advertisement data of the advertisement field; taking the user data and the advertisement data as input of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field; performing advertisement recall on the plurality of advertisements based on a recall model, and further screening the plurality of advertisements to obtain an advertisement list; the advertisement list comprises at least two advertisements screened from the plurality of advertisements; sorting at least two advertisements in the advertisement list, and performing price estimation according to a sorting result to obtain an advertisement quotation table; the advertisement quotation table comprises quotations of the at least two advertisements; and sending the advertisement quotation table to the advertiser.

Description

Technical Field

[0001] The present application relates to the technical field of advertisement push, and in particular to an advertisement quotation method and related devices. Background Art

[0002] In the modern advertising industry, advertisers and demanders are pursuing efficient and accurate pricing of advertising space. In traditional methods, advertisers and demanders usually make bids manually or semi-automatically, which may cause the following problems:

[0003] (1) Low efficiency: Manual or semi-automatic methods require a lot of time to process a large amount of advertising information and quotation requests, reducing the operating efficiency of the entire system;

[0004] (2) High subjectivity: Due to the subjective judgment and personal experience of the participants, it is easy to make wrong decisions and waste resources, and it is impossible to achieve the optimal advertising strategy;

[0005] (3) Lack of objective standards: Manual or semi-automatic methods rely on manpower and experience, making it difficult to ensure the objectivity and consistency of quotations, thus affecting the effectiveness of advertising and its profit prospects.

[0006] Therefore, for the current advertising market, it is of great significance to develop an efficient and automated advertising quotation system.

[0007] The methods described in this section are not necessarily methods that have been previously conceived or employed. Unless otherwise indicated, it should not be assumed that any method described in this section is considered to be prior art simply because it is included in this section. Similarly, unless otherwise indicated, the issues mentioned in this section should not be considered to have been recognized in any prior art. Summary of the invention

[0008] In view of this, an embodiment of the present application provides an advertisement quotation method and related devices.

[0009] In a first aspect, an embodiment of the present application provides an advertisement quotation method, the method comprising:

[0010] Obtaining an advertisement quotation request from an advertiser; the advertisement quotation request includes an advertisement type;

[0011] Performing advertising domain matching on the advertising type corresponding to the advertising quotation request to determine the advertising domain to which the advertising type belongs;

[0012] Acquire user data and advertisement data of the advertisement domain; wherein the user data includes user behavior data and context information corresponding to the advertisement domain, and the advertisement data includes historical advertisement delivery data and delivery results corresponding to the advertisement domain;

[0013] Using the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement domain;

[0014] Recalling the multiple advertisements based on the recall model, further screening the multiple advertisements to obtain an advertisement list; the advertisement list includes at least two advertisements screened from the multiple advertisements;

[0015] Sorting at least two advertisements in the advertisement list, and performing price estimation according to the sorting result to obtain an advertisement quotation table; the advertisement quotation table includes quotations of the at least two advertisements;

[0016] The advertisement quotation form is sent to the advertiser.

[0017] As an embodiment, the user data and the advertisement data are used as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field, including:

[0018] Cleaning the user data and the advertisement data to remove noise data and outliers;

[0019] Perform feature extraction on the cleaned user data and advertisement data to obtain feature vectors;

[0020] Inputting the feature vector into the traffic screening model to obtain classification results of each advertisement; the classification results include relevant advertisements and irrelevant advertisements;

[0021] The classified results are integrated into advertisements that are related advertisements to obtain the plurality of advertisements.

[0022] As an embodiment, the classification algorithm used by the traffic screening advertisement is a decision tree, a random forest or a support vector machine.

[0023] As an embodiment, the recalling the multiple advertisements based on the recall model and further screening the multiple advertisements to obtain an advertisement list includes:

[0024] Using a collaborative filtering model to recall the multiple advertisements, to obtain a first candidate advertisement list;

[0025] Recalling the multiple advertisements using a content recommendation model to obtain a second candidate advertisement list;

[0026] Using a deep learning model to perform advertisement recall on the multiple advertisements to obtain a third candidate advertisement list;

[0027] The first candidate advertisement list, the second candidate advertisement list and the third candidate advertisement list are integrated to obtain the advertisement list.

[0028] As an embodiment, the step of sorting at least two advertisements in the advertisement list includes:

[0029] Scoring the at least two advertisements according to the ranking index, and ranking them according to the scores combined with a weighted algorithm to obtain a rough ranking list;

[0030] The advertisements in the rough ranking list are further refined by using a gradient boosting tree model to obtain the ranking result.

[0031] As an embodiment, the ranking index includes historical advertising performance data, advertising delivery environment and user-advertising matching degree.

[0032] As an embodiment, the advertisement quotation request further includes an expected advertisement delivery duration and an expected conversion rate; before sending the advertisement quotation form to the advertiser, the following is further included:

[0033] Performing a weighted evaluation on the advertisement quotation table in combination with the expected advertisement delivery duration and the expected conversion rate;

[0034] The step of sending the advertisement quotation form to the advertiser comprises:

[0035] The weighted advertising quotation list is sent to the advertiser.

[0036] As an embodiment, the advertisement quotation request further includes the application software that the advertiser wants to place advertisements on; before sending the advertisement quotation form to the advertiser, the process further includes:

[0037] Combined with the application software's quotation for advertisement placement, the price of each advertisement in the advertisement quotation table is recalculated;

[0038] The step of sending the advertisement quotation form to the advertiser comprises:

[0039] The advertisement quotation sheet after price calculation is sent to the advertiser.

[0040] In a second aspect, an embodiment of the present application provides an advertisement quotation device, the device comprising:

[0041] A first acquisition module is used to acquire an advertisement quotation request from an advertiser; the advertisement quotation request includes an advertisement type;

[0042] A matching module, used to match the advertisement type corresponding to the advertisement quotation request with an advertisement field, and determine the advertisement field to which the advertisement type belongs;

[0043] A second acquisition module is used to acquire user data and advertisement data in the advertisement field; wherein the user data includes user behavior data and context information corresponding to the advertisement field, and the advertisement data includes historical advertisement delivery data and delivery results corresponding to the advertisement field;

[0044] A screening module, used to use the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field;

[0045] A recall module, configured to recall the multiple advertisements based on the recall model, further screen the multiple advertisements, and obtain an advertisement list; the advertisement list includes at least two advertisements screened from the multiple advertisements;

[0046] A sorting module, used to sort at least two advertisements in the advertisement list, and perform price estimation according to the sorting result to obtain an advertisement quotation table; the advertisement quotation table includes the quotations of the at least two advertisements;

[0047] The sending module is used to send the advertisement quotation form to the advertiser.

[0048] As an embodiment, the screening module is specifically used for:

[0049] Cleaning the user data and the advertisement data to remove noise data and outliers;

[0050] Perform feature extraction on the cleaned user data and advertisement data to obtain feature vectors;

[0051] Inputting the feature vector into the traffic screening model to obtain classification results of each advertisement; the classification results include relevant advertisements and irrelevant advertisements;

[0052] The classified results are integrated into advertisements that are related advertisements to obtain the plurality of advertisements.

[0053] As an embodiment, the classification algorithm used by the traffic screening advertisement is a decision tree, a random forest or a support vector machine.

[0054] As an embodiment, the recall module is specifically used for:

[0055] Using a collaborative filtering model to recall the multiple advertisements, to obtain a first candidate advertisement list;

[0056] Recalling the multiple advertisements using a content recommendation model to obtain a second candidate advertisement list;

[0057] Using a deep learning model to perform advertisement recall on the multiple advertisements to obtain a third candidate advertisement list;

[0058] The first candidate advertisement list, the second candidate advertisement list and the third candidate advertisement list are integrated to obtain the advertisement list.

[0059] As an embodiment, the sorting module is specifically used for:

[0060] Scoring the at least two advertisements according to the ranking index, and ranking them according to the scores combined with a weighted algorithm to obtain a rough ranking list;

[0061] The advertisements in the rough ranking list are further refined by using a gradient boosting tree model to obtain the ranking result.

[0062] As an embodiment, the ranking index includes historical advertising performance data, advertising delivery environment and user-advertising matching degree.

[0063] As an embodiment, the advertisement quotation request further includes expected advertisement delivery time and expected conversion rate; the sorting module is further used to:

[0064] Performing a weighted evaluation on the advertisement quotation table in combination with the expected advertisement delivery duration and the expected conversion rate;

[0065] The sending module is specifically used for:

[0066] The weighted advertising quotation list is sent to the advertiser.

[0067] As an embodiment, the advertisement quotation request further includes the application software for which the advertiser wishes to place advertisements; and the sorting module is further configured to:

[0068] Combined with the application software's quotation for advertisement placement, the price of each advertisement in the advertisement quotation table is recalculated;

[0069] The sending module is specifically used for:

[0070] The advertisement quotation sheet after price calculation is sent to the advertiser.

[0071] In a third aspect, an embodiment of the present application provides a device, comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the advertising bidding method described in any one of the first aspects above.

[0072] In a fourth aspect, an embodiment of the present application provides a computer storage medium, in which a code is stored. When the code is executed, a device executing the code implements the advertising quotation method described in any one of the first aspects above.

[0073] The advertising quotation method and related device provided by this application have the following technical effects:

[0074] 1. Improve efficiency

[0075] Through the automated advertising quotation system, a large number of advertising quotation requests can be processed quickly. The system automatically matches advertising fields, screens traffic, recalls advertisements, and sorts advertisements, reducing the need for manual operations. The system can obtain, process, and feedback advertising quotation forms in real time, allowing advertisers to quickly obtain quotations and adjust advertising strategies in a timely manner.

[0076] 2. Reduce subjectivity and improve decision-making quality

[0077] The system makes accurate pricing based on user data and advertising data (such as user behavior data, advertising history data, etc.), reducing the impact of human subjective judgment. It uses advanced machine learning models and algorithms (such as traffic screening models, recall models, sorting models, etc.) for data analysis and decision-making to improve the scientificity and accuracy of pricing.

[0078] 3. Provide objective and consistent pricing standards

[0079] The automated advertising quotation system ensures that every advertising quotation request is processed through the same process and standards, ensuring the consistency and objectivity of the quotation. The quotation system can be dynamically adjusted according to real-time data and market fluctuations to improve the flexibility and adaptability of the quotation.

[0080] 4. Optimize advertising strategies

[0081] Through advertising field matching and traffic screening models, the system can accurately match advertising types and target users, improving the accuracy of advertising delivery. The system can be integrated with DSP (demand-side platform) to further improve the efficiency and effectiveness of advertising delivery through methods such as real-time bidding (RTB).

[0082] 5. Enhance the experience of advertisers and demanders

[0083] Advertisers can view the advertising quotation table in real time, understand the basis of the quotation, and enhance their trust in the system. By optimizing the advertising delivery strategy, the system can help advertisers obtain higher advertising effects at a lower cost and improve the return on investment (ROI). BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0085] Figure 1 A flowchart of an advertising quotation method provided in an embodiment of the present application;

[0086] Figure 2 A schematic diagram of the structure of an advertising quotation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0087] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. The terms used in the following embodiments are only for the purpose of describing specific embodiments and are not intended to be used as limitations on the present application. As used in the specification and the appended claims of the present application, the singular expressions "one", "a kind", "said", "above", "the" and "this" are intended to also include expressions such as "one or more", unless there is a clear contrary indication in the context. It should also be understood that in the embodiments of the present application, "one or more" refers to one, two or more; "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist; for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0088] like Figure 1 As shown, Figure 1 A flowchart of an advertising quotation method provided in an embodiment of the present application.

[0089] The method includes:

[0090] S101: Obtaining an advertisement quotation request from an advertiser.

[0091] The advertising quote request includes the type of advertisement. An advertising quote request is a request sent by an advertiser (advertiser, demand-side platform, etc.) to an advertising platform or service provider to obtain detailed quote information about advertising placement. This request usually includes the advertiser's specific needs and budget so that the advertising platform can generate a suitable quote plan based on this information. Advertisers submit advertising quote requests through the interface or interface provided by the advertising platform, usually in the form of API calls or web forms. The advertising platform will verify the received request to ensure that all required information is complete and in the correct format.

[0092] Clarify the information that advertisers need to provide when applying for advertising quotes. This information usually includes the type of advertisement, advertising budget, target audience, and advertising duration. The system needs to understand these requirements in order to receive and process quotation requests.

[0093] Advertisement quotation requests occupy a key position in the advertising process. They are an important bridge for communication between advertising platforms and advertisers. Accurate and comprehensive ad quotation requests can help advertising platforms generate reasonable quotations to ensure the effectiveness and cost-effectiveness of advertising, while also helping advertisers better understand market conditions and optimize advertising strategies.

[0094] S102: Perform advertising domain matching on the advertisement type corresponding to the advertisement quotation request to determine the advertisement domain to which the advertisement type belongs.

[0095] In the advertising management system, the main task of step S102 is to determine the advertising field to which the advertisement belongs based on the advertisement type provided in the advertisement quotation request. The determination of the advertising field will provide a basis for the subsequent advertising delivery strategy and quotation generation. The following are the detailed technical implementation steps:

[0096] First, a mapping table is established to store the correspondence between the ad type and the ad domain to which it belongs. For example, the mapping table records that "video ad" corresponds to "video ad domain", "banner ad" corresponds to "display ad domain", etc. The data of the ad quotation request is obtained from the database or cache of the system. The request data includes detailed information such as ad type, budget, target audience, etc. According to the ad type in the ad quotation request, the pre-defined mapping table is queried to find the corresponding ad domain. For example, according to the "video ad" type in the request, the system finds the corresponding ad domain "video ad domain" from the mapping table. If the corresponding ad domain is found in the mapping table, it is determined as the final result; if the correspondence is not found, a default or error message is returned, indicating that the ad type is invalid or needs to be resubmitted. The matched ad domain information is updated to the record of the ad quotation request. The updated data needs to be stored back in the database to ensure that the subsequent processing links can use this information for accurate delivery and quotation generation. The matching results are recorded in the system for subsequent review and tracking. Once the match is successful, the result will be fed back to the next processing link to continue the process of generating the ad quote; if the match fails, the error log will be recorded, and the advertiser will be prompted to resubmit the request or provide a valid ad type. During the processing, if an ad type that does not exist in the mapping table is encountered, the system should record the log and return the error information to the advertiser as it handles regular exceptions. This information usually prompts the advertiser to enter a valid ad type or resubmit the quote request.

[0097] Through these steps, the system can effectively map the ad type in the ad quotation request to the corresponding ad field. This process ensures the accuracy of the input data and provides a necessary and accurate basis for subsequent ad delivery strategies and quotation generation. It not only improves the efficiency of the system, but also enhances the user experience of advertisers.

[0098] S103: Acquire user data and advertisement data in the advertisement field.

[0099] The user data includes user behavior data and context information corresponding to the advertising field, and the advertising data includes historical advertising delivery data and delivery results corresponding to the advertising field.

[0100] User data includes user behavior data (such as clicks, browsing, purchases, etc.) and contextual information (such as the user's geographic location, device type, time period, etc.). Advertising data includes historical advertising data (such as advertising content, delivery time, budget, etc.) and delivery results (such as click-through rate, conversion rate, ROI, etc.). Determine the data source that stores user behavior data and contextual information in the system. For example, user behavior data can come from website logs, application logs, or third-party data providers. Determine the data source of historical advertising data and delivery results. These data are usually stored in an advertising delivery management system or data warehouse.

[0101] According to the advertising field, extract the behavior data of relevant users from the user behavior data source. For example, for the "video advertising field", it is necessary to extract information such as the number of times the user watched the video, the duration, and the interactive behavior. Obtain the user's contextual information from the context data source, such as geographic location, device type, access time, etc., to fully understand the user's behavior pattern.

[0102] According to the advertising field, extract the historical delivery data of related advertisements from the advertising delivery management system. For example, for the "display advertising field", it is necessary to obtain data such as past display advertising content, delivery time, budget, etc. Extract the delivery results of advertisements from the same data source or other related data sources, including key indicators such as click-through rate, conversion rate, ROI, etc.

[0103] Integrate the extracted user behavior data, context information, historical advertising data and advertising results into a structured data format for subsequent processing and analysis. Store the integrated data in a unified storage system, such as a relational database, data warehouse or big data storage system, to ensure high availability and accessibility of the data.

[0104] The integrated user data and advertising data are fed back to the advertising optimization module and the quotation generation module to support accurate advertising strategy formulation and quotation optimization. The design mechanism ensures that user data and advertising data can be updated in real time or regularly to maintain the timeliness and accuracy of system data.

[0105] It should be noted that user data and advertising data can be obtained through third-party data providers and other channels.

[0106] Through the above steps, the system can fully obtain user data and advertising data related to the advertising field. This process ensures the accuracy and effectiveness of the advertising delivery strategy. By analyzing user behavior data and contextual information, the system can understand user preferences and needs, thereby formulating more accurate delivery strategies. Using historical advertising delivery data and delivery results, the system can evaluate the effectiveness of different advertising strategies and provide reference for future advertising delivery. Based on rich data support, the system can make advertising quotations more scientifically to ensure the maximum return on investment (ROI) of advertising delivery.

[0107] This data-driven approach not only effectively improves the accuracy and effectiveness of advertising, but also enhances advertisers’ trust and satisfaction with the system.

[0108] S104: Using the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field.

[0109] Traffic screening model is an algorithm model used to evaluate and screen target users for advertising. The model combines user data and advertising data, and uses certain rules or machine learning algorithms to screen out the most suitable advertising targets. Models may include the following types:

[0110] Rule engine model: Filters traffic based on a series of predefined rules. Machine learning model: Model trained based on historical data, such as decision tree, random forest, neural network, etc. Hybrid model: Combines the advantages of rule engine and machine learning algorithm.

[0111] User data includes user behavior data and contextual information, such as user click behavior, browsing history, geographic location, device used, access time, etc. Advertising data includes historical advertising delivery data and delivery results, such as ad impressions, click-through rate, conversion rate, ad content, delivery time, etc.

[0112] Before entering the traffic screening model, user data and advertisement data need to be preprocessed, including but not limited to the following operations:

[0113] Data cleaning: remove noise and outliers to ensure data accuracy and reliability.

[0114] Data format conversion: Convert data into the format required by the model, including feature vectorization, normalization, etc.

[0115] Feature selection: Select key features based on model requirements and remove irrelevant or redundant features.

[0116] The pre-processed user data and advertisement data are input into the traffic screening model. The specific steps are as follows:

[0117] Filter the data based on predefined rules. For example, users can be filtered based on their geographic location, device type, online time, etc. The screening process is run on the trained machine learning model to predict which users are most likely to be interested in the ad and calculate the matching score for each user. After the initial screening using the rule engine, the machine learning model is used for further screening and scoring.

[0118] According to the output of the traffic screening model, select multiple ads that best meet the screening criteria from the ad library in the advertising field for matching and recommendation. Sort the ads according to the user matching scores output by the model, and select the ads with the highest scores for delivery. Ensure that the recommended ads have a certain degree of diversity to avoid excessive delivery of a single ad. Generate a recommendation list for the multiple ads that are finally matched as a set of candidate ads for delivery. Record the running process and results of the screening model, including input data, screening rules, matching results, etc., to facilitate subsequent analysis and optimization.

[0119] In one embodiment, the using the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field includes:

[0120] Cleaning the user data and the advertisement data to remove noise data and outliers;

[0121] Perform feature extraction on the cleaned user data and advertisement data to obtain feature vectors;

[0122] Inputting the feature vector into the traffic screening model to obtain classification results of each advertisement; the classification results include relevant advertisements and irrelevant advertisements;

[0123] The classified results are integrated into advertisements that are related advertisements to obtain the plurality of advertisements.

[0124] Specifically, remove noise data and outliers, detect and remove errors or noise data (such as duplicate records, malformed data, etc.) in user behavior data and context information. For example, remove obviously erroneous geographic location data or device information. Detect and remove noise data in historical advertising data and delivery results (such as abnormally high click-through rates or conversion rates, which may be caused by cheating).

[0125] Extract features from the cleaned data. Extract user behavior and context features based on analysis requirements, such as the number of clicks, browsing time, device type, geographic location, access frequency, etc. of each user, to form a user feature vector. Extract features from historical advertising data, such as the click-through rate, conversion rate, number of impressions, and delivery time period of each advertisement, to form an advertising feature vector.

[0126] The extracted user feature vectors and ad feature vectors are input into the pre-trained traffic screening model. The traffic screening model uses these feature vectors for analysis and classification to evaluate the matching degree of each ad with each user. The model outputs the classification results for each ad, identifying whether each ad is "relevant" or "irrelevant". Based on the classification output results of the traffic screening model, the relevant ads are integrated and the items identified as "relevant ads" are extracted from the classification results. All the items of relevant ads are integrated to form a list of ads that are in line with the current advertising field and suitable for delivery.

[0127] The integrated list of related ads is used as the final ad recommendation result for the ad delivery system. The process information and final output of cleaning, feature extraction, model operation and classification results are recorded to ensure traceability and subsequent optimization and adjustment.

[0128] In one embodiment, the classification algorithm used by the traffic screening advertisement is a decision tree, a random forest, or a support vector machine.

[0129] S105: recalling the multiple advertisements based on the recall model, and further screening the multiple advertisements to obtain an advertisement list.

[0130] The advertisement list includes at least two advertisements screened from the multiple advertisements.

[0131] The recall model is used to re-screen the ads that have been initially screened to improve the accuracy of ad recommendations. This model usually refers to information such as the user's historical behavior, interest tags, and contextual features.

[0132] Combine user features and ad features to generate a more comprehensive feature vector. Common features include user features and ad features. User features include interest tags, browsing history, click history, purchase history, region, etc. Ad features include content keywords, categories, past performance (such as click-through rate, conversion rate), target audience, etc. Select appropriate recall algorithms based on the application scenario, such as collaborative filtering (based on user behavior similarity (user-user collaborative filtering) or ad similarity (item-item collaborative filtering)), content recommendation (based on the matching degree between ad content and user interests) and deep learning models (such as DNN, RNN, CNN, etc., for deep feature learning and matching of users and ads).

[0133] The recall model is trained using the user's historical behavior data. The training data may include records of ads that the user has clicked, browsed, or purchased. The ads that have been initially screened are input into the trained recall model to calculate the matching score between each ad and the user. These scores reflect the relevance and potential interest of the ad to the user. The ads are further screened and sorted based on the scores output by the recall model. The ads with the highest scores are selected to form the final ad recommendation list.

[0134] In one embodiment, the recalling the multiple advertisements based on the recall model and further screening the multiple advertisements to obtain an advertisement list includes:

[0135] Using a collaborative filtering model to recall the multiple advertisements, to obtain a first candidate advertisement list;

[0136] Recalling the multiple advertisements using a content recommendation model to obtain a second candidate advertisement list;

[0137] Using a deep learning model to perform advertisement recall on the multiple advertisements to obtain a third candidate advertisement list;

[0138] The first candidate advertisement list, the second candidate advertisement list and the third candidate advertisement list are integrated to obtain the advertisement list.

[0139] First, use the collaborative filtering model to recall advertisements and obtain the first candidate advertisement list.

[0140] The collaborative filtering model makes recommendations based on the similarity between users and ads. Collaborative filtering can be divided into user-based collaborative filtering and item-based collaborative filtering. Collect user behavior data, such as user browsing, clicking, purchasing, etc. Build a user-ad interaction matrix. Calculate the similarity between users or between ads to predict ads that users may be interested in. Based on the similarity score, recommend the most relevant ads to form the first candidate ad list.

[0141] Secondly, use the content recommendation model to recall advertisements and obtain the second candidate advertisement list

[0142] The content recommendation model makes recommendations based on the matching degree between the ad content and the user's interest characteristics. Extract ad content features, such as keywords, categories, descriptions, and images. Extract user interest features, such as the user's browsing history, interest tags, etc. Calculate the ad matching degree by comparing the similarity between the user's interest features and the ad content features. Recommend the most relevant ads based on the matching score to form a second candidate ad list.

[0143] Then, the deep learning model is used to recall advertisements and obtain the third candidate advertisement list.

[0144] Deep learning models (such as deep neural networks) can capture more complex relationships between users and ads, and can usually provide more accurate recommendations. Collect user behavior data and ad feature data. Build deep learning models, such as DNN, CNN, and RNN, and select the model architecture based on specific application requirements. Use historical data to train the model so that it can accurately predict user preferences for ads. Use the trained model to predict the user's interest score for each ad. Based on the predicted interest score, recommend the most relevant ads to form a third candidate ad list.

[0145] Finally, advertisement integration generates the final advertisement list. The candidate advertisement lists obtained from the collaborative filtering model, content recommendation model, and deep learning model are integrated, and the comprehensive score or other fusion strategies are considered to obtain the final advertisement recommendation list. The first candidate advertisement list, the second candidate advertisement list, and the third candidate advertisement list are merged. Duplicate advertisements are removed and sorted according to the combined score or weight. The advertisement with the highest comprehensive score is selected to form the final advertisement recommendation list.

[0146] S106: sorting at least two advertisements in the advertisement list, and performing price estimation according to the sorting result to obtain an advertisement quotation table.

[0147] The advertisement quotation table includes quotations of the at least two advertisements.

[0148] The purpose of sorting is to determine the order in which each ad is displayed. The ranking can be based on a variety of factors, such as relevance score, user interest matching, historical performance of the ad, etc. In general, the total score of each ad is composed of the weighted sum of different factors.

[0149] Get a list of ads that have been comprehensively screened by the recall model. Calculate the comprehensive score of each ad based on factors such as the relevance score, user interest score, historical click-through rate, and conversion rate of each ad. Measure the matching degree between ad content and user interests. Predicted score based on user historical interaction behavior. Such as click-through rate (CTR), conversion rate (CVR), and other indicators. Sort ads according to the comprehensive score and determine the display order.

[0150] The price estimation process is based on the advertiser's bidding strategy and the pricing mechanism of the advertising system, such as cost per click (CPC), cost per thousand impressions (CPM) or cost per acquisition (CPA), and the bids of each advertisement in the ad space are determined by ranking. Select an appropriate bidding mechanism (such as CPC, CPM, CPA, etc.). Commonly used in CPC mechanism, advertisers pay the bid of the second-ranked advertisement plus a certain increment. Dynamic pricing mechanism determines the ranking and price based on real-time auction results. By analyzing the comprehensive score and ranking of each advertisement and combining the bidding strategy, the bid of each advertisement is estimated. The ranking and price estimation results are organized into an advertising quotation table as a reference for advertising placement and revenue analysis. It includes key information such as advertisement ID, advertisement position, comprehensive score, and estimated price. The detailed information of each advertisement is listed in a table to facilitate further analysis and decision-making.

[0151] In one embodiment, the step of sorting at least two advertisements in the advertisement list includes:

[0152] Scoring the at least two advertisements according to the ranking index, and ranking them according to the scores combined with a weighted algorithm to obtain a rough ranking list;

[0153] The advertisements in the rough ranking list are further refined by using a gradient boosting tree model to obtain the ranking result.

[0154] Score at least two ads according to the ranking index, and sort them according to the scores combined with the weighted algorithm to get a rough ranking list. This is the process of preliminary sorting (rough sorting), which aims to quickly sort the ads preliminarily through some simple but effective sorting indicators. Determine the sorting index: Select a set of indicators that can reflect the quality and relevance of the ads, such as click-through rate (CTR), conversion rate (CVR), impressions, and user feature matching.

[0155] Each advertisement is scored according to multiple ranking indicators and can be assigned different weights. According to the comprehensive score of each advertisement, it is sorted from high to low to obtain a rough ranking list.

[0156] The scoring formula is, for example, comprehensive score of advertisement = CTR×weight 1+CVR×weight 2+…+user feature matching degree×weight N.

[0157] The ads in the rough ranking list are further refined using the gradient boosting tree model to obtain the ranking results. The gradient boosting tree model (GBDT) is a powerful machine learning model that can handle nonlinear relationships and capture complex interactions between features. By further refining the rough ranking list, the accuracy of the ranking can be improved.

[0158] Construct a feature vector for each ad. These features may include the indicators used in the rough sorting stage and other potential influencing factors (such as the user's real-time behavior, the historical performance of the ad, etc.).

[0159] Common features include user characteristics (age, gender, interest tags and historical behavior), advertising characteristics (ad content, delivery time and target audience) and interaction characteristics (user interaction history with advertising).

[0160] Use historical data (such as click and conversion records) to train the gradient boosting tree model to learn the relationship between features and ad priorities. Include ad features, user features, and ad effect (such as click behavior) labels. Input the ads in the rough ranking list into the trained gradient boosting tree model to predict the exact score of the ads. The final ranking score of each ad will be adjusted based on the prediction results of the gradient boosting tree model. Re-rank the ads based on the predicted scores of the gradient boosting tree model to get the ranking results.

[0161] In one embodiment, the ranking index includes historical advertising performance data, advertising delivery environment, and user-advertising matching degree.

[0162] S107: Send the advertisement quotation form to the advertiser.

[0163] In one embodiment, the advertisement quotation request further includes an expected advertisement delivery duration and an expected conversion rate; before sending the advertisement quotation form to the advertiser, the process further includes:

[0164] Performing a weighted evaluation on the advertisement quotation table in combination with the expected advertisement delivery duration and the expected conversion rate;

[0165] The step of sending the advertisement quotation form to the advertiser comprises:

[0166] The weighted advertising quotation list is sent to the advertiser.

[0167] In addition to the comprehensive score and estimated price of the advertisement itself, the following additional parameters are included in the advertising quotation request: expected advertising duration (the duration of the planned advertising delivery) and expected conversion rate (the expected conversion success rate after the advertising delivery). In order to improve the accuracy and rationality of the quotation, it is necessary to weight the advertisement in combination with the expected advertising duration and the expected conversion rate. The specific steps are: Calculate the expected revenue of the advertisement based on the expected conversion rate and delivery duration. The value of a single conversion can be estimated based on historical data or industry averages.

[0168] Expected revenue formula: Expected revenue = expected advertising duration × expected conversion rate × single conversion value

[0169] Considering the expected duration and conversion rate of the ad, the overall score of the ad is weighted and adjusted.

[0170] Adjustment formula: Comprehensive score after adjustment = original comprehensive score × (1 + weight_duration × delivery duration + weight_conversion rate × conversion rate)

[0171] Weight_duration and weight_conversion rate need to be set according to actual conditions and business strategies.

[0172] The advertising quotation sheet after weighted valuation is sorted and updated to ensure that it contains the latest estimated information. Then the advertising quotation sheet after weighted valuation is sent to the advertiser.

[0173] In one embodiment, the advertisement quotation request further includes the application software that the advertiser wants to place advertisements on; before sending the advertisement quotation form to the advertiser, the process further includes:

[0174] Combined with the application software's quotation for advertisement placement, the price of each advertisement in the advertisement quotation table is recalculated;

[0175] The step of sending the advertisement quotation form to the advertiser comprises:

[0176] The advertisement quotation sheet after price calculation is sent to the advertiser.

[0177] In addition to the expected duration of the advertisement and the expected conversion rate, the advertisement quotation request also includes the following additional parameters: the application software where the advertiser expects the advertisement to be placed (specifically, which application platforms or software the advertiser expects to place the advertisement on).

[0178] In order to further refine the quotation, it is necessary to recalculate the quotation in combination with the application software that the advertiser wants to place the advertisement. The specific steps are as follows:

[0179] Determine the quotation coefficient of each application software: Determine the quotation coefficient of each application software based on factors such as market value and user coverage. For example: the quotation coefficient of application A = 1.2; the quotation coefficient of application B = 1.0; the quotation coefficient of application C = 0.8. Apply the quotation coefficient to adjust the price: adjust the initial quotation of the advertisement based on the application software selected by the advertiser.

[0180] Adjustment formula: Final quotation = preliminary quotation × application software quotation coefficient

[0181] The advertising quotation sheet after price calculation is collated and updated to ensure that it contains the latest estimated information. The updated advertising quotation sheet is then sent to the advertiser.

[0182] The above are some specific implementations of the advertising quotation method provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization.

[0183] See also Figure 2 The structure diagram of the advertisement quotation device shown in the figure includes a first acquisition module 201, a matching module 202, a second acquisition module 203, a screening module 204, a recall module 205, a sorting module 206 and a sending module 207.

[0184] The first acquisition module 201 is used to acquire an advertisement quotation request from an advertiser; the advertisement quotation request includes an advertisement type;

[0185] A matching module 202, configured to match the advertisement type corresponding to the advertisement quotation request with an advertisement domain, and determine the advertisement domain to which the advertisement type belongs;

[0186] The second acquisition module 203 is used to acquire user data and advertisement data in the advertisement field; wherein the user data includes user behavior data and context information corresponding to the advertisement field, and the advertisement data includes historical advertisement delivery data and delivery results corresponding to the advertisement field;

[0187] A screening module 204, configured to use the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field;

[0188] A recall module 205 is used to recall the multiple advertisements based on the recall model, and further screen the multiple advertisements to obtain an advertisement list; the advertisement list includes at least two advertisements screened from the multiple advertisements;

[0189] A sorting module 206 is used to sort at least two advertisements in the advertisement list, and perform price estimation according to the sorting result to obtain an advertisement quotation table; the advertisement quotation table includes the quotations of the at least two advertisements;

[0190] The sending module 207 is used to send the advertisement quotation form to the advertiser.

[0191] As an embodiment, the screening module is specifically used for:

[0192] Cleaning the user data and the advertisement data to remove noise data and outliers;

[0193] Perform feature extraction on the cleaned user data and advertisement data to obtain feature vectors;

[0194] Inputting the feature vector into the traffic screening model to obtain classification results of each advertisement; the classification results include relevant advertisements and irrelevant advertisements;

[0195] The classified results are integrated into advertisements that are related advertisements to obtain the plurality of advertisements.

[0196] As an embodiment, the classification algorithm used by the traffic screening advertisement is a decision tree, a random forest or a support vector machine.

[0197] As an embodiment, the recall module is specifically used for:

[0198] Using a collaborative filtering model to recall the multiple advertisements, to obtain a first candidate advertisement list;

[0199] Recalling the multiple advertisements using a content recommendation model to obtain a second candidate advertisement list;

[0200] Using a deep learning model to perform advertisement recall on the multiple advertisements to obtain a third candidate advertisement list;

[0201] The first candidate advertisement list, the second candidate advertisement list and the third candidate advertisement list are integrated to obtain the advertisement list.

[0202] As an embodiment, the sorting module is specifically used for:

[0203] Scoring the at least two advertisements according to the ranking index, and ranking them according to the scores combined with a weighted algorithm to obtain a rough ranking list;

[0204] The advertisements in the rough ranking list are further refined by using a gradient boosting tree model to obtain the ranking result.

[0205] As an embodiment, the ranking index includes historical advertising performance data, advertising delivery environment and user-advertising matching degree.

[0206] As an embodiment, the advertisement quotation request further includes expected advertisement delivery time and expected conversion rate; the sorting module is further used to:

[0207] Performing a weighted evaluation on the advertisement quotation table in combination with the expected advertisement delivery duration and the expected conversion rate;

[0208] The sending module is specifically used for:

[0209] The weighted advertising quotation list is sent to the advertiser.

[0210] As an embodiment, the advertisement quotation request further includes the application software for which the advertiser wishes to place advertisements; and the sorting module is further configured to:

[0211] Combined with the application software's quotation for advertisement placement, the price of each advertisement in the advertisement quotation table is recalculated;

[0212] The sending module is specifically used for:

[0213] The advertisement quotation sheet after price calculation is sent to the advertiser.

[0214] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0215] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes an advertising quotation method described in any embodiment of the present application.

[0216] The computer storage medium stores codes, and when the codes are executed, a device executing the codes implements an advertisement quotation method described in any embodiment of the present application.

[0217] The "first" and "second" (if any) in the names mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0218] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0219] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0220] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. An advertising quotation method, characterized in that: include: Obtain advertising quotation requests from advertisers; The advertisement quotation request includes an advertisement type; Performing advertising domain matching on the advertising type corresponding to the advertising quotation request to determine the advertising domain to which the advertising type belongs; Acquire user data and advertisement data of the advertisement domain; wherein the user data includes user behavior data and context information corresponding to the advertisement domain, and the advertisement data includes historical advertisement delivery data and delivery results corresponding to the advertisement domain; Using the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement domain; Recalling the multiple advertisements based on the recall model, further screening the multiple advertisements to obtain an advertisement list; the advertisement list includes at least two advertisements screened from the multiple advertisements; Sorting at least two advertisements in the advertisement list, and performing price estimation according to the sorting result to obtain an advertisement quotation table; the advertisement quotation table includes quotations of the at least two advertisements; The advertisement quotation form is sent to the advertiser.

2. The method according to claim 1, characterized in that The step of using the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field includes: Cleaning the user data and the advertisement data to remove noise data and outliers; Perform feature extraction on the cleaned user data and advertisement data to obtain feature vectors; Inputting the feature vector into the traffic screening model to obtain classification results of each advertisement; the classification results include relevant advertisements and irrelevant advertisements; The classified results are integrated into advertisements that are related advertisements to obtain the plurality of advertisements.

3. The method according to claim 1, characterized in that The classification algorithm used by the traffic screening advertisement is a decision tree, a random forest or a support vector machine.

4. The method according to claim 1, characterized in that: The step of recalling the multiple advertisements based on the recall model and further screening the multiple advertisements to obtain an advertisement list includes: Using a collaborative filtering model to recall the multiple advertisements, to obtain a first candidate advertisement list; Recalling the multiple advertisements using a content recommendation model to obtain a second candidate advertisement list; Using a deep learning model to perform advertisement recall on the multiple advertisements to obtain a third candidate advertisement list; The first candidate advertisement list, the second candidate advertisement list and the third candidate advertisement list are integrated to obtain the advertisement list.

5. The method according to claim 1, characterized in that The step of sorting at least two advertisements in the advertisement list comprises: Scoring the at least two advertisements according to the ranking index, and ranking them according to the scores combined with a weighted algorithm to obtain a rough ranking list; The advertisements in the rough ranking list are further refined by using a gradient boosting tree model to obtain the ranking result.

6. The method according to claim 5, characterized in that The ranking indicators include historical advertising performance data, advertising delivery environment and user-advertising matching degree.

7. The method according to claim 1, characterized in that The advertisement quotation request also includes the expected advertisement delivery time and the expected conversion rate; before sending the advertisement quotation form to the advertiser, the process also includes: Performing a weighted evaluation on the advertisement quotation table in combination with the expected advertisement delivery duration and the expected conversion rate; The step of sending the advertisement quotation form to the advertiser comprises: The weighted advertising quotation list is sent to the advertiser.

8. The method according to claim 1, characterized in that The advertisement quotation request also includes the application software that the advertiser wants to place advertisements on; before sending the advertisement quotation form to the advertiser, the process further includes: Combined with the application software's quotation for advertisement placement, the price of each advertisement in the advertisement quotation table is recalculated; The step of sending the advertisement quotation form to the advertiser comprises: The advertisement quotation sheet after price calculation is sent to the advertiser.

9. An advertising quotation device, characterized in that: include: A first acquisition module is used to acquire an advertisement quotation request from an advertiser; The advertisement quotation request includes an advertisement type; A matching module, used to match the advertisement type corresponding to the advertisement quotation request with an advertisement field, and determine the advertisement field to which the advertisement type belongs; A second acquisition module is used to acquire user data and advertisement data in the advertisement field; wherein the user data includes user behavior data and context information corresponding to the advertisement field, and the advertisement data includes historical advertisement delivery data and delivery results corresponding to the advertisement field; A screening module, used to use the user data and the advertisement data as inputs of a traffic screening model to obtain a plurality of advertisements belonging to the advertisement field; A recall module, configured to recall the multiple advertisements based on the recall model, further screen the multiple advertisements, and obtain an advertisement list; the advertisement list includes at least two advertisements screened from the multiple advertisements; A sorting module, used to sort at least two advertisements in the advertisement list, and perform price estimation according to the sorting result to obtain an advertisement quotation table; the advertisement quotation table includes the quotations of the at least two advertisements; The sending module is used to send the advertisement quotation form to the advertiser.

10. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the advertising quotation method according to any one of claims 1 to 8 is implemented.

Images