Advertisement recommendation method and device based on artificial intelligence and storage medium
By building a feature quantification benchmark model and continuous data evaluation, the problem of low advertising recommendation accuracy is solved, and efficient advertising recommendations that dynamically adapt to changes in user interests are achieved.
Patent Information
- Application Number
- CN202510717070.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-30
- Publication Date
- 2025-09-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing advertising recommendation methods rely on manually set rules and are difficult to adapt to rapidly changing user behavior and market environment, resulting in low advertising recommendation accuracy and serious waste of resources.
Through an AI-based advertising recommendation method, a feature quantification benchmark model is constructed, key features are extracted using a neural network model, user basic data and advertising data are compared, the recommendation value is calculated, the best advertisement is selected for recommendation, and the recommendation quality is evaluated through continuous data acquisition.
It improves the accuracy of advertising recommendations, dynamically captures changes in user interests, optimizes recommendation strategies, and reduces resource waste.
Smart Images
Figure CN120634645A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of advertising push, and specifically relates to an artificial intelligence-based advertising recommendation method, device and storage medium. Background Art
[0002] With the continuous expansion of the Internet user base and the popularization of mobile devices, the digital advertising market has shown explosive growth. Early advertising recommendations mainly relied on manually set rules. For example, based on basic attributes such as user location, age, and gender, advertising delivery groups were divided to make fixed recommendations. However, due to factors such as users' cyclical preferences or product popularity, advertising recommendations are subject to many variables. The formulation of simple rules relies more on the experience of domain experts and is difficult to adapt to rapidly changing user behavior and market environments. In addition, this method can only process limited explicit features and cannot tap into users' potential interests and preferences, resulting in low accuracy of advertising recommendations and a large amount of advertising resources being wasted on non-target user groups. Summary of the Invention
[0003] The purpose of the present invention is to provide an artificial intelligence-based advertising recommendation method, device and storage medium to solve the problems faced in the above-mentioned background technology.
[0004] The purpose of the present invention can be achieved through the following technical solutions:
[0005] An artificial intelligence-based advertising recommendation method, the recommendation method comprising:
[0006] Step 1: Collect data information related to advertising recommendations from multiple data terminals, including advertising data information to be pushed and basic user data information;
[0007] Step 2: Storing and processing the collected data information;
[0008] Step 3: Based on a large amount of historical data information, key features are extracted and a feature quantification benchmark model is trained using a neural network model;
[0009] Step 4: Extract key features based on the advertising data information and input them into the feature quantification benchmark model to obtain the advertising feature quantification model;
[0010] Step 5: Based on the basic data information, extract key features and input them into the feature quantification benchmark model to obtain the user's basic feature quantification model;
[0011] Step 6: Using the user's basic feature quantification model as a benchmark, compare it with the advertising feature quantification model of each advertisement to calculate the recommendation value of each advertisement, and select the advertisement with the highest recommendation value for recommendation;
[0012] Step 7: After making an ad recommendation, continue to obtain corresponding data parameters to evaluate this recommendation.
[0013] Furthermore, the operation method of step three is:
[0014] Based on the historical advertising effectiveness, multiple key features are extracted. Each key feature is set with a quantitative scoring value of 0-9 levels. Each quantitative scoring value is set with corresponding data information content. Through continuous training and optimization of the neural network model, a feature quantitative benchmark model is obtained.
[0015] Furthermore, the operation method of step 4 is:
[0016] According to the data information of each advertisement in the advertisement library, key features are extracted, and the data information of each key feature of the advertisement is input into the feature quantification benchmark model to obtain the quantitative value of each key feature of the advertisement, thereby forming an advertisement feature quantification model A=(TG1, ..., TG i ,…,TG n ), where TG i is the quantitative value of the i-th key feature of the recommended advertisement, n is the number of key features, and i∈[1,n].
[0017] Furthermore, the operation method of step five is:
[0018] Based on the user's basic data information, the corresponding key features are extracted, and the data information of each key feature of the user is input into the feature quantization benchmark model, thereby obtaining the user's basic feature quantization model B = (TF1, ..., TF i ,…,TF n ), where TF i is the quantitative value of the i-th key feature of the recommended user.
[0019] Furthermore, the operation method of step six is:
[0020] Based on the advertising feature quantification model A and the basic feature quantification model B, we get the model
[0021] C=(|TF1-TG1|,…,|TF i -TG i |,…,|TF n -TG n |), so that by the formula Obtain the recommendation value H of each recommended advertisement;
[0022] According to the size of the recommendation value, the advertisements to be recommended are sorted from large to small, and the advertisement with the highest recommendation value is selected for recommendation.
[0023] Furthermore, the processing operations in step 2 include data cleaning and data conversion operations, wherein data cleaning includes outlier processing, missing value filling, data deduplication and standardization processing, and the data conversion includes normalization and standardization processing, discrete coding processing, and text vectorization processing.
[0024] Furthermore, the operation method of step seven is:
[0025] After making an ad recommendation, the click-through rate and conversion rate of the recommended ad are obtained within a period of Δt, so as to formulate the function of click-through rate variation over period CTR(t) and the function of conversion rate variation over time CR(t). At the same time, according to the ad content, the sales volume N of the corresponding product is obtained. U and the rate of profit, Pr;
[0026] By formula
[0027] Obtain the evaluation value K;
[0028] When K>K2, it is judged that the quality of this ad recommendation is good;
[0029] When K1≤K≤K2, the quality of this ad recommendation is considered normal;
[0030] When K<K1, it is judged that the quality of this ad recommendation is poor;
[0031] Where a1 and a2 are weight coefficients, and a1+a2=1, t1 is the start point of Δt time, t2 is the end point of Δt time, N U0 is the sales volume of the product when the ad is not recommended, Pr0 is the profit margin of the product when the ad is not recommended, ΔCTR is the click-through rate comparison value, ΔCR is the conversion rate comparison value, ΔN U is the sales volume comparison value, ΔPr is the profit rate comparison value, K1 and K2 are the evaluation judgment threshold coefficients.
[0032] An artificial intelligence-based advertising recommendation device includes a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the artificial intelligence-based advertising recommendation method.
[0033] An artificial intelligence-based advertising recommendation storage medium is provided, wherein the storage medium is a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the artificial intelligence-based advertising recommendation method is implemented.
[0034] Beneficial effects of the present invention:
[0035] The present invention uses artificial intelligence technology to construct a feature quantification benchmark model, and finds the most suitable advertisement for the user for push based on the similarity between the quantified values of each key feature of the recommended advertisement and the quantified values of the key features of the user, thereby improving the push accuracy. It can derive different quantified values according to the corresponding key features based on factors such as the user's interests and hobbies at different times, and can dynamically capture the changes in the user's interests at different times, making the advertisement recommendation more in line with the user's real-time needs.
[0036] After making an advertisement recommendation, the present invention continuously obtains corresponding data parameters, such as the click-through rate, conversion rate, sales volume and profit margin of the advertisement, and conducts a comprehensive analysis to judge the quality of the advertisement recommendation, so as to better optimize the advertisement recommendation strategy.
[0037] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0039] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION
[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.
[0041] In one embodiment, an artificial intelligence-based advertising recommendation method is disclosed, such as Figure 1 As shown, recommended methods include:
[0042] Step 1: Collect data information related to advertising recommendations from multiple data terminals, including advertising data information to be pushed and basic user data information;
[0043] Step 2: Storing and processing the collected data information;
[0044] Step 3: Based on a large amount of historical data information, key features are extracted and a feature quantification benchmark model is trained using a neural network model;
[0045] Step 4: Extract key features based on the advertising data information and input them into the feature quantification benchmark model to obtain the advertising feature quantification model;
[0046] Step 5: Based on the basic data information, extract key features and input them into the feature quantification benchmark model to obtain the user's basic feature quantification model;
[0047] Step 6: Using the user's basic feature quantification model as a benchmark, compare it with the advertising feature quantification model of each advertisement to calculate the recommendation value of each advertisement, and select the advertisement with the highest recommendation value for recommendation;
[0048] Step 7: After making an ad recommendation, continue to obtain corresponding data parameters to evaluate this recommendation.
[0049] Through the above technical solution, the present application collects data information related to advertising recommendations from multiple data terminals, including advertising data information to be pushed and basic data information of users. For example, the advertising data information includes information such as the content of the advertisement to be pushed, the target audience, the advertising spokesperson, and the duration. The basic data information includes data information such as the user's consumption age, gender, historical browsing behavior, purchase history, favorites, and active time period. Then, a feature quantification benchmark model is automatically constructed through artificial intelligence technology. The recommendation value of each advertisement is obtained based on the comparison of the basic data information of the user and the advertising data information, so as to select the advertisement with the highest recommendation value for recommendation. In this way, the present application can construct a feature quantification benchmark model through artificial intelligence technology. Based on the similarity between the quantified values of each key feature of the advertisement to be recommended and the quantified values of the key features of the user, the best advertisement suitable for the user can be found for push, thereby improving the push accuracy. Moreover, different quantified values can be obtained based on factors such as the user's interests and hobbies at different times, which can dynamically capture the changes in the user's interests at different times, making the advertising recommendation more in line with the user's real-time needs. In addition, after the advertisement recommendation is made, the corresponding data parameters are continuously obtained, such as the click-through rate, conversion rate, and the sales volume N of the corresponding product. U And the profit margin Pr, so as to conduct a comprehensive analysis to judge the quality of this advertising recommendation, so as to better optimize the advertising recommendation strategy.
[0050] The processing operations in step 2 include data cleaning and data conversion operations, where data cleaning includes outlier processing, missing value filling, data deduplication and standardization processing, and data conversion includes normalization and standardization processing, discrete coding processing, and text vectorization processing.
[0051] First, by processing the data for outliers, invalid data (such as robot clicks, repeated submissions) can be filtered out, and erroneous records (such as abnormally high purchase amounts) can be corrected. Missing value filling can be used to fill in missing user attributes (such as unfilled age) with mean filling, mode filling, or model-based prediction filling. Deduplication and standardization can unify the data format (such as converting timestamps to standard time) and remove duplicate records (such as the same user clicking on the same ad multiple times). The normalization and standardization steps in data conversion can normalize numerical features (such as number of clicks, length of stay), thereby eliminating dimensional differences and facilitating subsequent analysis. Discrete coding can divide continuous features (such as age) into discrete categories (such as "18-25 years old" and "26-35 years old") to facilitate model processing. Text vectorization can convert text such as advertising copy and user comments into numerical vectors through TF-IDF or Word2Vec to facilitate subsequent processing.
[0052] The operation method of step three is to extract multiple key features based on historical advertising effectiveness. Each key feature is set with a quantitative score value of 0-9 levels. Each quantitative score value is set with corresponding data information content. Through continuous training and optimization of the neural network model, a feature quantitative benchmark model is obtained.
[0053] The operation method of step 4 is as follows: extract key features based on the data information of each advertisement in the advertisement library, input the data information of each key feature of the advertisement into the feature quantification benchmark model, thereby obtaining the quantitative value of each key feature of the advertisement, and thus forming the advertisement feature quantification model A=(TG1, ..., TG i ,…,TG n ), where TG i is the quantitative value of the i-th key feature of the recommended advertisement, n is the number of key features, and i∈[1,n];
[0054] The operation method of step five is: based on the user's basic data information, the corresponding key features are extracted, and the data information of each key feature of the user is input into the feature quantification benchmark model to obtain the user's basic feature quantification model.
[0055] B=(TF1,…,TF i ,…,TF n ), where TF i is the quantitative value of the i-th key feature of the recommended user;
[0056] The operation method of step six is: based on the advertising feature quantification model A and the basic feature quantification model B, obtain the model
[0057] C=(|TF1-TG1|,…,|TFi -TG i |,…,|TF n -TG n |), so that by the formula Obtain the recommendation value H of each recommended advertisement;
[0058] According to the size of the recommendation value, the advertisements to be recommended are sorted from large to small, and the advertisement with the highest recommendation value is selected for recommendation.
[0059] The above scheme provides a specific method for advertising recommendation. First, based on the historical advertising effectiveness, multiple key features are extracted. Each key feature is set with a quantitative rating value of 0-9 levels. Each quantitative rating value is set with corresponding data information content. Through continuous training and optimization of the neural network model, a feature quantitative benchmark model is obtained. Then, based on the data information of each advertisement in the advertisement library, key features are extracted. According to the data information of each key feature of the advertisement, it is input into the feature quantitative benchmark model to obtain the quantitative value of each key feature of the advertisement, thereby forming an advertisement feature quantitative model.
[0060] A=(TG1,…,TG i ,…,TG n ), where TG i is the quantitative value of the i-th key feature of the recommended advertisement, n is the number of key features, and i∈[1,n]; based on the user's basic data information, the corresponding key features are extracted, and the data information of each key feature of the user is input into the feature quantification benchmark model, thereby obtaining the user's basic feature quantization model B=(TF1,…,TF i ,…,TF n ), where TF i is the quantitative value of the i-th key feature of the recommended user; finally, based on the advertising feature quantification model A and the basic feature quantification model B, the model
[0061] C=(|TF1-TG1|,…,|TF i -TG i |,…,|TF n -TG n |), so that by the formula Obtain the recommendation value H of each recommended advertisement, sort each advertisement to be recommended from large to small according to the size of the recommendation value, and select the advertisement with the highest recommendation value for recommendation. For example, taking sports shoes as an example, construct key features such as amount, age, and quality. Each key feature is set with a quantitative rating value of 0-9 levels. Each quantitative rating value is set with corresponding data information content. Then, based on the advertising data information of the sports shoes, extract the data information under the corresponding key feature to obtain the corresponding advertising feature quantitative model. Then, based on the data information carried by each key feature of the user, obtain the user's basic feature quantitative model. Each key feature of this user and the advertisement to be recommended has a corresponding quantitative value. Then, through the quantitative difference of each key feature of the two, the formula is used. By deriving the recommendation value H for each recommended ad, it can be seen that the smaller the difference, that is, the larger the recommendation value, the higher the match between the user and the corresponding recommended ad. This allows the ad with the highest recommendation value to be selected for recommendation and precise push notification. This method can generate different quantitative values based on user data changes over time, thereby recommending ads that better meet the user's real-time needs and increase accuracy. It can also identify potential users based on data parameters such as user behavior habits, make ad recommendations, and ensure the rational use of advertising resources. For example, based on characteristics such as user browsing rate and keyword search, the corresponding quantitative value can be derived, and appropriate ads can be recommended based on the quantitative value.
[0062] The operation method of step seven is: after making an advertisement recommendation, obtain the click-through rate and conversion rate of the recommended advertisement within a period of Δt, so as to formulate the function of click-through rate variation over period CTR(t) and the function of conversion rate variation over time CR(t), and at the same time, obtain the sales volume N of the corresponding product according to the advertisement content. U and the rate of profit, Pr;
[0063] By formula
[0064] Obtain the evaluation value K;
[0065] When K>K2, it is judged that the quality of this ad recommendation is good;
[0066] When K1≤K≤K2, the quality of this ad recommendation is considered normal;
[0067] When K<K1, it is judged that the quality of this ad recommendation is poor;
[0068] Where a1 and a2 are weight coefficients, and a1+a2=1, t1 is the start point of Δt time, t2 is the end point of Δt time, N U0is the sales volume of the product when the ad is not recommended, Pr0 is the profit margin of the product when the ad is not recommended, ΔCTR is the click-through rate comparison value, ΔCR is the conversion rate comparison value, ΔN U is the sales volume comparison value, ΔPr is the profit rate comparison value, K1 and K2 are the evaluation judgment threshold coefficients.
[0069] The above technical solution provides a specific method for evaluating the quality of advertising recommendations. Generally speaking, after an advertisement is pushed, the sales volume and profit of the corresponding product will increase, and the click-through rate and the corresponding conversion rate will also increase. Therefore, after the corresponding advertisement is recommended to multiple users, the click-through rate and conversion rate of the recommended advertisement are obtained within a continuous time of Δt, thereby formulating the function of click-through rate variation with period CTR(t) and the function of conversion rate variation with time CR(t). At the same time, according to the advertisement content, the sales volume N of the corresponding product is obtained. U And the profit rate Pr, through the formula
[0070] The evaluation value K is obtained, a1 and a2 are weight coefficients, determined according to experience, a1+a2=1, t1 is the starting point of Δt time, t2 is the end point of Δt time, N U0 is the sales volume of the product when the ad is not recommended, Pr0 is the profit margin of the product when the ad is not recommended, ΔCTR is the click-through rate comparison value, ΔCR is the conversion rate comparison value, ΔN U is the sales volume comparison value, ΔPr is the profit rate comparison value, K1 and K2 are the evaluation threshold coefficients, which can be determined based on historical data and empirical data. It represents the cumulative change of the click-through rate and conversion rate of the advertisement in Δt time. High click-through rate and high conversion rate indicate that the quality of the advertisement push is good. It indicates the change in sales volume and profit margin after the ad push compared to before the ad push. Obviously, the larger the value, the better the quality of the ad push. To facilitate judgment, two evaluation threshold coefficients K1 and K2 are set based on experience. When K>K2, the quality of the ad recommendation is judged to be good. When K1≤K≤K2, the quality of the ad recommendation is judged to be normal. When K<K1, the quality of the ad recommendation is judged to be poor. In this way, a comprehensive analysis can be conducted based on parameters such as click-through rate, conversion rate, profit margin, and sales volume after the ad push to evaluate the quality of the ad push, so as to better optimize the ad recommendation strategy.
[0071] An artificial intelligence-based advertising recommendation device includes a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the above-mentioned artificial intelligence-based advertising recommendation method.
[0072] An artificial intelligence-based advertising recommendation storage medium is provided, wherein the storage medium is a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned artificial intelligence-based advertising recommendation method is implemented.
[0073] The above content is merely an example and explanation of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.
Claims
1. An artificial intelligence-based advertising recommendation method, characterized in that: The recommended methods include: Step 1: Collect data information related to advertising recommendations from multiple data terminals, including advertising data information to be pushed and basic user data information; Step 2: Storing and processing the collected data information; Step 3: Based on a large amount of historical data information, key features are extracted and a feature quantification benchmark model is trained using a neural network model; Step 4: Extract key features based on the advertising data information and input them into the feature quantification benchmark model to obtain the advertising feature quantification model; Step 5: Based on the basic data information, extract key features and input them into the feature quantification benchmark model to obtain the user's basic feature quantification model; Step 6: Using the user's basic feature quantification model as a benchmark, compare it with the advertising feature quantification model of each advertisement to calculate the recommendation value of each advertisement, and select the advertisement with the highest recommendation value for recommendation; Step 7: After making an ad recommendation, continue to obtain corresponding data parameters to evaluate this recommendation.
2. The artificial intelligence-based advertising recommendation method according to claim 1 is characterized in that: The operation method of step three is: Based on the historical advertising effectiveness, multiple key features are extracted. Each key feature is set with a quantitative scoring value of 0-9 levels. Each quantitative scoring value is set with corresponding data information content. Through continuous training and optimization of the neural network model, a feature quantitative benchmark model is obtained.
3. The artificial intelligence-based advertising recommendation method according to claim 2, characterized in that: The operation method of step 4 is: According to the data information of each advertisement in the advertisement library, key features are extracted, and the data information of each key feature of the advertisement is input into the feature quantification benchmark model to obtain the quantitative value of each key feature of the advertisement, thereby forming an advertisement feature quantification model A=(TG1, ..., TG i ,…,TG n ), where TG i is the quantitative value of the i-th key feature of the recommended advertisement, n is the number of key features, and i∈[1,n].
4. The artificial intelligence-based advertising recommendation method according to claim 3, characterized in that: The operation method of step five is: Based on the user's basic data information, the corresponding key features are extracted, and the data information of each key feature of the user is input into the feature quantification benchmark model to obtain the user's basic feature quantification model B=(TF1,…,TF i ,…,TF n ), where TF i is the quantitative value of the i-th key feature of the recommended user.
5. The artificial intelligence-based advertising recommendation method according to claim 4, characterized in that: The operation method of step six is: Based on the advertising feature quantification model A and the basic feature quantification model B, we get the model C=(|TF1-TG1|,…,|TF i -TG i |,…,|TF n -TG n |), so that by the formula Obtain the recommendation value H of each recommended advertisement; According to the size of the recommendation value, the advertisements to be recommended are sorted from large to small, and the advertisement with the highest recommendation value is selected for recommendation.
6. The artificial intelligence-based advertising recommendation method according to claim 5, characterized in that: The processing operations in step 2 include data cleaning and data conversion operations, wherein data cleaning includes outlier processing, missing value filling, data deduplication and standardization processing, and the data conversion includes normalization and standardization processing, discretization coding processing, and text vectorization processing.
7. The artificial intelligence-based advertising recommendation method according to claim 6, characterized in that: The operation method of step seven is: After making an ad recommendation, the click-through rate and conversion rate of the recommended ad are obtained within a period of Δt, so as to formulate the function of click-through rate variation over period CTR(t) and the function of conversion rate variation over time CR(t). At the same time, according to the ad content, the sales volume N of the corresponding product is obtained. U and the rate of profit, Pr; By formula Obtain the evaluation value K; When K>K2, it is judged that the quality of this ad recommendation is good; When K1≤K≤K2, the quality of this ad recommendation is considered normal; When K<K1, it is judged that the quality of this ad recommendation is poor; Where a1 and a2 are weight coefficients, and a1+a2=1, t1 is the start point of Δt time, t2 is the end point of Δt time, N U0 is the sales volume of the product when the ad is not recommended, Pr0 is the profit margin of the product when the ad is not recommended, ΔCTR is the click-through rate comparison value, ΔCR is the conversion rate comparison value, ΔN U is the sales volume comparison value, ΔPr is the profit rate comparison value, K1 and K2 are the evaluation judgment threshold coefficients.
8. An artificial intelligence-based advertising recommendation device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the artificial intelligence-based advertising recommendation method according to any one of claims 1 to 7.
9. An artificial intelligence-based advertising recommendation storage medium, characterized in that: The storage medium is a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the artificial intelligence-based advertising recommendation method according to any one of claims 1 to 7 is implemented.