E-commerce advertisement pushing method and system

By calculating the comprehensive preference coefficient of user interest and behavior data, analyzing product similarity, and dynamically adjusting the advertising push strategy, the problem of lack of accuracy and low efficiency of advertising push in the existing technology is solved, and the click rate of e-commerce advertising is improved. Sales value.

CN119006064APending Publication Date: 2024-11-22张璇
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411016344.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-27
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing e-commerce advertising push methods fail to fully consider the user's preference for products and the specific return rate of advertising push and the strategy correlation, resulting in a lack of accuracy and low efficiency in advertising push.

Method used

By obtaining user interest data and behavior data, calculating the comprehensive preference coefficient, analyzing product similarity, and dynamically adjusting the push time and number of times according to the advertising return rate to achieve personalized advertising recommendations and resource optimization.

Benefits of technology

It improves the click-through rate and conversion rate of advertising, enhances user experience, optimizes the efficiency and sales value of advertising push, reduces resource waste, and ensures effective utilization of advertising budget.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006064A_ABST
    Figure CN119006064A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of advertisement pushing, in particular to an e-commerce advertisement pushing method and system. First, first interest data and second behavior data of a user are obtained; comprehensively analyzing the two kinds of data to obtain a comprehensive preference coefficient; secondly, analyzing the comprehensive preference coefficient to obtain a first commodity and a first associated commodity set; then, analyzing a first similarity between the first commodity and each commodity in the first associated commodity set; obtaining an advertisement push commodity set; finally, calculating a first advertisement return rate of each commodity in the second associated commodity set, and summing the first advertisement return rate to obtain a first total advertisement return rate; and according to the first advertisement return rate, the first total advertisement return rate and a preset advertisement return rate threshold, dynamically adjusting the advertisement push duration and the push frequency of each commodity in the advertisement push commodity set in the next time period. Through the method, the accuracy, efficiency and resource optimization of advertisement pushing are realized, and the advertisement pushing effect and economic benefits of the e-commerce platform are remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement push, and in particular to an e-commerce advertisement push method and system. Background Art

[0002] E-commerce advertising push refers to the process of delivering product or service information to users through various digital advertising channels on e-commerce platforms. Its main purpose is to increase users' willingness to buy, increase sales, and enhance brand awareness and user loyalty. E-commerce advertising push takes various forms, including but not limited to emails, text messages, social media ads, search engine ads, display ads, video ads, and APP push notifications.

[0003] Accurate e-commerce advertising push methods are not only an effective means to increase sales and brand awareness, but also an important strategy to enhance user experience and satisfaction. With the development of big data and artificial intelligence technology, the accuracy and intelligence level of e-commerce advertising push are constantly improving, bringing greater commercial value to e-commerce platforms.

[0004] With the continuous development of science and technology, some methods for pushing e-commerce advertisements have been proposed. However, the existing advertising push methods still have certain defects. For example, the Chinese patent with the technical application number CN202410080233.0 provides an IoT e-commerce advertising delivery method and system. The method steps include collecting user interaction behaviors from e-commerce platforms and IoT devices; preprocessing user interaction behaviors, including data cleaning and analyzing user interaction behaviors to form user data portraits; using machine learning models to process the obtained user data portraits to identify user potential demand signals; adjusting advertising targets and keyword settings based on the identified user potential demand signals, and then pushing the advertising content to IoT devices for display to users; using programmatic buying technology to adjust the real-time delivery of advertisements; implementing real-time advertising performance monitoring, and then further combining network topology analysis to adjust the spatial accuracy of advertising targeting; for example, a Chinese patent with technical application number CN202310298391.9 provides an advertising delivery method based on e-commerce big data, including: generating a corresponding transaction subgraph sequence based on snapshot data of commodity sales data at different times; obtaining all first-order neighboring user nodes associated with the user node, the number of loops, and the edges contained in the self-network to construct the first Node features, based on the first node features of the neighboring user nodes, the second node features of the corresponding user node are obtained; based on the first node features and the second node features, the migration node set of the commodity trading network at the corresponding moment is obtained; based on the migration node set and the expanded first node features of each user node, the attribute set of the corresponding user node is generated; based on the attribute set, the real-time migration user node of the corresponding user node at the next moment is predicted, and based on the real-time migration user node, the target user is used as the interested user for the corresponding target advertisement delivery; the above method solves some problems of advertisement push from a certain aspect, but the consideration is relatively single. On the one hand, it does not fully consider the user's preference for commodities and lacks the consideration of the correlation between the specific return rate of advertisement push and the advertisement push strategy, which makes the e-commerce advertisement lack of accuracy in pushing commodity advertisements to users and cannot accurately match user needs; on the other hand, it does not fully consider the push value of related commodities other than the pushed commodities, which ultimately leads to the problem of low efficiency of e-commerce advertisement push.

[0005] To this end, an e-commerce advertisement push method and system are proposed. Summary of the invention

[0006] The purpose of the present invention is to provide an e-commerce advertising push method and system. First, obtain the user's first interest data and second behavior data; pre-process the two data respectively to obtain the second interest data and the second behavior data; and analyze the second interest data and the second behavior data to obtain a comprehensive preference coefficient; secondly, by analyzing the comprehensive preference coefficient, obtain the first product and the first associated product set; then, by analyzing the first similarity between the first product and each product in the first associated product set; obtain the advertising push product set; finally, calculate the first advertising return rate of each product in the advertising push product set, and sum them up to obtain the first total advertising return rate; dynamically adjust the advertising push duration and number of pushes for each product in the advertising push product set in the next time period according to the first advertising return rate, the first total advertising return rate and the preset advertising return rate threshold.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] An e-commerce advertisement pushing method, comprising:

[0009] Obtaining first interest data and second behavior data of a user on an e-commerce platform; the first interest data includes search information, consultation information and evaluation information of each product by the user; the second behavior data includes product browsing information, shopping cart addition information, purchase information, return information, sharing information and collection information of the user; pre-processing the first interest data and the first behavior data respectively to obtain second interest data and second behavior data respectively;

[0010] Analyze the second interest data and the second behavior data respectively to obtain a first interest preference coefficient and a second behavior preference coefficient, and obtain a comprehensive preference coefficient according to the first interest preference coefficient and the second behavior preference coefficient;

[0011] By analyzing the comprehensive preference coefficient, a first product is obtained; and a first associated product set of the first product is determined according to a screening condition;

[0012] Calculating a first similarity between the first product and each product in the first associated product set; determining a second associated product set according to the first similarity and a preset similarity threshold, and then merging the first product with the first associated product set to obtain an advertisement push product set;

[0013] Calculate the first advertising return rate of each product in the advertising push product set within a preset time period, and sum them up to obtain a first total advertising return rate; and dynamically adjust the advertising push duration and number of pushes for each product in the advertising push product set in the next time period based on the first total advertising return rate and a preset return rate threshold.

[0014] Preferably, the preprocessing includes denoising, stop word removal and standardization of the first interest data, and denoising and standardization of the first behavior data; and the process of obtaining the first interest preference coefficient includes:

[0015] Separate the search information, consultation information and evaluation information of the user on each product in the second interest data to obtain a word set; the word set includes the user's search information word set, consultation information word set and evaluation information word set; calculate the importance of the words in the above word sets; and perform weighted summation on each importance to obtain the first interest preference coefficient;

[0016] The first interest preference coefficient is:

[0017]

[0018] TFIDF s u,i (t) = TF s u,i (t)*IDF(t);

[0019] TFIDF q u,i (t) = TF q u,i (t)*IDF(t);

[0020] TFIDF r u,i (t) = TF r u,i (t)*IDF(t);

[0021] Among them, IP u,i It is expressed as the first interest preference coefficient of user u for product i; TFIDF s u,i (t) represents the importance of word t in the search information word set; TFIDF q u,i (t) represents the importance of word t in the query information word set; TFIDF r u,i (t) represents the importance of word t in the evaluation information word set; TF s u,i (t) is represented by the frequency of occurrence of word t in the search information word set; TF q u,i (t) is represented by the frequency of occurrence of word t in the query information word set; TF r u,i(T) represents the frequency of occurrence of word t in the evaluation information word set; IDF(t) represents the rarity of word t in each word set; α s represents the first adjustment parameter; α q represents the second adjustment parameter; α r represents the third adjustment parameter; T s u,i is represented as the search information word set; T q u,i It is represented as the consulting information word set; T r u,i It is represented as the evaluation information word set.

[0022] Preferably, the calculation formula of the second behavior preference coefficient is:

[0023] AP u,i =β c *Score C +β b *Score B -β r *Score R +β s *Score S +β f *Score F ;

[0024] Score C =e P(C|V) ;

[0025] Score B =e P(B|V) ;

[0026] Score R =e -P(R|V) ;

[0027] Score S =e P(S|V) ;

[0028] Score F =e P(F|V) ;

[0029] Among them, AP u,i Indicates the second behavior preference coefficient of user u for product i; Score C 、Score B 、Score R 、Score S 、Score Frepresents the first, second, third, fourth and fifth preference ratings of user u for product i respectively; P(C|V), P(B|V), P(R|V), P(S|V) and P(F|V) represent the probability of adding to shopping cart, purchase, return, share and favorite of user u for product i under the number of browsing times V within the preset time period respectively; C, B, R, S and F represent the number of times user u adds to shopping cart, purchases, returns, shares and favorites of product i within the preset time period respectively; β c Expressed as the first preference score weight; β b Expressed as the second preference score weight; β r Expressed as the third preference score weight; β s Expressed as the fourth preference score weight; β f Expressed as the fifth preference score weight.

[0030] Preferably, the calculation formula of the comprehensive preference coefficient is:

[0031] CP u,i =γ IP *IP u,i +γ AP *AP u,i ;

[0032] Among them, CP u,i Represented as the comprehensive preference coefficient of user u for product i; γ IP Expressed as the first interest preference coefficient weight; γ AP Expressed as the second behavior preference coefficient weight.

[0033] Preferably, the screening conditions include: same brand, same category and price range error within a preset range; the acquisition process of the first similarity formula is:

[0034] Obtain the first description text of each product in the first associated product set; preprocess the first description text to obtain a second description text; convert the second description text into a high-dimensional vector representation through the BERT model; obtain the first high-dimensional vector of each product in the first associated product set; and analyze the first high-dimensional vector of each product in the first associated product set based on the multi-head attention mechanism to obtain a first similarity; the first similarity formula is:

[0035]

[0036] Wherein, Sim(i,j) represents the first similarity; V i,hV is represented as the first high-dimensional vector representation of the first product at the hth attention head; j,h It is represented as the first high-dimensional vector representation of each product in the first associated product set at the hth attention head; Attention(V i,h ,V j,h ) represents the weighted vector calculated by the multi-head attention mechanism; a h represents the attention mechanism weight;

[0037] Select the products in the first joint product set whose first similarity is higher than the preset threshold to obtain the second joint product set, and merge the first product and the second joint product set to obtain the advertisement push product set.

[0038] Preferably, the calculation formula of the first total advertising return rate is:

[0039] Total_ROI u,i =∑ i∈L (λ*(α*CTR u,i +γ*RPC u,i *Click u,i +δ*TWT u,i )-Cost i );

[0040] Among them, Total_ROI u,i is represented by the first total advertising return rate; L is represented by the advertising push product set; λ is represented by the overall adjustment parameter; CTR u,i Represented as the click rate of user u on product i; RPC u,i Represents the advertising revenue of a single click of product i by user u; Click u,i Represented as the number of clicks by user u on product i; TWT u,i Cost is the total time that user u spends on product i. i It is represented by the total advertising cost of product i; α is the click-through rate adjustment parameter; γ is the single click revenue adjustment parameter; δ is the dwell time adjustment parameter;

[0041] The calculation formula for the advertising push duration and push times of each product in the advertising push product set is:

[0042]

[0043] Among them, T new,u,i The adjustment value of the duration of a single advertisement push of product i to user u; T base,u,i The initial value of the duration of a single advertisement push of product i to user u; Total_ROI u,i Expressed as the first total advertising return rate; ROIthreshold N is represented by the preset advertising return rate threshold; new,u,i The adjustment value of the number of advertisement pushes of product i to user u; N base,u,i The initial value of the number of times product i pushes advertisements to user u; k T represents the first advertisement adjustment parameter, which is used to control the adjustment range of the duration of a single advertisement push; k N Represents the second advertisement adjustment parameter, which is used to control the adjustment range of the number of advertisement pushes.

[0044] An e-commerce advertising push system, the system comprising a data acquisition module, a user preference analysis module, a commodity analysis module and an advertising push module;

[0045] A data acquisition module, used to acquire second interest data and second behavior data;

[0046] A user preference analysis module, used to obtain a first interest preference coefficient, a second behavior preference coefficient, a comprehensive preference coefficient and a first product;

[0047] A product analysis module, used to determine a first associated product set, a second associated product set, and an advertisement push product set;

[0048] The advertisement push module is used to obtain the advertisement push duration and push times of each product in the advertisement push product set in the next period of time.

[0049] Preferably, the second interest data and the second behavior data are obtained by preprocessing the first interest data and the first behavior data, wherein the first interest data includes the user's search information, consultation information and evaluation information on each product; the second behavior data includes the user's product browsing information, shopping cart addition information, purchase information, return information, sharing information and collection information;

[0050] The first interest preference coefficient and the second behavior preference coefficient are obtained by analyzing the second interest data and the second behavior data, and the comprehensive preference coefficient is obtained by weighting; the first commodity is the commodity with the highest comprehensive preference coefficient value; and the first associated commodity set is determined according to the screening condition;

[0051] Calculating a first similarity between the first product and each product in the first associated product set; determining a second associated product set according to the first similarity and a preset similarity threshold, and then merging the first product with the second associated product set to obtain an advertisement push product set;

[0052] Calculate the first advertising return rate of each product in the advertising push product set within a preset time period, and sum them up to obtain a first total advertising return rate; and dynamically adjust the advertising push duration and number of pushes for each product in the advertising push product set in the next time period based on the first total advertising return rate and a preset return rate threshold.

[0053] Preferably, the calculation formula of the comprehensive preference coefficient is:

[0054] CP u,i =γ IP *IP u,i +γ AP *AP u,i ;

[0055] Among them, CP u,i It is expressed as the comprehensive preference coefficient of user u for product i; IP u,i It is represented as the first interest preference coefficient of user u for product i; AP u,i represents the second behavior preference coefficient of user u for product i; γ IP Expressed as the first interest preference coefficient weight; γ AP Expressed as the second behavior preference coefficient weight;

[0056] The first similarity formula is:

[0057]

[0058] Wherein, Sim(i,j) represents the first similarity; V i,h V is represented as the first high-dimensional vector representation of the first product at the hth attention head; j,h It is represented as the first high-dimensional vector representation of each product in the first associated product set at the hth attention head; Attention(V i,h ,V j,h ) represents the weighted vector calculated by the multi-head attention mechanism; a h Represents the attention mechanism weight.

[0059] Preferably, the calculation formula of the first total advertising return rate is:

[0060] Total_ROI u,i =∑ i∈L (λ*(α*CTR u,i +γ*RPC u,i *Click u,i +δ*TWT u,i )-Cost i );

[0061] Among them, Total_ROI u,iis represented by the first total advertising return rate; L is represented by the second associated product set; λ is represented by the overall adjustment parameter; CTR u,i Represented as the click rate of user u on product i; RPC u,i Represents the advertising revenue of a single click of product i by user u; Click u,i Represented as the number of clicks by user u on product i; TWT u,i Cost is the total time that user u spends on product i. i It is represented by the total advertising cost of product i; α is the click-through rate adjustment parameter; γ is the single click revenue adjustment parameter; δ is the dwell time adjustment parameter;

[0062] The calculation formula for the advertising push duration and push times of each product in the advertising push product set is:

[0063]

[0064] Among them, T new,u,i The adjustment value of the duration of a single advertisement push of product i to user u; T base,u,i The initial value of the duration of a single advertisement push of product i to user u; Total_ROI u,i Expressed as the first total advertising return rate; ROI threshold N is represented by the preset advertising return rate threshold; new,u,i The adjustment value of the number of advertisement pushes of product i to user u; N base,u,i The initial value of the number of times product i pushes advertisements to user u; k T represents the first advertisement adjustment parameter, which is used to control the adjustment range of the duration of a single advertisement push; k N Represents the second advertisement adjustment parameter, which is used to control the adjustment range of the number of advertisement pushes.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. The present invention comprehensively and accurately reflects the user's real interest and demand for products by integrating the user's search information, consultation information and evaluation information, and analyzing various user behavior data. It also generates personalized advertising recommendation plans for each user using the calculated comprehensive preference coefficient, thereby improving the user experience, thereby improving the advertising click-through rate and conversion rate, and effectively improving the efficiency and sales value of e-commerce advertising push.

[0067] 2. The present invention captures the user's preferred products and further identifies the associated products corresponding to the user's preferred products, thereby achieving diversity in advertising push and better catering to the user's diverse shopping needs, which is conducive to improving the user's acceptance and satisfaction with recommended products, thereby improving the overall shopping experience, avoiding invalid advertising recommendations, and reducing the waste of advertising resources, which is conducive to improving the efficiency of e-commerce advertising push and the overall benefits of advertising push.

[0068] 3. The present invention obtains the advertising push return rate of e-commerce advertising push to dynamically adjust the push duration and push frequency of e-commerce advertising, thereby ensuring that the advertising budget is used most effectively, reducing investment in inefficient advertising, and concentrating more resources on high-return advertising, thereby helping to improve the overall advertising push effect and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A flow chart of an e-commerce advertisement push method provided by an embodiment of the present invention;

[0070] Figure 2 A diagram of an e-commerce advertising push system provided by an embodiment of the present invention;

[0071] Figure 3 A flowchart of obtaining an advertisement push product set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0073] Accurate e-commerce advertising push methods are not only an effective means to increase sales and brand awareness, but also an important strategy to enhance user experience and satisfaction. With the development of big data and artificial intelligence technology, the accuracy and intelligence level of e-commerce advertising push are constantly improving, bringing greater commercial value to e-commerce platforms.

[0074] There are still some deficiencies in the existing e-commerce advertising push methods. In order to achieve more accurate advertising push, the present invention will be described in detail with the help of the following two embodiments to prove that the method proposed in the present invention has certain effectiveness in actual advertising push.

[0075] Embodiment 1

[0076] This example will analyze the advertising push process of e-commerce platform A;

[0077] Reference Figure 1 A flow chart of an e-commerce advertising push method provided for an embodiment of the present invention includes S10. obtaining first interest data and second behavior data of a user on an e-commerce platform, and preprocessing the first interest data and the second behavior data to obtain second interest data and second behavior data respectively; S20. analyzing the second interest data and the second behavior data to obtain a first interest preference coefficient and a second behavior preference coefficient, and comprehensively weighting them to obtain a comprehensive preference coefficient; S30. determining a first commodity according to the comprehensive preference coefficient, and determining a first associated commodity set for the first commodity according to a screening condition; S40. calculating a first similarity between the first commodity and each commodity in the first associated commodity set, and screening out a second associated commodity set according to the similarity and a preset similarity threshold, and obtaining an advertising push commodity set by merging the first commodity and the second associated commodity set; S50. calculating a first advertising return rate for each commodity in the advertising push commodity set within a preset time period, and summing them to obtain a first total advertising return rate, and dynamically adjusting the advertising push duration and number of pushes for each commodity in the advertising push commodity set in the next time period according to the first total advertising return rate and the preset return rate threshold.

[0078] According to step S10, the first interest data and the second behavior data of the user on the A e-commerce platform are obtained; the first interest data includes the user's search information, consultation information and evaluation information for each product; the second behavior data includes the user's product browsing information, shopping cart addition information, purchase information, return information, sharing information and collection information; the first interest data and the first behavior data are preprocessed respectively to obtain the second interest data and the second behavior data respectively; the preprocessing includes denoising, stop word removal and standardization of the first interest data, and denoising and standardization of the second behavior data;

[0079] Furthermore, the user ID of the above user is U1;

[0080] According to step S20, the second interest data and the second behavior data are analyzed respectively to obtain a first interest preference coefficient and a second behavior preference coefficient, and a comprehensive preference coefficient is obtained according to the first interest preference coefficient and the second behavior preference coefficient;

[0081] Furthermore, the preprocessing includes denoising, removing stop words and standardizing the first interest data, and denoising and standardizing the second behavior data; and the process of obtaining the first interest preference coefficient includes:

[0082] Separate the search information, consultation information and evaluation information of the user on each product to obtain a word set; the word set includes the user's search information word set, consultation information word set and evaluation information word set; calculate the importance of the words in the above word sets; and perform weighted summation on each importance to obtain the first interest preference coefficient;

[0083] The first interest preference coefficient is:

[0084]

[0085] TFIDF s u,i (t) = TF s u,i (t)*IDF(t);

[0086] TFIDF q u,i (t) = TF q u,i (t)*IDF(t);

[0087] TFIDF r u,i (t) = TF r u,i (t)*IDF(t);

[0088] Among them, IP u,i It is expressed as the first interest preference coefficient of user u for product i; TFIDF s u,i (t) represents the importance of word t in the search information word set; TFIDF q u,i (t) represents the importance of word t in the query information word set; TFIDF r u,i (t) represents the importance of word t in the evaluation information word set; TF s u,i (t) is represented by the frequency of occurrence of word t in the search information word set; TF q u,i (t) is represented by the frequency of occurrence of word t in the query information word set; TF r u,i (T) represents the frequency of occurrence of word t in the evaluation information word set; IDF(t) represents the rarity of word t in each word set; α s represents the first adjustment parameter; α q represents the second adjustment parameter; α r represents the third adjustment parameter; T su,i is represented as the search information word set; T q u,i It is represented as the consulting information word set; T r u,i It is represented as the evaluation information word set.

[0089] Furthermore, the calculation formula of the second behavior preference coefficient is:

[0090] AP u,i =β c *Score C +β b *Score B -β r *Score R +β s *Score S +β f *Score F ;

[0091] Score C =e P(C|V) ;

[0092] Score B =e P(B|V) ;

[0093] Score R =e -P(R|V) ;

[0094] Score S =e P(S|V) ;

[0095] Score F =e P(F|V) ;

[0096] Among them, AP u,i Indicates the second behavior preference coefficient of user u for product i; Score C 、Score B 、Score R 、Score S 、Score Frepresents the first, second, third, fourth and fifth preference ratings of user u for product i respectively; P(C|V), P(B|V), P(R|V), P(S|V) and P(F|V) represent the probability of adding to shopping cart, purchase, return, share and favorite of user u for product i under the number of browsing times V within the preset time period respectively; C, B, R, S and F represent the number of times user u adds to shopping cart, purchases, returns, shares and favorites of product i within the preset time period respectively; β c Expressed as the first preference score weight; β b Expressed as the second preference score weight; β r Expressed as the third preference score weight; β s Expressed as the fourth preference score weight; β f Expressed as the fifth preference score weight.

[0097] Furthermore, the calculation formula of the comprehensive preference coefficient is:

[0098] CP u,i =γ IP *IP u,i +γ AP *AP u,i ;

[0099] Among them, CP u,i Represented as the comprehensive preference coefficient of user u for product i; γ IP Expressed as the first interest preference coefficient weight; γ AP Expressed as the second behavior preference coefficient weight.

[0100] Furthermore, it is obtained that the first product ID with the highest comprehensive preference coefficient value of user U1 is P1;

[0101] This embodiment comprehensively and accurately reflects the user's real interest and demand for products by integrating the user's search information, consultation information and evaluation information and analyzing various user behavior data. It also uses the calculated comprehensive preference coefficient to generate personalized advertising recommendation plans for each user, thereby improving the user experience, thereby increasing the advertising click-through rate and conversion rate, and effectively improving the efficiency of e-commerce advertising push and sales value.

[0102] By analyzing the comprehensive preference coefficient, a first product is obtained; and a first associated product set of the first product is determined according to a screening condition;

[0103] Furthermore, the screening conditions include: same brand, same category and price range error within a preset range;

[0104] Calculate the first similarity between the first product P1 and each product in the first associated product set; and determine the second associated product set according to the first similarity and a preset similarity threshold; the process of obtaining the first similarity formula is:

[0105] Obtain the first description text of each product in the first associated product set; preprocess the first description text to obtain a second description text; convert the second description text into a high-dimensional vector representation through the BERT model; obtain the first high-dimensional vector of each product in the first associated product set; and analyze the first high-dimensional vector of each product in the first associated product set based on the multi-head attention mechanism to obtain a first similarity; the first similarity formula is:

[0106]

[0107] Wherein, Sim(i,j) represents the first similarity; V i,h V is represented as the first high-dimensional vector representation of the first product at the hth attention head; j,h It is represented as the first high-dimensional vector representation of each product in the first associated product set at the hth attention head; Attention(V i,h ,V j,h ) represents the weighted vector calculated by the multi-head attention mechanism; a h Represents the attention mechanism weight;

[0108] The associated products of the first product P1 and their first similarity data are obtained, and part of the data is shown in Table 1;

[0109] Table 1 The first associated product set and the first similarity data table of some first products P1

[0110] First associated product ID First similarity P2 0.81 P3 0.62 P4 0.65 P5 0.85 P6 0.67

[0111] Table 2 Data table of the second associated product set of part of the first product P1

[0112] Second associated product ID First similarity P2 0.81 P5 0.85

[0113] In this embodiment, the preset similarity threshold is 0.7; therefore, the second associated product set of part of the first product P1 is obtained, as shown in Table 2;

[0114] Further, the first product and the second associated product set are combined to obtain an advertisement push product set; therefore, the advertisement push product set includes products P1, P2, and P5;

[0115] This embodiment captures the user's preferred products and further identifies related products corresponding to the user's preferred products, thereby achieving diversity in advertising push and better catering to the user's diverse shopping needs, which is conducive to improving the user's acceptance and satisfaction with recommended products, thereby improving the overall shopping experience, avoiding invalid advertising recommendations, and reducing the waste of advertising resources, which is conducive to improving the efficiency of e-commerce advertising push and the overall benefits of advertising push.

[0116] Calculate the first advertising return rate of each product in the advertisement push product set within a preset time period, and sum them up to obtain a first total advertising return rate; and dynamically adjust the advertising push duration and number of pushes for each product in the second associated product set in the next time period based on the first total advertising return rate and the preset return rate threshold.

[0117] Select the products in the first joint product set whose first similarity is higher than the preset threshold to obtain the second joint product set, and merge the first product and the second joint product set to obtain the advertising push product set. The specific acquisition flow chart is as follows: Figure 3 shown.

[0118] Furthermore, the calculation formula of the first total advertising return rate is:

[0119] Total_ROI u,i =∑ i∈L (λ*(α*CTR u,i +γ*RPC u,i *Click u,i +δ*TWT u,i )-Cost i );

[0120] Among them, Total_ROI u,i is represented by the first total advertising return rate; L is represented by the advertising push product set; λ is represented by the overall adjustment parameter; CTR u,i Represented as the click rate of user u on product i; RPC u,i Represents the advertising revenue of a single click of product i by user u; Click u,i Represented as the number of clicks by user u on product i; TWT u,i Cost is the total time that user u spends on product i. i It is represented by the total advertising cost of product i; α is the click-through rate adjustment parameter; γ is the single click revenue adjustment parameter; δ is the dwell time adjustment parameter;

[0121] The calculation formula for the advertising push duration and push times of each product in the advertising push product set is:

[0122]

[0123]

[0124] Among them, T new,u,i The adjustment value of the duration of a single advertisement push of product i to user u; T base,u,i The initial value of the duration of a single advertisement push of product i to user u; Total_ROI u,i Expressed as the first total advertising return rate; ROI threshold N is represented by the preset advertising return rate threshold; new,u,i The adjustment value of the number of advertisement pushes of product i to user u; N base,u,i The initial value of the number of times product i pushes advertisements to user u; k T represents the first advertisement adjustment parameter, which is used to control the adjustment range of the duration of a single advertisement push; k N Represents the second advertisement adjustment parameter, which is used to control the adjustment range of the number of advertisement pushes.

[0125] This embodiment obtains the advertising push return rate of e-commerce advertising push to dynamically adjust the push duration and number of e-commerce advertising pushes, which can ensure that the advertising budget is used most effectively, reduce investment in inefficient advertising, and concentrate more resources on high-return advertising, thereby helping to improve the overall advertising push effect and efficiency.

[0126] This embodiment comprehensively considers the user's interest data and behavior data to generate the user's comprehensive preferred products. According to the preset screening conditions, the preferred products are deeply analyzed to identify the associated related products. Through specific similarity analysis, the final set of advertising push products is determined, and the advertising push return rate of the product set is calculated. Based on the results of the return rate, the duration and number of advertising pushes are dynamically adjusted. The above method achieves the accuracy, efficiency and resource optimization of advertising push through precise user data analysis and dynamic adjustment strategies, significantly improving the advertising push effect and economic benefits of the e-commerce platform.

[0127] Embodiment 2

[0128] This embodiment will analyze the advertising push process of B e-commerce platform;

[0129] Reference Figure 1A flow chart of an e-commerce advertisement push method provided for an embodiment of the present invention includes S10. acquiring first interest data and second behavior data of a user on B e-commerce platform, and preprocessing the first interest data and the second behavior data to obtain second interest data and second behavior data respectively; S20. analyzing the second interest data and the second behavior data to obtain a first interest preference coefficient and a second behavior preference coefficient, and comprehensively weighting to obtain a comprehensive preference coefficient; S30. determining a first commodity according to the comprehensive preference coefficient, and determining a first associated commodity set for the first commodity according to a screening condition; S40. calculating a first similarity between the first commodity and each commodity in the first associated commodity set, and screening out a second associated commodity set according to the similarity and a preset similarity threshold, and obtaining an advertisement push commodity set by merging the first commodity and the second associated commodity set; S50. calculating a first advertising return rate for each commodity in the advertisement push commodity set within a preset time period, and summing them to obtain a first total advertising return rate, and dynamically adjusting the advertisement push duration and number of pushes for each commodity in the advertisement push commodity set in the next time period according to the first total advertising return rate and the preset return rate threshold;

[0130] The above steps correspond to Figure 2 The modules in the provided e-commerce advertising push system correspond to each other; the modules include a data acquisition module, a user preference analysis module, a commodity analysis module, and an advertising push module; they correspond to the above process steps respectively;

[0131] The data acquisition module includes: acquiring the first interest data and the second behavior data of the user on the B e-commerce platform; the first interest data includes the user's search information, consultation information and evaluation information on each product; the second behavior data includes the user's product browsing information, shopping cart information, purchase information, return information, sharing information and collection information; pre-processing the first interest data and the first behavior data respectively to obtain the second interest data and the second behavior data respectively; the pre-processing includes denoising, stop word removal and standardization of the first interest data, and denoising and standardization of the second behavior data;

[0132] Furthermore, the user ID of the above user is M1;

[0133] The user preference analysis module comprises: analyzing the second interest data and the second behavior data respectively to obtain a first interest preference coefficient and a second behavior preference coefficient, and obtaining a comprehensive preference coefficient according to the first interest preference coefficient and the second behavior preference coefficient;

[0134] Furthermore, the preprocessing includes denoising, removing stop words and standardizing the first interest data, and denoising and standardizing the second behavior data; and the process of obtaining the first interest preference coefficient includes:

[0135] Separate the search information, consultation information and evaluation information of the user on each product to obtain a word set; the word set includes the user's search information word set, consultation information word set and evaluation information word set; calculate the importance of the words in the above word sets; and perform weighted summation on each importance to obtain the first interest preference coefficient;

[0136] The first interest preference coefficient is:

[0137]

[0138] TFIDF s u,i (t) = TF s u,i (t)*IDF(t);

[0139] TFIDF q u,i (t) = TF q u,i (t)*IDF(t);

[0140] TFIDF r u,i (t) = TF r u,i (t)*IDF(t);

[0141] Among them, IP u,i It is expressed as the first interest preference coefficient of user u for product i; TFIDF s u,i (t) represents the importance of word t in the search information word set; TFIDF q u,i (t) represents the importance of word t in the query information word set; TFIDF r u,i (t) represents the importance of word t in the evaluation information word set; TF s u,i (t) is represented by the frequency of occurrence of word t in the search information word set; TF q u,i (t) is represented by the frequency of occurrence of word t in the query information word set; TF r u,i (T) represents the frequency of occurrence of word t in the evaluation information word set; IDF(t) represents the rarity of word t in each word set; α s represents the first adjustment parameter; α q represents the second adjustment parameter; α r represents the third adjustment parameter; Ts u,i is represented by the search information word set; T q u,i It is represented as the consulting information word set; T r u,i It is represented as the evaluation information word set.

[0142] Furthermore, the calculation formula of the second behavior preference coefficient is:

[0143] AP u,i =β c *Score C +β b *Score B -β r *Score R +β s *Score S +β f *Score F ;

[0144] Score C =e P(C|V) ;

[0145] Score B =e P(B|V) ;

[0146] Score R =e -P(R|V) ;

[0147] Score S =e P(S|V) ;

[0148] Score F =e P(F|V) ;

[0149] Among them, AP u,i Indicates the second behavior preference coefficient of user u for product i; Score C 、Score B 、Score R 、Score S 、Score Frepresents the first, second, third, fourth and fifth preference ratings of user u for product i respectively; P(C|V), P(B|V), P(R|V), P(S|V) and P(F|V) represent the probability of adding to shopping cart, purchase, return, share and favorite of user u for product i under the number of browsing times V within the preset time period respectively; C, B, R, S and F represent the number of times user u adds to shopping cart, purchases, returns, shares and favorites of product i within the preset time period respectively; β c Expressed as the first preference score weight; β b Expressed as the second preference score weight; β r Expressed as the third preference score weight; β s Expressed as the fourth preference score weight; β f Expressed as the fifth preference score weight.

[0150] Furthermore, the calculation formula of the comprehensive preference coefficient is:

[0151] CP u,i =γ IP *IP u,i +γ AP *AP u,i ;

[0152] Among them, CP u,i Represented as the comprehensive preference coefficient of user u for product i; γ IP Expressed as the first interest preference coefficient weight; γ AP Expressed as the second behavior preference coefficient weight.

[0153] The first product ID with the highest comprehensive preference coefficient value of user M1 in this embodiment is obtained as Q1;

[0154] By analyzing the comprehensive preference coefficient, a first product is obtained; and a first associated product set of the first product is determined according to a screening condition;

[0155] Furthermore, the screening conditions include: same brand, same category and price range error within a preset range;

[0156] Calculate the first similarity between the first product Q1 and each product in the first associated product set; and determine the second associated product set according to the first similarity and a preset similarity threshold; the process of obtaining the first similarity formula is:

[0157] Obtain the first description text of each product in the first associated product set; preprocess the first description text to obtain a second description text; convert the second description text into a high-dimensional vector representation through the BERT model; obtain the first high-dimensional vector of each product in the first associated product set; and analyze the first high-dimensional vector of each product in the first associated product set based on the multi-head attention mechanism to obtain a first similarity; the first similarity formula is:

[0158]

[0159] Wherein, Sim(i,j) represents the first similarity; V i,h V is represented as the first high-dimensional vector representation of the first product at the hth attention head; j,h It is represented as the first high-dimensional vector representation of each product in the first associated product set at the hth attention head; Attention(V i,h ,V j,h ) represents the weighted vector calculated by the multi-head attention mechanism; a h represents the attention mechanism weight;

[0160] Furthermore, the associated products of the first product P1 and their first similarity data are obtained, and part of the data is shown in Table 3;

[0161] Table 3 The first associated product set and the first similarity data table of some first products Q1

[0162] First associated product ID First similarity Q2 0.83 Q3 0.55 Q4 0.62 Q5 0.64 Q6 0.78

[0163] Table 4 Data table of the second associated commodity set of part of the first commodity Q1

[0164] Second associated product ID First similarity Q2 0.83 Q6 0.78

[0165] In this embodiment, the preset similarity threshold is 0.7; therefore, the second associated product set of part of the first product Q1 is obtained, as shown in Table 4;

[0166] Further, the first product and the second associated product set are combined to obtain an advertisement push product set; therefore, the advertisement push product set includes products Q1, Q2, and Q6;

[0167] Calculate the first advertising return rate of each product in the advertisement push product set within a preset time period, and sum them up to obtain a first total advertising return rate; and dynamically adjust the advertising push duration and number of pushes for each product in the second associated product set in the next time period based on the first total advertising return rate and the preset return rate threshold.

[0168] Select the products in the first joint product set whose first similarity is higher than the preset threshold to obtain the second joint product set, and merge the first product and the second joint product set to obtain the advertising push product set. The specific acquisition flow chart is as follows: Figure 3 shown.

[0169] Furthermore, the calculation formula of the first total advertising return rate is:

[0170] Total_ROI u,i =∑ i∈L (λ*(α*CTR u,i +γ*RPC u,i *Click u,i +δ*TWT u,i )-Cost i );

[0171] Among them, Total_ROI u,i is represented by the first total advertising return rate; L is represented by the advertising push product set; λ is represented by the overall adjustment parameter; CTR u,i Represented as the click rate of user u on product i; RPC u,i Represents the advertising revenue of a single click of product i by user u; Click u,i Represented as the number of clicks by user u on product i; TWT u,i Cost is the total time that user u spends on product i. i It is represented by the total advertising cost of product i; α is the click-through rate adjustment parameter; γ is the single click revenue adjustment parameter; δ is the dwell time adjustment parameter;

[0172] The calculation formula for the advertising push duration and push times of each product in the advertising push product set is:

[0173]

[0174] Among them, T new,u,i The adjustment value of the duration of a single advertisement push of product i to user u; T base,u,i The initial value of the duration of a single advertisement push of product i to user u; Total_ROI u,i Expressed as the first total advertising return rate; ROI threshold N is represented by the preset advertising return rate threshold; new,u,i The adjustment value of the number of advertisement pushes of product i to user u; N base,u,i The initial value of the number of times product i pushes advertisements to user u; k T represents the first advertisement adjustment parameter, which is used to control the adjustment range of the duration of a single advertisement push; k NRepresents the second advertisement adjustment parameter, which is used to control the adjustment range of the number of advertisement pushes.

[0175] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An e-commerce advertisement push method, characterized in that: include: Obtain the user's first interest data and first behavior data on the e-commerce platform; The first interest data includes the user's search information, consultation information and evaluation information of each product; the first behavior data includes the user's product browsing information, shopping cart addition information, purchase information, return information, sharing information and collection information; the first interest data and the first behavior data are pre-processed respectively to obtain second interest data and second behavior data respectively; Analyze the second interest data and the second behavior data respectively to obtain a first interest preference coefficient and a second behavior preference coefficient; obtain a comprehensive preference coefficient according to the first interest preference coefficient and the second behavior preference coefficient; By analyzing the comprehensive preference coefficient, a first product is obtained; and determining a first associated product set of the first product according to the screening condition; Calculating a first similarity between the first product and each product in the first associated product set; determining a second associated product set according to the first similarity and a preset similarity threshold, and then merging the first product with the second associated product set to obtain an advertisement push product set; Calculating a first advertising return rate for each product in the advertisement push product set within a preset time period, and summing them up to obtain a first total advertising return rate; According to the first total advertising return rate and a preset return rate threshold, the advertisement push duration and the number of pushes for each product in the advertisement push product set in the next time period are dynamically adjusted.

2. The e-commerce advertisement push method according to claim 1, characterized in that: The preprocessing includes denoising, stop word removal and standardization of the first interest data, and denoising and standardization of the first behavior data; The process of obtaining the first interest preference coefficient includes: Separate the search information, consultation information and evaluation information of each product by the user in the second interest data to obtain a word set; the word set includes the user's search information word set, consultation information word set and evaluation information word set; calculate the importance of the words in the above word sets; and perform weighted summation on each importance to obtain the first interest preference coefficient; The first interest preference coefficient is: TFIDF s u,i (t)=TF s u,i (t)*IDF(t); TFIDF q u,i (t)=TF q u,i (t)*IDF(t); TFIDF r u,i (t)=TF r u,i (t)*IDF(t); Among them, IP u,i It is expressed as the first interest preference coefficient of user u for product i; TFIDF s u,i (t) represents the importance of word t in the search information word set; TFIDF q u,i (t) represents the importance of word t in the query information word set; TFIDF r u,i (t) represents the importance of word t in the evaluation information word set; TF s u,i (t) is represented by the frequency of occurrence of word t in the search information word set; TF q u,i (t) is represented by the frequency of occurrence of word t in the query information word set; TF r u,i (T) represents the frequency of occurrence of word t in the evaluation information word set; IDF(t) represents the rarity of word t in each word set; α s represents the first adjustment parameter; α q represents the second adjustment parameter; α r represents the third adjustment parameter; T s u,i is represented as the search information word set; T q u,i It is represented as the consulting information word set; T r u,i It is represented as the evaluation information word set.

3. The e-commerce advertisement push method according to claim 1, characterized in that: The calculation formula of the second behavior preference coefficient is: AP u,i =β c *Score C +β b *Score B -β r *Score R +β s *Score S +β f *Score F ; Score C =e P(C|V) ; Score B =e P(B|V) ; Score R =e -P(R|V) ; Score S =e P(S|V) ; Score F =e P(F|V) ; Among them, AP u,i Indicates the second behavior preference coefficient of user u for product i; Score C 、Score B 、Score R 、Score S 、Score F represents the first, second, third, fourth and fifth preference ratings of user u for product i respectively; P(C|V), P(B|V), P(R|V), P(S|V) and P(F|V) represent the probability of adding to shopping cart, purchase, return, share and favorite of user u for product i under the number of browsing times V within the preset time period respectively; C, B, R, S and F represent the number of times user u adds to shopping cart, purchases, returns, shares and favorites of product i within the preset time period respectively; β c Expressed as the first preference score weight; β b Expressed as the second preference score weight; β r Expressed as the third preference score weight; β s Expressed as the fourth preference score weight; β f Expressed as the fifth preference score weight.

4. The e-commerce advertisement push method according to claim 1, characterized in that: The calculation formula of the comprehensive preference coefficient is: CP u,i =c IP *IP u,i +g AP *AP u,i ; Among them, CP u,i Represented as the comprehensive preference coefficient of user u for product i; γ IP Expressed as the first interest preference coefficient weight; γ AP Expressed as the second behavior preference coefficient weight.

5. The e-commerce advertisement push method according to claim 1, characterized in that: The screening conditions include: same brand, same category and price range error within a preset range; the acquisition process of the first similarity formula is: Obtain the first description text of each product in the first associated product set; preprocess the first description text to obtain a second description text; convert the second description text into a high-dimensional vector representation through the BERT model to obtain the first high-dimensional vector of each product in the first associated product set; and analyze the first high-dimensional vector of each product in the first associated product set based on the multi-head attention mechanism to obtain a first similarity; the first similarity formula is: Wherein, Sim(i,j) represents the first similarity; V i,h V is represented as the first high-dimensional vector representation of the first product at the hth attention head; j,h It is represented as the first high-dimensional vector representation of each product in the first associated product set at the hth attention head; Attention(V i,h ,V j,h ) represents the weighted vector calculated by the multi-head attention mechanism; a h Represents the attention mechanism weight; Select the products in the first joint product set whose first similarity is higher than the preset threshold to obtain the second joint product set, and merge the first product and the second joint product set to obtain the advertisement push product set.

6. The e-commerce advertisement push method according to claim 1, characterized in that:The calculation formula of the first total advertising return rate is: Total_ROI u,i =∑ i∈L (λ*(α*CTR u,i +γ*RPC u,i *Click u,i +δ*TWT u,i )-Cost i ); Among them, Total_ROI u,i is represented by the first total advertising return rate; L is represented by the advertising push product set; λ is represented by the overall adjustment parameter; CTR u,i Represented as the click rate of user u on product i; RPC u,i Represents the advertising revenue of a single click of product i by user u; Click u,i Represented as the number of clicks by user u on product i; TWT u,i Cost is the total time that user u spends on product i. i It is represented by the total advertising cost of product i; α is the click-through rate adjustment parameter; γ is the single click revenue adjustment parameter; δ is the dwell time adjustment parameter; The calculation formula for the advertising push duration and push times of each product in the advertising push product set is: Among them, T new,u,i The adjustment value of the duration of a single advertisement push of product i to user u; T base,u,i The initial value of the duration of a single advertisement push of product i to user u; Total_ROI u,i Expressed as the first total advertising return rate; ROI threshold N is represented by the preset advertising return rate threshold; new,u,i The adjustment value of the number of advertisement pushes of product i to user u; N base,u,i The initial value of the number of times product i pushes advertisements to user u; k T represents the first advertisement adjustment parameter, which is used to control the adjustment range of the duration of a single advertisement push; k N Represents the second advertisement adjustment parameter, which is used to control the adjustment range of the number of advertisement pushes.

7. An e-commerce advertising push system, characterized in that: The system includes a data acquisition module, a user preference analysis module, a product analysis module and an advertisement push module; A data acquisition module, used to acquire second interest data and second behavior data; A user preference analysis module, used to obtain a first interest preference coefficient, a second behavior preference coefficient, a comprehensive preference coefficient and a first product; A product analysis module, used to determine a first associated product set, a second associated product set, and an advertisement push product set; The advertisement push module is used to obtain the advertisement push duration and push times of each product in the advertisement push product set in the next period of time.

8. An e-commerce advertising push system according to claim 7, characterized in that: The second interest data and the second behavior data are obtained by preprocessing the first interest data and the first behavior data, wherein the first interest data includes the user's search information, consultation information and evaluation information on each product; the first behavior data includes the user's product browsing information, shopping cart addition information, purchase information, return information, sharing information and collection information; The first interest preference coefficient and the second behavior preference coefficient are obtained by analyzing the second interest data and the second behavior data, and the comprehensive preference coefficient is obtained by weighting; the first commodity is the commodity with the highest comprehensive preference coefficient value; and determining the first associated product set according to the screening condition; Calculating a first similarity between the first product and each product in the first associated product set; determining a second associated product set according to the first similarity and a preset similarity threshold, and then merging the first product with the second associated product set to obtain the advertisement push product set; Calculate the first advertising return rate of each product in the advertising push product set within a preset time period, and sum them up to obtain a first total advertising return rate; and dynamically adjust the advertising push duration and number of pushes for each product in the advertising push product set in the next time period based on the first total advertising return rate and a preset return rate threshold.

9. An e-commerce advertising push system according to claim 8, characterized in that: The calculation formula of the comprehensive preference coefficient is: CP u,i =c IP *IP u,i +g AP *AP u,i ; Among them, CP u,i Represented as the comprehensive preference coefficient of user u for product i; γ IP Expressed as the first interest preference coefficient weight; γ AP Expressed as the second behavior preference coefficient weight; The first similarity formula is: Wherein, Sim(i,j) represents the first similarity; V i,h V is represented as the first high-dimensional vector representation of the first product at the hth attention head; j,h It is represented as the first high-dimensional vector representation of each product in the first associated product set at the hth attention head; Attention(V i,h ,V j,h ) represents the weighted vector calculated by the multi-head attention mechanism; a h Represents the attention mechanism weight.

10. An e-commerce advertising push system according to claim 8, characterized in that: The calculation formula of the first total advertising return rate is: Total_ROI u,i =∑ i∈L (λ*(α*CTR u,i +γ*RPC u,i *Click u,i +δ*TWT u,i )-Cost i ); Among them, Total_ROI u,i is represented by the first total advertising return rate; L is represented by the advertising push product set; λ is represented by the overall adjustment parameter; CTR u,i Represented as the click rate of user u on product i; RPC u,i Represents the advertising revenue of a single click of product i by user u; Click u,i Represented as the number of clicks by user u on product i; TWT u,i Cost is the total time that user u spends on product i. i It is represented by the total advertising cost of product i; α is the click-through rate adjustment parameter; γ is the single click revenue adjustment parameter; δ is the dwell time adjustment parameter; The calculation formula for the advertising push duration and push times of each product in the advertising push product set is: Among them, T new,u,i The adjustment value of the duration of a single advertisement push of product i to user u; T base,u,i The initial value of the duration of a single advertisement push of product i to user u; Total_ROI u,i Expressed as the first total advertising return rate; ROI threshold N is represented by the preset advertising return rate threshold; new,u,i The adjustment value of the number of advertisement pushes of product i to user u; N base,u,i The initial value of the number of times product i pushes advertisements to user u; k T represents the first advertisement adjustment parameter, which is used to control the adjustment range of the duration of a single advertisement push; k N Represents the second advertisement adjustment parameter, which is used to control the adjustment range of the number of advertisement pushes.