Advertisement pushing method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of advertisement pushing, and particularly relates to an advertisement pushing method and system based on artificial intelligence, and the method comprises the steps: obtaining feature vocabularies of a target user based on a word segmentation model; establishing a long-term portrait of the target user according to the time interval difference of the feature vocabularies of the target user; obtaining the immersion degree of the target user; combining the immersion degree of the target user with the long-term portrait to obtain a short-term portrait of the target user; based on the short-term portrait of the target user, screening a to-be-recommended advertisement of the target user through an artificial intelligence technology; and sorting the to-be-recommended advertisements of the target user based on the advertisement recommendability fed back by the purchase record of each user. According to the invention, the advertisement pushing effect is improved, and the long-term total income of the three parties of the advertisement party, the user and the platform is increased.

Description

Technical Field

[0001] The present invention relates to the field of advertising push technology. More specifically, the present invention relates to an artificial intelligence-based advertising push method and system. Background Art

[0002] To maintain platform operations, platforms often collaborate with advertisers, providing content and products to users while ensuring a return on investment for advertisers. This ensures that the platform, advertisers, and users all profit. Inappropriate advertising and inappropriate delivery timing can impact user retention, leading to ineffective advertising investment, user loss for the platform, and a poor user experience. Therefore, the method of ad delivery is crucial to ensuring profitability for all three parties.

[0003] To improve the effectiveness of ad push, a commonly used approach is user profile-based recommendations. This approach leverages artificial intelligence to analyze user information and behavior, determine user profiles, and then predict ad preferences based on these behaviors. Related technologies include Chinese patent application CN112308626B, which discloses a blockchain- and AI-based ad push method and big data mining center. This method analyzes user needs based on user profile tags, pairs this need information with data trajectory identification results, and ensures a high degree of match between push ads and users. Furthermore, it reduces the disruption of ad push to users by determining target push time periods.

[0004] However, ad push can cause conflicting ads. When users are active on the platform, multiple ads may be pushed simultaneously. High ad density or low user interest can negatively impact user experience and, in turn, user activity on the platform. Furthermore, product quality issues can lead to a decline in long-term user trust in the platform, ignoring the long-term interests of all three parties. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems of unreasonable distribution of advertisements and low-quality advertisements affecting user retention, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides an artificial intelligence-based advertising push method, comprising: Obtain a push information sequence for several time periods of several users, wherein the push information sequence consists of several push information sorted in chronological order, and the push information includes text and browsing time; obtain the purchase record of each user, wherein the purchase record includes several commodities and corresponding advertisements, as well as transaction status and transaction time; the transaction status is divided into successful transaction and failed transaction; record any user as the target user, and record the most recent time period as the current time period; based on the word segmentation model, obtain several words of each push information sequence of the target user, take the union of all the words of the target user, and obtain several related words of the target user; obtain the characteristic words of the target user based on the similarity of the words in different push information of the target user; establish a long-term profile of the target user based on the time period difference of the target user's characteristic words; obtain the target user's immersion level based on the target user's browsing time of the push information and the distribution of characteristic words of the push information in the current time period; combine the target user's immersion level with the long-term profile to obtain the target user's short-term profile; based on the target user's short-term profile, screen the target user's to-be-recommended advertisements through artificial intelligence technology; and sort the target user's to-be-recommended advertisements based on the recommendability of the advertisements fed back by the purchase records of each user.

[0007] This invention adjusts the type of ads pushed to users by establishing long-term and short-term user profiles, making them more timely and relevant to user interests, thereby improving the success rate of ad push. This invention also ranks ads based on user feedback, increasing the likelihood of recommending high-quality ads and boosting long-term profits for advertisers, users, and the platform.

[0008] Preferably, the acquiring of characteristic words of the target user includes: Get the push information of the target user's i-th related word, record it as the target user's i-th related word related push information, get the word intersection and word union of any two related push information of the target user's i-th related word, compare them, and get the correlation between the two related push information; combine the target user's i-th related word related push information in pairs, record the number of combinations as , add up the correlations of the two related push information obtained by all pairwise combinations, and divide by , get the directionality of the target user's i-th related word; The N related words with the greatest directivity are recorded as the characteristic words of the target user, where N satisfies that any push information of the target user contains at least one characteristic word.

[0009] The present invention obtains characteristic words that are strongly related to the user through the user's historical push information, narrows down the analysis data, and can improve the efficiency of obtaining the advertisements to be pushed.

[0010] Preferably, the step of establishing a long-term profile of the target user includes: Intersection of all words in the target user's push information sequence for each time period with the target user's characteristic words to obtain the target user's characteristic word sequence; obtaining the frequency of the target user's wth characteristic word in each time period, sorting them in ascending order to obtain the frequency-time period sequence of the target user's wth characteristic word; recording the D time periods with the highest frequency in the frequency-time period sequence as high-frequency time periods; calculating the reproducibility of the target user's wth characteristic word; All characteristic words of the target user and their corresponding reproducibility constitute the long-term portrait of the target user.

[0011] The present invention combines the reproducibility of characteristic words to establish a long-term profile of the user, so that the user profile can distinguish the reproducibility possibility of different characteristic words, thereby providing a basis for obtaining recommended advertisements and ensuring stable push of relevant advertisements for content that the user is interested in.

[0012] Preferably, the reproducibility of the w-th characteristic vocabulary of the target user satisfies the expression: ; Where, represents the reproducibility of the w-th feature word of the target user; Indicates the number of time slots for target users; represents the frequency set of the w-th characteristic word of the target user in each time period, represents the hth frequency of the wth characteristic word of the target user in the corresponding frequency-time period sequence; Indicates the number of high-frequency time periods of the w-th characteristic word of the target user; The difference between the kth and k+1th high-frequency time periods of the target user's wth characteristic word; represents the absolute value function; represents the maximum value function; represents the natural exponential function.

[0013] Preferably, the immersion level of the target user satisfies the expression: ; Where, Indicates the target user's level of immersion; Indicates the proportion of push messages with a browsing time greater than 0 among the push messages sequence of the target user in the current time period; Indicates the Euclidean distance between the first sequence and the second sequence of the target user; represents the mean of the Euclidean distance between the first sequence and the second sequence of all users; Indicates the maximum frequency of all temporary feature words of the target user in the current time period; the temporary feature words are the feature words of the push information with a browsing time greater than 0 in the current time period; The mean of the maximum frequencies of all temporary feature words of each user in the current period; represents the maximum value function; Represents the normalization function.

[0014] The present invention analyzes the status of users browsing information on the platform to obtain the user's immersion level, thereby judging the user's current degree of acceptance of advertisements, thereby improving the success rate of advertisement push.

[0015] Preferably, obtaining the Euclidean distance between the first sequence and the second sequence includes: The ratio of the frequency of the kth temporary feature vocabulary of the target user in the current period to the total frequency of the temporary feature vocabulary in the current period is recorded as the frequency of the kth temporary feature vocabulary of the target user in the current period. All temporary feature vocabulary of the target user in the current period are sorted in descending order of frequency and recorded as the first sequence of the target user. All temporary feature vocabulary of the target user in the current period are sorted in descending order of reproducibility and recorded as the second sequence of the target user. The Euclidean distance between the first and second sequences of the target user is calculated.

[0016] Preferably, obtaining a short-term profile of the target user includes: The target user's acceptance of any feature vocabulary satisfies the expression: ; Where, represents the target user's acceptance of the wth feature word; S represents the target user's immersion; represents the reproducibility of the w-th feature word of the target user; Indicates the frequency of the w-th characteristic word of the target user in the current period; represents the normalization function; A first threshold is preset, and feature words with an acceptance degree greater than the first threshold are recorded as target feature words of the target user. The target feature words of the target user and the corresponding acceptance degrees constitute a short-term profile of the target user.

[0017] The present invention establishes a short-term profile of the user so that the advertising push to the user can more quickly discover the change of the user's interest, thereby pushing advertisements more targetedly and improving the success rate of advertising push.

[0018] Preferably, the step of sorting the advertisements to be recommended to the target user includes: Record any advertisement to be recommended to the target user as the target advertisement, obtain users whose purchase records include the product corresponding to the target advertisement and record them as the relevant users of the target advertisement, and record the proportion of relevant users with successful transactions among the relevant users of the target advertisement as the first recommendability of the target advertisement; based on the change in purchase frequency of any relevant user of the target advertisement after purchasing the product corresponding to the target advertisement, obtain the behavior change coefficient of any relevant user of the target advertisement; The mean of the behavior change coefficients of all relevant users of the target advertisement is multiplied by the first recommendability of the target advertisement, and normalized for positive correlation to obtain the second recommendability of the target advertisement; all advertisements to be recommended to the target user are sorted in descending order according to the second recommendability.

[0019] Preferably, obtaining the behavior change coefficient of any user related to the target advertisement includes: Obtain the number of purchase records of the qth related user of the target advertisement after purchasing the corresponding product of the target advertisement, and compare it with the length of the corresponding transaction time range, which is recorded as the first ratio; obtain the number of purchase records of the qth related user of the target advertisement before purchasing the corresponding product of the target advertisement, and compare it with the length of the corresponding transaction time range, which is recorded as the second ratio; compare the first ratio with the second ratio, and record it as the behavior change coefficient of the qth related user of the target advertisement.

[0020] In a second aspect, the present invention provides an artificial intelligence-based advertising push system, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned artificial intelligence-based advertising push method is implemented.

[0021] By adopting the above technical solution, the above-mentioned artificial intelligence-based advertising push method is generated into a computer program and stored in a memory to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

[0022] The beneficial effects of the present invention are: (1) The present invention analyzes the user's current level of immersion and adjusts the type of advertisements pushed to the user in real time to disperse or concentrate them, thereby increasing the user's overall interest in the advertisement, thereby increasing the probability of successful advertisement push, and thus increasing the overall benefits of the advertiser, user, and platform; (2) The present invention ranks the pushed advertisements based on the users' purchase feedback, so that high-quality advertisements are more likely to be clicked by users, which can improve the overall benefits of advertisers, users and platforms in the long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1It is a flow chart schematically illustrating an artificial intelligence-based advertising push method in the present invention. DETAILED DESCRIPTION

[0024] The embodiment of the present invention discloses an artificial intelligence-based advertising push method, referring to Figure 1 , including steps S1 to S4: S1: Obtain a push information sequence for several time periods of several users, wherein the push information sequence consists of several push information sorted in chronological order, and the push information includes text and browsing time; obtain the purchase record of each user, wherein the purchase record includes several products and corresponding advertisements, as well as transaction status and transaction time; the transaction status is divided into successful transaction and failed transaction.

[0025] It should be noted that, taking social media platforms as an example, users will receive a large amount of information, such as current affairs and personal interests. In order to operate better, the platform will cooperate with advertisers. When the advertisements overlap with the content that the users are interested in, the users are more likely to be converted into paying users of the products corresponding to the advertisements. Therefore, the pushed advertisements are often related to the users' browsing information. Therefore, the present invention first obtains the users' push information.

[0026] Specifically, a sequence of push messages for several users over several time periods is obtained. The push message sequence consists of several push messages sorted in chronological order, each containing text and browsing duration. It should be noted that the number of users and time periods is set by the implementer based on actual implementation circumstances. To ensure sufficient data volume and effective analysis, the number of users can be set to 1000, and the time periods can be set to all browsing periods of the user in the previous month. Browsing periods can be distinguished based on the time interval between users browsing push messages. For example, adjacent push messages with a time interval greater than half an hour can be divided into two time periods.

[0027] Obtain each user's purchase history, which includes several products and corresponding advertisements, as well as transaction details and transaction time. Transaction details are categorized as either successful or failed. A successful transaction indicates that the user has confirmed receipt of the product, while a failed transaction indicates that the user has returned the product.

[0028] So far, the push information sequences and purchase records of several users for several time periods have been obtained.

[0029] S2: Record any user as the target user, and record the most recent time period as the current time period; based on the word segmentation model, obtain several words in each push information sequence of the target user, take the union of all the words of the target user, and obtain several related words of the target user; obtain the characteristic words of the target user based on the similarity of the words in different push information of the target user; establish a long-term portrait of the target user based on the time period differences of the target user's characteristic words.

[0030] It should be noted that the browsing situation of the user for all push messages reflects all the user's interest areas. Extracting keywords to form a user portrait can describe the user. However, as the user's interests change and current events hotspots change, the user portrait also changes. Therefore, the present invention establishes a long-term portrait of the user to describe the long-term characteristics of the user and enrich the scope of pushable advertisements for the user.

[0031] Specifically, taking any user as the target user, using the Jieba word segmentation model to segment the text of the push message sequence of all time periods of the target user, obtaining a number of words in each push message sequence of the target user, and taking the union of all words in all push message sequences of the target user to obtain a number of relevant words of the target user.

[0032] It should be noted that a large number of non-referential words such as "is" and "of" are included in the relevant words of the target user. These words have a high frequency but are not related to the user's interests. Therefore, it is necessary to extract words that can distinguish the user's interests as the user's characteristic words. For characteristic words, they should be words that only appear in one type of push message. Therefore, when a word appears in multiple push messages with low similarity, it indicates that the pointing of the word is weak and the possibility of being a characteristic word is low.

[0033] Preferably, according to the similarity of the words in different push messages of the target user, obtain the characteristic words of the target user, including: Obtain the push message to which the i-th relevant word of the target user belongs, denoted as the relevant push message of the i-th relevant word of the target user, and obtain the word intersection and word union of any two relevant push messages of the i-th relevant word of the target user.

[0034] The pointing of any relevant word of the target user satisfies the expression: ; In the formula, represents the pointing of the i-th relevant word of the target user; represents the number of relevant push messages of the i-th relevant word of the target user; [[ID=2七十二]]represents the number of combinations obtained by pairwise combination of the relevant push messages of the i-th relevant word of the target user; 、 represent the number of words in the word intersection and word union corresponding to the a-th and b-th relevant push messages of the i-th relevant word of the target user.

[0035] In the formula, The ratio of the number of words in the intersection and union of the corresponding combinations of the a-th and b-th related push information for the target user's i-th related word. This value reflects the similarity between the a-th and b-th related push information for the target user's i-th related word. A larger value indicates that the a-th and b-th related push information for the target user's i-th related word have more identical words, and therefore a higher similarity. represents the sum of similarities of all relevant push information combinations of the target user’s i-th related word, It represents the mean similarity of all push information combinations related to the target user's i-th related word. The larger the value, the higher the overall similarity of the push information related to the target user's i-th related word. This means that the target user's i-th related word only appears in one type of push information, so the target user's i-th related word has greater directionality.

[0036] The N related words with the greatest directivity are recorded as the characteristic words of the target user, where N satisfies that any pushed information contains at least one characteristic word.

[0037] It's important to note that different users' interests change differently in the short term. For example, users who follow real-time hot topics may stop paying attention to related information after the popularity of a certain type of information has faded, while users with long-term interests may periodically or continuously pay attention to the same type of information. Therefore, the present invention analyzes the recurrence and update of users' characteristic words to create a more accurate long-term user profile.

[0038] Preferably, a long-term profile of the target user is established based on the time periods when the target user's characteristic words appear, including: The intersection of all the words in the target user's push information sequence in each time period and the target user's characteristic words is taken to obtain the target user's characteristic words in each time period. It should be noted that the target user's characteristic words in each time period represent the target user's attention in each time period.

[0039] Sort the target user's characteristic vocabulary in each period in chronological order to obtain the target user's characteristic vocabulary sequence , , where m represents the number of time periods for target users, Represents the characteristic vocabulary set of the target user in the mth period.

[0040] It should be noted that The cth feature word in ,like If the number of times the data appears in a certain number of adjacent time periods but not in other time periods, it means that the user The attention of relevant information is related to time, so as time changes, users' attention to relevant information The attention of If it appears in different periods with a large time span, it means that the user The correlation between attention and time is low, so Therefore, the present invention obtains the reproducibility of the characteristic words of the target user according to the time difference of the time period in which the characteristic words appear.

[0041] Obtain the frequency of the target user's w-th characteristic word in each time period, sort them in ascending order, and obtain the frequency-time period sequence of the target user's w-th characteristic word. Record the D time periods with the largest frequencies in the frequency-time period sequence of the target user's w-th characteristic word as the high-frequency time periods of the target user's w-th characteristic word. It should be noted that in the frequency-time period sequence of the target user's w-th characteristic word, the greater the frequency difference and the more continuous the high-frequency time periods, the lower the online linearity of the target user's w-th characteristic word. The D value is a preset value, which is set by the implementer according to the actual implementation situation. For example, the D value can be set to 10.

[0042] The reproducibility of any characteristic vocabulary of the target user satisfies the expression: ; Where, represents the reproducibility of the w-th feature word of the target user; Indicates the number of time slots for target users; represents the frequency set of the w-th characteristic word of the target user in each time period, represents the hth frequency of the wth characteristic word of the target user in the corresponding frequency-time period sequence; Indicates the number of high-frequency time periods of the w-th characteristic word of the target user; The difference between the kth and k+1th high-frequency time periods of the target user's wth characteristic word; represents the absolute value function; represents the maximum value function; represents the natural exponential function.

[0043] Where, represents the difference between the maximum frequency of the wth characteristic word of the target user in each time period and the hth frequency in the corresponding frequency-time period sequence, Indicates the average difference between the maximum frequency of the target user's w-th characteristic word in each time period and the frequencies in the corresponding frequency-time period sequence. A larger value indicates a larger average difference between other frequencies and the maximum frequency, indicating that the overall frequency difference of the target user's w-th characteristic word is larger, the target user's w-th characteristic word is more correlated with time, and thus indicates that the target user's w-th characteristic word has lower reproducibility; The larger the value, the farther the high-frequency time period distribution of the w-th characteristic word of the target user is, which means that the w-th characteristic word of the target user has a smaller correlation with time and a greater reproducibility of the w-th characteristic word of the target user.

[0044] All characteristic words of the target user and their corresponding reproducibility constitute the long-term portrait of the target user.

[0045] At this point, a long-term portrait of the target users has been obtained.

[0046] S3: Obtain the target user's immersion level based on the target user's browsing time for the pushed information and the characteristic vocabulary distribution of the pushed information during the current period; combine the target user's immersion level with the long-term portrait to obtain the target user's short-term portrait.

[0047] It's important to note that when pushing ads to users, the user's long-term profile provides the basis for the content and order of recommended ads. A higher reproducibility indicates a higher likelihood that the user will view related ads. However, the ads a user might view are also related to their current interests. While the long-term profile reflects the user's historical interest distribution, the user may currently be focused on information related to specific key words, for which the corresponding reproducibility may not be high. Therefore, it's also necessary to analyze the user's current browsing information to create a temporary profile.

[0048] It's important to further clarify that the user's state during information browsing influences their receptiveness to ads. When a user is immersed in information about a topic, their interest is high and they are more receptive to related ads. When a user is not immersed, their receptiveness to pushed information or ads is generally low. Therefore, the types of pushed ads can be expanded rather than focusing on a single type. Therefore, a user's temporary profile should include not only the characteristic words in the information they are currently browsing, but also the type distribution of these characteristic words and the characteristic words included in their long-term profile.

[0049] Specifically, the target user's immersion level is obtained based on the target user's browsing time for the pushed information and the characteristic vocabulary distribution of the pushed information during the current period, including: Obtain a set of characteristic vocabulary of the push information whose browsing time of the target user in the current period is greater than 0, and record it as the temporary characteristic vocabulary set of the target user.

[0050] It should be noted that, when browsing time is non-zero, the closer the frequency of characteristic words in each pushed information during the target user's current period is to the reproducibility of these characteristic words, the more consistent the target user's browsing habits are with the target user, thus indicating a higher level of immersion. Furthermore, the longer the browsing time, the higher the target user's immersion. Furthermore, the more concentrated the characteristic words are in the domain, the more immersed the user is in the pushed information, thus increasing their immersion.

[0051] The ratio of the frequency of the kth temporary feature vocabulary of the target user in the current period to the total frequency of the temporary feature vocabulary in the current period is recorded as the frequency of the kth temporary feature vocabulary of the target user in the current period. All temporary feature vocabulary of the target user in the current period are sorted in descending order of frequency, and recorded as the first sequence of the target user. All temporary feature vocabulary of the target user in the current period are sorted in descending order of reproducibility, and recorded as the second sequence of the target user. The Euclidean distance between the first sequence and the second sequence of the target user is obtained.

[0052] The target user's immersion level satisfies the expression: ; Where, Indicates the target user's level of immersion; Indicates the proportion of push messages with a browsing time greater than 0 among the push messages of the target user in the current time period; Indicates the Euclidean distance between the first sequence and the second sequence of the target user; represents the mean of the Euclidean distance between the first sequence and the second sequence of all users; Indicates the maximum frequency of all temporary feature words of the target user in the current period; The mean of the maximum frequencies of all temporary feature words of each user in the current period; represents the maximum value function; Represents the normalization function.

[0053] Where, Indicates the degree to which the target user's browsing behavior during the current period is consistent with the target user's historical browsing habits. Indicates the average degree to which all users' browsing habits in the current period conform to their historical browsing habits. Indicates the degree to which the target user conforms to the historical browsing habits of all users in the current period. The larger the value, the more consistent the target user is with the historical browsing habits. Indicates the relative maximum frequency of the target user's temporary characteristic words relative to all users in the current period. The larger the value, the more concentrated the target user's current browsing content; It indicates the degree to which the target user complies with the historical browsing habits and the maximum value of the concentration of the target user's current browsing content. The immersion in the two directions is compared to obtain the final immersion of the target user. At the same time, The larger the value is, the more information the target user chooses to browse in the pushed information, which means the target user has a higher interest in the platform and content, and the higher the target user's immersion.

[0054] At this point, the target user's immersion level is obtained.

[0055] It should be noted that the higher the target user's immersion level, the more receptive they are to ads related to temporary feature words. Furthermore, if these temporary feature words are more reproducible in the target user's long-term profile, the target user's receptiveness is even higher. When the target user's immersion level is low, multiplying the target user's immersion level by the reproducibility of these temporary feature words in the target user's long-term profile can dilute the reproducibility of these temporary feature words, thereby providing space for ads related to other feature words to be pushed, thereby increasing the likelihood that the user will view other ads and successfully convert.

[0056] Preferably, the target user's immersion level is combined with the long-term profile to obtain the target user's short-term profile, including: The target user's acceptance of any feature vocabulary satisfies the expression: ; Where, represents the target user's acceptance of the wth feature word; S represents the target user's immersion; represents the reproducibility of the w-th feature word of the target user; Indicates the frequency of the w-th characteristic word of the target user in the current period; Represents the normalization function.

[0057] A first threshold is preset, and feature words with an acceptance degree greater than the first threshold are recorded as target feature words of the target user. The target feature words of the target user and the corresponding acceptance degrees constitute a short-term profile of the target user.

[0058] At this point, the user's short-term portrait has been obtained.

[0059] S4: Based on the short-term profile of the target user, the target user's advertisements to be recommended are screened through artificial intelligence technology; based on the advertisement recommendability feedback from each user's purchase record, the target user's advertisements to be recommended are sorted.

[0060] Specifically, using semantic matching technology based on Natural Language Processing (NLP), a number of ads to be pushed are retrieved from the Real-time Available Pool that match the target user's target feature vocabulary. It should be noted that the Real-time Available Pool is the ad inventory that is instantly accessible via a direct API connection and is available for push. The number of ads to be pushed that match each target feature vocabulary can be set to five, and this is also determined by the implementer based on actual implementation circumstances.

[0061] It should be noted that since the Internet cannot demonstrate the authenticity of the quality of the products in the advertising content, low-quality products will affect the user experience, and then affect the user's trust in the platform, and ultimately affect the revenue of other advertisers, resulting in a decrease in the long-term overall revenue of the three parties. Therefore, it is necessary to determine the quality of the advertisements to be recommended, and the behavior of users who have purchased related advertisements can reflect the product quality of the advertisements. Among the relevant users of any advertisement to be recommended, the more users who significantly reduce the purchase frequency and the lower the proportion of successful product transactions, the lower the quality of the advertisement to be recommended.

[0062] Preferably, the advertisements to be recommended to the target user are sorted based on the recommendability of the advertisements as reflected in the purchase records of each user, including: Any ad to be recommended to a target user is recorded as the target ad. Users whose purchase records include the product corresponding to the target ad are recorded as the target ad's related users. The proportion of related users with successful transactions among the target ad's related users is recorded as the target ad's first recommendability. It should be noted that the first recommendability of a target ad is represented by the purchase success rate of the product corresponding to the target ad, ensuring the subsequent purchase success rate of the product corresponding to the target ad. A higher value indicates that the target ad is ranked higher.

[0063] Obtain the number of purchase records of the qth related user of the target ad after purchasing the corresponding product of the target ad, and compare it with the length of the corresponding transaction time range, recording it as the first ratio; obtain the number of purchase records of the qth related user of the target ad before purchasing the corresponding product of the target ad, and compare it with the length of the corresponding transaction time range, recording it as the second ratio; compare the first ratio with the second ratio, and record it as the behavior change coefficient of the qth related user of the target ad. What is needed is that the behavior change coefficient represents the change in purchase frequency of the qth related user of the target ad after purchasing the corresponding product of the target ad. A larger value indicates that the qth related user of the target ad purchases the product more frequently, thereby indicating that the corresponding product of the target ad has the characteristics of positive shopping guidance, and therefore can be ranked higher.

[0064] The second recommendability of the target advertisement is obtained by multiplying the mean of the behavior change coefficients of all relevant users of the target advertisement by the first recommendability of the target advertisement and performing positive correlation normalization.

[0065] All the advertisements to be recommended to the target user are sorted in descending order of the second recommendability, and inserted into the recommendation information in the order of the advertisements to be recommended, thus completing the advertisement push.

[0066] At this point, the advertising push is completed.

[0067] An embodiment of the present invention further discloses an artificial intelligence-based advertising push system, comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, an artificial intelligence-based advertising push method according to the present invention is implemented.

[0068] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0069] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. An artificial intelligence-based advertising push method, characterized in that: include: Obtain a push information sequence for a plurality of users during a plurality of time periods, wherein the push information sequence is composed of a plurality of push information sorted in chronological order, and the push information includes text and browsing duration; Obtaining each user's purchase records, which include several products and corresponding advertisements, as well as transaction status and transaction time; the transaction status is divided into successful transactions and failed transactions; Record any user as the target user, and record the most recent time period as the current time period; Based on the word segmentation model, several words from each push information sequence of the target user are obtained, and all the words of the target user are unioned to obtain several related words of the target user; based on the similarity of the words in different push information of the target user, the characteristic words of the target user are obtained; based on the time periods when the characteristic words of the target user appear, a long-term profile of the target user is established; based on the target user's browsing time of the push information and the distribution of characteristic words in the push information during the current period, the target user's immersion level is obtained; the target user's immersion level is combined with the long-term profile to obtain a short-term profile of the target user; Based on the short-term profile of target users, AI technology is used to screen the recommended ads for target users. Based on the ad recommendability of each user's purchase record feedback, the ads to be recommended to the target user are ranked.

2. The artificial intelligence-based advertising push method according to claim 1, characterized in that: The step of obtaining the target user's characteristic vocabulary includes: Get the push information of the target user's i-th related word, record it as the target user's i-th related word related push information, get the word intersection and word union of any two related push information of the target user's i-th related word, compare them, and get the correlation between the two related push information; combine the target user's i-th related word related push information in pairs, record the number of combinations as , add up the correlations of the two related push information obtained by all pairwise combinations, and divide by , get the directionality of the target user's i-th related word; The N related words with the greatest directivity are recorded as the characteristic words of the target user, where N satisfies that any push information of the target user contains at least one characteristic word.

3. The artificial intelligence-based advertising push method according to claim 1, characterized in that: The establishment of a long-term profile of target users includes: Intersection of all words in the target user's push information sequence for each time period with the target user's characteristic words to obtain the target user's characteristic word sequence; obtaining the frequency of the target user's wth characteristic word in each time period, sorting them in ascending order to obtain the frequency-time period sequence of the target user's wth characteristic word; recording the D time periods with the highest frequency in the frequency-time period sequence as high-frequency time periods; calculating the reproducibility of the target user's wth characteristic word; All characteristic words of the target user and their corresponding reproducibility constitute the long-term portrait of the target user.

4. The artificial intelligence-based advertising push method according to claim 3, characterized in that: The reproducibility of the w-th characteristic vocabulary of the target user satisfies the expression: ； Where, represents the reproducibility of the w-th feature word of the target user; Indicates the number of time slots for target users; represents the frequency set of the w-th characteristic word of the target user in each time period, represents the hth frequency of the wth characteristic word of the target user in the corresponding frequency-time period sequence; Indicates the number of high-frequency time periods of the w-th characteristic word of the target user; The difference between the time period numbers of the kth and k+1th high-frequency time periods of the wth characteristic word of the target user; represents the absolute value function; represents the maximum value function; represents the natural exponential function.

5. The artificial intelligence-based advertising push method according to claim 1, characterized in that: The target user's immersion level satisfies the expression: ； Where, Indicates the target user's level of immersion; Indicates the proportion of push messages with a browsing time greater than 0 among the push messages sequence of the target user in the current time period; Indicates the Euclidean distance between the first sequence and the second sequence of the target user; represents the mean of the Euclidean distance between the first sequence and the second sequence of all users; Indicates the maximum frequency of all temporary feature words of the target user in the current time period; the temporary feature words are the feature words of the push information with a browsing time greater than 0 in the current time period; The mean of the maximum frequencies of all temporary feature words of each user in the current period; represents the maximum value function; Represents the normalization function.

6. The artificial intelligence-based advertising push method according to claim 5, characterized in that: The acquisition of the Euclidean distance between the first sequence and the second sequence includes: The ratio of the frequency of the kth temporary feature vocabulary of the target user in the current period to the total frequency of the temporary feature vocabulary in the current period is recorded as the frequency of the kth temporary feature vocabulary of the target user in the current period. All temporary feature vocabulary of the target user in the current period are sorted in descending order of frequency and recorded as the first sequence of the target user. All temporary feature vocabulary of the target user in the current period are sorted in descending order of reproducibility and recorded as the second sequence of the target user. The Euclidean distance between the first and second sequences of the target user is calculated.

7. The artificial intelligence-based advertising push method according to claim 1, characterized in that: The obtaining of a short-term profile of a target user includes: The target user's acceptance of any feature vocabulary satisfies the expression: ； Where, represents the target user's acceptance of the wth feature word; S represents the target user's immersion; represents the reproducibility of the w-th feature word of the target user; Indicates the frequency of the w-th characteristic word of the target user in the current period; represents the normalization function; A first threshold is preset, and feature words with an acceptance degree greater than the first threshold are recorded as target feature words of the target user. The target feature words of the target user and the corresponding acceptance degrees constitute a short-term profile of the target user.

8. The artificial intelligence-based advertising push method according to claim 1, characterized in that: The sorting of advertisements to be recommended to target users includes: Record any advertisement to be recommended for the target user as the target advertisement, obtain users whose purchase records include the corresponding products of the target advertisement and record them as relevant users of the target advertisement, and record the proportion of relevant users with successful transactions among the relevant users of the target advertisement as the first recommendability of the target advertisement; obtain the behavior change coefficient of any relevant user of the target advertisement based on the change in the purchase frequency of any relevant user of the target advertisement after purchasing the corresponding products of the target advertisement; multiply the mean of the behavior change coefficients of all relevant users of the target advertisement by the first recommendability of the target advertisement, and perform positive correlation normalization to obtain the second recommendability of the target advertisement; sort all advertisements to be recommended for the target user in descending order according to the second recommendability.

9. The artificial intelligence-based advertising push method according to claim 8, characterized in that: The step of obtaining the behavior change coefficient of any user related to the target advertisement includes: Obtain the number of purchase records of the qth related user of the target advertisement after purchasing the corresponding product of the target advertisement, and compare it with the length of the corresponding transaction time range, which is recorded as the first ratio; obtain the number of purchase records of the qth related user of the target advertisement before purchasing the corresponding product of the target advertisement, and compare it with the length of the corresponding transaction time range, which is recorded as the second ratio; compare the first ratio with the second ratio, and record it as the behavior change coefficient of the qth related user of the target advertisement.

10. An artificial intelligence-based advertising push system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an artificial intelligence-based advertising push method according to any one of claims 1 to 9 is implemented.

Images