Advertisement pushing method and system based on Internet of Things
By building user portrait and interest prediction models, combined with IoT technology, the problem of difficulty in accurately serving and timely modification of traditional advertising push is solved, accurate prediction of user interests and dynamic adjustment of advertising content is achieved, and personalization and real-time advertising is improved. sex.
Patent Information
- Application Number
- CN202410807405.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-21
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional advertising push is difficult to accurately serve every user, and the advertising content is difficult to modify and optimize in a timely manner, so timely and detailed feedback cannot be obtained.
By obtaining user historical click data, building user portraits, using LRU algorithms and interest prediction models, adjusting user preference sequences, generating user advertising recommendation strategies, and combining IoT technology to achieve dynamic advertising push.
It realizes accurate prediction of user interests and dynamic adjustment of advertising content, improves the personalization and real-timeness of advertising, and meets changes in user needs.
Smart Images

Figure CN118941337A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of advertisement push, and in particular to an advertisement push method and system based on the Internet of Things. Background Art
[0002] Advertisement is a way to promote items or events to the public, and it is a very important marketing tool. With the development of the Internet, advertising is no longer limited to traditional media such as television, radio, newspapers and magazines, but has expanded to various forms such as web ads and social platform ads. Traditional ads are often aimed at a wider audience, and it is difficult to accurately deliver content that each user is interested in. In addition, it is difficult to change the content of traditional ads after they are published, and it is impossible to get timely and detailed feedback after the ads are published, making it very difficult to modify the content of the next ad. Summary of the invention
[0003] In order to overcome the shortcomings that advertisement push cannot be accurately generated and cannot be modified at any time, the present invention provides an advertisement push method and system based on the Internet of Things.
[0004] The technical solution is as follows: An advertising push method based on the Internet of Things includes the following steps:
[0005] S1: Obtain relevant advertisement data of the user's historical clicks, integrate the data to build a user profile, and use the LRU algorithm to build a first user preference sequence based on the user profile;
[0006] S2: Obtain relevant interest data according to the user portrait, and input the relevant interest data into the interest prediction model after preprocessing to obtain the user's predicted interest data;
[0007] S3: setting weights for the first user preference sequence, and adjusting the first user preference sequence according to the user's predicted interest data to obtain a second user preference sequence;
[0008] S4: Perform combined adjustment according to the second user preference sequence and the corresponding weight to obtain a user advertisement recommendation strategy.
[0009] Preferably, the method of obtaining relevant advertisement data of user's historical clicks, integrating the data and constructing a user portrait, and before constructing the first user preference sequence based on the user portrait using the LRU algorithm, includes: obtaining the size specifications of the advertisement on the delivery interface according to the display interface size of the user's client device and relevant policies of the advertisement delivery platform, and obtaining the length of the first user preference sequence according to a correspondence table between the size specifications of the delivery interface and the number of sequence elements.
[0010] Preferably, the method of obtaining relevant advertising data of user's historical clicks, integrating the data to construct a user portrait, and using an LRU algorithm to construct a first user preference sequence according to the user portrait includes: obtaining relevant advertising data of multiple clicks of the user, and integrating them to construct a user portrait, obtaining various advertising style elements according to the user portrait, sorting the advertising style elements according to the proportion of each advertising style element, and obtaining the first user preference sequence according to each advertising style element and time-related factors using an LRU algorithm.
[0011] Preferably, after sorting the advertising style elements according to the proportion of each advertising style element, the first user preference sequence is obtained using the LRU algorithm according to each advertising style element and time-related factors, including: after sorting a preset number of advertisements according to the user's recent advertisement click time, a preset number of advertising sequence frames are obtained, wherein each advertising sequence frame contains a corresponding advertising style element, and after sorting the advertising style elements in each advertising sequence frame according to the proportion of the advertising style elements, a final advertising style element sequence is obtained, and the first user preference sequence is obtained using the LRU algorithm according to the final advertising style element sequence.
[0012] Preferably, the advertising style elements within each advertising sequence framework are sorted according to the proportion of the advertising style elements, including: the larger the proportion of the advertising style elements, the more recent the corresponding time; if the same phenomenon occurs in the proportion of the advertising style elements, the advertising style elements with the same proportion of the style elements are randomly sorted.
[0013] Preferably, the method of obtaining relevant advertising data of user historical clicks, integrating the data to construct a user portrait, and constructing a first user preference sequence based on the user portrait using an LRU algorithm includes: after constructing the first user preference sequence using the LRU algorithm, assigning different weight intervals to different positions, the corresponding weight intervals decrease in sequence and the total value is a fixed value.
[0014] Preferably, the method of obtaining relevant interest data according to the user portrait, and preprocessing the relevant interest data and inputting it into the interest prediction model to obtain the user's predicted interest data, includes: using various user portraits to obtain the user's most recent relevant interest data, and preprocessing it and inputting it into the interest prediction model to obtain the user's predicted interest data, collecting a large number of advertising categories and dividing them into interest categories and then preprocessing them to obtain preprocessed interest categories, and dividing the preprocessed interest categories into a test set and a training set to train the interest prediction model using Transformer, wherein the interest prediction model uses cross entropy as the loss function to optimize the model parameters.
[0015] Preferably, the first user preference sequence is weighted and adjusted according to the user's predicted interest data to obtain a second user preference sequence, including: if the predicted interest data is the same as the data in the first user preference sequence, then the final advertising style element sequence is updated and then continued to be updated using the LRU algorithm; if the predicted interest data is different from the data in the first user preference sequence, then the first user preference sequence is replaced according to the user's historical data to finally obtain the second user preference sequence.
[0016] Preferably, if the predicted interest data is different from the data in the first user preference sequence, the first user preference sequence is replaced according to the user history data, and finally the second user preference sequence is obtained, including: using the predicted interest data, replacing the lowest weight and the leftmost advertising style element in the first user preference sequence whose weight is lower than the preset weight threshold to obtain a modified complete sequence, obtaining an incomplete modified sequence whose weight is lower than the preset weight threshold according to the modified complete sequence, calculating the respective second weights of the incomplete modified sequence by a weight calculation formula, and after re-sorting according to the second weight, restoring the original weight and splicing it with the advertising style element whose weight is greater than or equal to the preset weight threshold to obtain the second user preference sequence, and the weight calculation formula is:
[0017]
[0018] Where s is the weight of the advertising style element, x is the number of times the target advertising style element is clicked by users, z is the total number of times users click on advertising style elements below the preset weight threshold, b is the total number of types of advertising style elements below the preset weight threshold, and k is a constant.
[0019] Preferably, an advertisement push system based on the Internet of Things further includes:
[0020] A length planning module, used to determine the number of sequence elements according to the size of the display interface of the user client device, thereby determining the length of the first user's preferred sequence;
[0021] A sequence construction module, used to construct a first user preference sequence using an LRU algorithm according to user data;
[0022] A sequence determination module, used to sort the advertisements and the style elements in the advertisements to obtain a final sequence of advertisement style elements;
[0023] A weight allocation module allocates different weight intervals to each advertising style element position in the first user preference sequence;
[0024] An interest prediction module is used to obtain the advertising style elements that the user will be interested in in the future based on the user portrait using an interest prediction model;
[0025] The interest prediction model training module is used to use the Transformer model training to predict the advertising style elements that users will be interested in in the future;
[0026] A first judgment module, used to judge whether the predicted interest data is the same as the data of the first user preference sequence;
[0027] The interest replacement module is used to replace the first user preference sequence when the predicted interest data is different from the data in the first user preference sequence.
[0028] Beneficial effect: The present invention determines the length of the first user preference sequence by the size of the display interface of the user client device, and uses the LRU algorithm to sort the first user preference sequence, sets different weights for different positions of the first user preference sequence, and predicts the user's future advertising style elements to obtain predicted advertising style elements, adjusts the first user preference sequence according to the weights and the predicted advertising style elements to obtain a second user preference sequence, and finally generates an advertisement that meets the user's preferences based on the second user preference sequence and the weights of each part. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 The present invention is a flowchart of an advertisement push method based on the Internet of Things. DETAILED DESCRIPTION
[0030] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following, in combination with the accompanying drawings and preferred embodiments, an advertising push method and system based on the Internet of Things proposed by the present invention, its specific implementation method, structure, characteristics and effects are described in detail as follows.
[0031] Embodiment 1: An advertisement push method based on the Internet of Things, such as Figure 1 As shown, the following steps are included:
[0032] S1: Obtain the relevant advertising data of the user's historical clicks, integrate the data to build a user profile, and use the LRU algorithm to build the first user preference sequence based on the user profile;
[0033] According to the display interface size of the user client device and the relevant policies of the advertising delivery platform, the size specifications of the advertisement on the delivery interface are obtained, and according to the correspondence table between the size specifications of the delivery interface and the number of sequence elements, the length of the first user preference sequence is obtained.
[0034] Obtain relevant advertising data of multiple user clicks, integrate them to build a user portrait, obtain various advertising style elements based on the user portrait, sort the advertising style elements according to the proportion of each advertising style element, and use the LRU algorithm according to each advertising style element and time-related factors to obtain the first user preference sequence.
[0035] After sorting a preset number of advertisements according to the user's recent advertisement click time, a preset number of advertisement sequence frames are obtained, wherein each advertisement sequence frame contains corresponding advertisement style elements, and after sorting the advertisement style elements in each advertisement sequence frame according to the proportion of the advertisement style elements, a final advertisement style element sequence is obtained, and based on the final advertisement style element sequence, the LRU algorithm is used to obtain the first user preference sequence.
[0036] The greater the proportion of advertising style elements, the more recent the corresponding time is. If the proportion of advertising style elements is the same, the advertising style elements with the same proportion of style elements will be randomly sorted.
[0037] After the first user preference sequence is constructed using the LRU algorithm, different weight intervals are assigned to different positions, and the corresponding weight intervals decrease in sequence and the total value is a fixed value.
[0038] It should be noted that by obtaining the display interface size of the user's client device and judging the type of advertising style elements, when the display interface is larger, a variety of advertising style elements can be included without appearing cluttered. When the display interface is smaller, more advertising style elements will reduce the user's relevant interest. Therefore, the number of advertising style elements needs to be adjusted according to the user's display interface.
[0039] According to the ads that users have clicked, first sort each ad in chronological order, for example: arrange them into the first ad sequence frame, the second ad sequence frame, and the third ad sequence frame, where the first ad sequence frame is the earliest, the third ad sequence frame is closest to the current time, and the first ad sequence frame is the first ad. Then obtain the proportion of ad style elements in each ad sequence frame respectively. The greater the proportion of ad style elements, the closer it is to the current time. For example: the proportion of style elements in the first ad sequence frame is 50% ancient style, 30% cyberpunk, and 20% hyperrealism, then the order in the first ad sequence frame is hyperrealism, cyberpunk, and ancient style.
[0040] When the same advertising style elements appear, the advertising style elements with the same proportions are randomly sorted. For example: the proportions of the style elements in the first advertising sequence frame are 60% ancient style, 20% cyberpunk, and 20% hyperrealism, then the sorting in the first advertising sequence frame is hyperrealism, cyberpunk, ancient style, or one of cyberpunk, hyperrealism, and ancient style is randomly selected.
[0041] After obtaining the final sequence of advertising style elements, the LRU algorithm is used to obtain the first user preference sequence, and different weight intervals are assigned to different positions. The corresponding weight intervals decrease in sequence and the total value is a fixed value, for example: cyberpunk 60%-50%, hyperrealism 40%-30%, ancient style 30%-0%, but cyberpunk 50%, hyperrealism 30%, ancient style 20%, the sum of all their weights must be 1.
[0042] S2: Obtain relevant interest data based on the user portrait, and after preprocessing, input the relevant interest data into the interest prediction model to obtain the user's predicted interest data;
[0043] Various user portraits are used to obtain the user's recent relevant interest data, which is preprocessed and input into the interest prediction model to obtain the user's predicted interest data. A large number of advertising categories are collected and divided into interest categories, and then preprocessed to obtain preprocessed interest categories. The preprocessed interest categories are divided into test sets and training sets, and then the interest prediction model using Transformer is trained. The interest prediction model uses cross entropy as the loss function to optimize the model parameters.
[0044] It should be noted that by adopting the Transformer model, it is possible to capture the user's temporal relationship well, thereby more accurately predicting the user's future preferred advertising style elements.
[0045] S3: setting weights for the first user preference sequence, and adjusting the first user preference sequence according to the user's predicted interest data to obtain a second user preference sequence;
[0046] If the predicted interest data is the same as the data in the first user preference sequence, the final advertising style element sequence is updated and then continued to be updated using the LRU algorithm. If the predicted interest data is different from the data in the first user preference sequence, the first user preference sequence is replaced according to the user's historical data, and finally the second user preference sequence is obtained.
[0047] Using the predicted interest data, the advertising style element with the lowest weight and the leftmost sorting whose weight is lower than the preset weight threshold in the first user preference sequence is replaced to obtain a modified complete sequence, and an incomplete modified sequence whose weight is lower than the preset weight threshold is obtained according to the modified complete sequence. The second weights of the incomplete modified sequences are calculated by the weight calculation formula, and after re-sorting according to the second weight, the original weights are restored and spliced with the advertising style elements whose weight is greater than or equal to the preset weight threshold to obtain the second user preference sequence. The weight calculation formula is:
[0048]
[0049] Where s is the weight of the advertising style element, x is the number of times the target advertising style element is clicked by users, z is the total number of times users click on advertising style elements below the preset weight threshold, b is the total number of types of advertising style elements below the preset weight threshold, and k is a constant.
[0050] It should be noted that the predicted interest data is matched with the data in the first user preference sequence to determine how to proceed. A specific example is as follows: if the first user preference sequence is cyberpunk, hyperrealism, and ancient style, if the predicted interest data is cyberpunk, the LRU algorithm is used to update the sequence to hyperrealism, ancient style, and cyberpunk; if the predicted interest data is minimalism, the original first user preference sequence is replaced. For example, if the original first user preference sequence is cyberpunk 20%, hyperrealism 30%, and ancient style 50%, then the cyberpunk 20%, hyperrealism 30%, and ancient style 50% with weights less than 50% are replaced. The hyperrealism 30% sequence is adjusted. Specifically, cyberpunk in cyberpunk and hyperrealism is replaced with minimalism and hyperrealism. Then the second weights of minimalism and hyperrealism for users are calculated according to the weight calculation formula. After sorting according to the second weights, the original weights are restored. For example, the second weights of minimalism and hyperrealism are: minimalism 70%, hyperrealism 30%, then the sorting is hyperrealism, minimalism. After restoring to the original weights, the overall sequence is hyperrealism 20%, minimalism 30%, and ancient style 50%, so as to obtain the second user preference sequence.
[0051] S4: Perform combination adjustment according to the second user preference sequence and the corresponding weight to obtain a user advertising recommendation strategy.
[0052] It should be noted that advertisements are arranged according to the elements and weights of the second user preference sequence, wherein the weights are equal to the proportions of the various elements, and finally an advertisement recommendation strategy suitable for the user is obtained.
[0053] Embodiment 2: Based on Embodiment 1, an advertisement push system based on the Internet of Things is also included:
[0054] A length planning module, used to determine the number of sequence elements according to the size of the display interface of the user client device, thereby determining the length of the first user's preferred sequence;
[0055] A sequence construction module, used to construct a first user preference sequence using an LRU algorithm according to user data;
[0056] A sequence determination module, used to sort the advertisements and the style elements in the advertisements to obtain a final sequence of advertisement style elements;
[0057] A weight allocation module allocates different weight intervals to each advertising style element position in the first user preference sequence;
[0058] An interest prediction module is used to obtain the advertising style elements that the user will be interested in in the future based on the user portrait using an interest prediction model;
[0059] The interest prediction model training module is used to use the Transformer model training to predict the advertising style elements that users will be interested in in the future;
[0060] A first judgment module, used to judge whether the predicted interest data is the same as the data of the first user preference sequence;
[0061] The interest replacement module is used to replace the first user preference sequence when the predicted interest data is different from the data in the first user preference sequence.
[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. An advertisement push method based on the Internet of Things, characterized in that: The following steps are involved: S1: Obtain relevant advertisement data of the user's historical clicks, integrate the data to build a user profile, and use the LRU algorithm to build a first user preference sequence based on the user profile; S2: Obtain relevant interest data according to the user portrait, and input the relevant interest data into the interest prediction model after preprocessing to obtain the user's predicted interest data; S3: setting weights for the first user preference sequence, and adjusting the first user preference sequence according to the user's predicted interest data to obtain a second user preference sequence; S4: Perform combined adjustment according to the second user preference sequence and the corresponding weight to obtain a user advertisement recommendation strategy.
2. The method for pushing advertisements based on the Internet of Things according to claim 1, characterized in that: The method of obtaining relevant advertising data of user's historical clicks, integrating the data to construct a user portrait, and before using the LRU algorithm to construct the first user preference sequence based on the user portrait, includes: obtaining the size specifications of the advertisement on the delivery interface based on the display interface size of the user's client device and the relevant policies of the advertising delivery platform, and obtaining the length of the first user preference sequence based on the correspondence table between the size specifications of the delivery interface and the number of sequence elements.
3. The method for pushing advertisements based on the Internet of Things according to claim 1, characterized in that: The method of obtaining relevant advertising data of user's historical clicks, integrating the data to construct a user portrait, and using the LRU algorithm to construct a first user preference sequence according to the user portrait includes: obtaining relevant advertising data of multiple user clicks, integrating the data to construct a user portrait, obtaining various advertising style elements according to the user portrait, sorting the advertising style elements according to the proportion of each advertising style element, and obtaining the first user preference sequence according to each advertising style element and time-related factors using the LRU algorithm.
4. The method for pushing advertisements based on the Internet of Things according to claim 3, characterized in that: After sorting the advertising style elements according to the proportion of each advertising style element, the first user preference sequence is obtained using the LRU algorithm according to each advertising style element and time-related factors, including: after sorting a preset number of advertisements according to the user's recent advertisement click time, a preset number of advertising sequence frames are obtained, wherein each advertising sequence frame contains a corresponding advertising style element, and after sorting the advertising style elements in each advertising sequence frame according to the proportion of the advertising style elements, a final advertising style element sequence is obtained, and the first user preference sequence is obtained using the LRU algorithm according to the final advertising style element sequence.
5. The method for pushing advertisements based on the Internet of Things according to claim 4, characterized in that: The advertising style elements in each advertising sequence framework are sorted according to the proportion of the advertising style elements, including: the larger the proportion of the advertising style elements, the more recent the corresponding time; if the proportion of the advertising style elements is the same, the advertising style elements with the same proportion of the style elements are randomly sorted.
6. The method for pushing advertisements based on the Internet of Things according to claim 1, characterized in that: The method obtains relevant advertising data of user historical clicks, integrates the data to construct a user portrait, and constructs a first user preference sequence based on the user portrait using an LRU algorithm, including: after using the LRU algorithm to construct the first user preference sequence, different weight intervals are assigned to different positions, and the corresponding weight intervals decrease in sequence and the total value is a fixed value.
7. The method for pushing advertisements based on the Internet of Things according to claim 1, characterized in that: The method of obtaining relevant interest data according to the user portrait, and preprocessing the relevant interest data and inputting it into the interest prediction model to obtain the user's predicted interest data includes: using various user portraits to obtain the user's most recent relevant interest data, and preprocessing it and inputting it into the interest prediction model to obtain the user's predicted interest data, collecting a large number of advertising categories and dividing them into interest categories and then preprocessing them to obtain preprocessed interest categories, and then dividing the preprocessed interest categories into a test set and a training set to train the interest prediction model using Transformer, wherein the interest prediction model uses cross entropy as a loss function to optimize model parameters.
8. The method for pushing advertisements based on the Internet of Things according to claim 1, characterized in that: The weighting of the first user preference sequence is set, and the first user preference sequence is adjusted according to the user's predicted interest data to obtain the second user preference sequence, including: if the predicted interest data is the same as the data in the first user preference sequence, the final advertising style element sequence is updated and then continued to be updated using the LRU algorithm; if the predicted interest data is different from the data in the first user preference sequence, the first user preference sequence is replaced according to the user's historical data, and finally the second user preference sequence is obtained.
9. The method for pushing advertisements based on the Internet of Things according to claim 1, characterized in that: If the predicted interest data is different from the data in the first user preference sequence, the first user preference sequence is replaced according to the user history data, and finally the second user preference sequence is obtained, including: using the predicted interest data, replacing the lowest weight and the leftmost advertising style element in the first user preference sequence whose weight is lower than the preset weight threshold, to obtain a modified complete sequence, obtaining an incomplete modified sequence whose weight is lower than the preset weight threshold according to the modified complete sequence, calculating the respective second weights of the incomplete modified sequence by a weight calculation formula, sorting again according to the second weight, restoring the original weight and splicing it with the advertising style element greater than or equal to the preset weight threshold to obtain the second user preference sequence, and the weight calculation formula is: Where s is the weight of the advertising style element, x is the number of times the target advertising style element is clicked by users, z is the total number of times users click on advertising style elements below the preset weight threshold, b is the total number of types of advertising style elements below the preset weight threshold, and k is a constant.
10. An Internet of Things-based advertising push system according to claims 1-9, characterized in that: Also includes: A length planning module, used to determine the number of sequence elements according to the size of the display interface of the user client device, thereby determining the length of the first user's preferred sequence; A sequence construction module, used to construct a first user preference sequence using an LRU algorithm according to user data; A sequence determination module, used to sort the advertisements and the style elements in the advertisements to obtain a final sequence of advertisement style elements; A weight allocation module allocates different weight intervals to each advertising style element position in the first user preference sequence; An interest prediction module is used to obtain the advertising style elements that the user will be interested in in the future based on the user portrait using an interest prediction model; The interest prediction model training module is used to use the Transformer model training to predict the advertising style elements that users will be interested in in the future; A first judgment module, used to judge whether the predicted interest data is the same as the data of the first user preference sequence; The interest replacement module is used to replace the first user preference sequence when the predicted interest data is different from the data in the first user preference sequence.
